Computer program, information processing method, and information processing device

Abstract

The invention provides a computer program, an information processing method, and an information processing apparatus. The computer executes the following processes: acquiring data related to substrate processing; a feature extraction unit that extracts a feature amount of the acquired data using a first learning model that is a model learned so as to output the feature amount of the data in accordance with the input of the data; converting the extracted feature quantity into a feature quantity of the set object dimension; and obtaining a predicted value by inputting the dimensionally converted feature quantity into a second learning model learned so as to output the predicted value relating to the substrate processing in accordance with the input of the feature quantity having the target dimensionality.

Description

Technical Field

[0001] The present invention relates to a computer program, an information processing method, and an information processing device. Background Art

[0002] Conventionally, virtual metrology has been utilized in the field of substrate processing. For example, in virtual metrology, measurement data obtained during the processing of an object such as a substrate is analyzed, and a predicted value for the resulting object is calculated.

[0003] Patent Document 1: Japanese Patent Application No. 2019-537240 Summary of the Invention

[0004] The present disclosure provides a computer program, an information processing method, and an information processing device capable of performing analysis taking spatial correlation into account using a learning model.

[0005] A computer program involved in a technical solution of the present invention is used to enable a computer to perform the following processing: acquiring data related to substrate processing; using a first learning model to extract feature quantities of the acquired data, wherein the first learning model is a model that has been learned in a manner that outputs feature quantities of the above data based on input of the above data; converting the extracted feature quantities into feature quantities of a set object dimension; and inputting the dimension-converted feature quantities into a second learning model to obtain a predicted value, wherein the second learning model is a model that has been learned in a manner that outputs a predicted value related to substrate processing based on input of feature quantities having the above object dimension.

[0006] According to the present disclosure, analysis taking spatial correlation into consideration can be performed using a learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is an explanatory diagram illustrating the configuration of an information processing system according to an embodiment.

[0008] Figure 2 This is an explanatory diagram for explaining the prediction method in the first embodiment.

[0009] Figure 3 This is a block diagram showing the internal structure of an information processing device.

[0010] Figure 4 This is a flowchart showing the steps for generating a prediction model.

[0011] Figure 5 This is a flowchart showing the prediction steps using the prediction model.

[0012] Figure 6 This is an explanatory diagram for explaining the performance evaluation of the prediction model.

[0013] Figure 7 It is a graph showing the spatial distribution of the importance of each observation data.

[0014] Figure 8 This is a flowchart showing the steps of processing executed by the information processing device involved in the second embodiment.

[0015] Figure 9 This is an explanatory diagram for explaining the prediction method in the third embodiment.

[0016] Figure 10 This is a flowchart showing the steps of processing executed by the information processing device involved in the fourth embodiment.

[0017] Figure 11 This is a flowchart showing the steps of processing executed by the information processing device involved in Embodiment 5. DETAILED DESCRIPTION

[0018] Hereinafter, one embodiment will be described with reference to the accompanying drawings. In the description, the same elements or elements having the same function are denoted by the same reference numerals, and repeated descriptions are omitted.

[0019] (Implementation Method 1)

[0020] Figure 1 1 and 2 are explanatory diagrams illustrating the configuration of an information processing system according to an embodiment. The information processing system according to the embodiment includes an information processing apparatus 100 and a substrate processing apparatus 200 that are communicatively connected to each other.

[0021] The substrate processing apparatus 200 is, for example, a semiconductor manufacturing apparatus including at least one of an exposure apparatus, an etching apparatus, a film forming apparatus, an ion implantation apparatus, an ashing apparatus, a sputtering apparatus, etc. Alternatively, the substrate processing apparatus 200 may be a display manufacturing apparatus for manufacturing FDPs (Flat Display Panels), such as liquid crystal display panels and organic EL (Electro-Luminescence) panels.

[0022] When the process is started in the substrate processing device 200, various set values ​​such as the temperature of the substrate, the pressure in the chamber, the gas flow rate, the voltage applied from the high-frequency power supply, etc. are set. For example, the set values ​​are given by the process recipe. In addition, various sensors and equipment for measuring the temperature of the substrate, the pressure in the chamber, the gas flow rate, the voltage applied to the upper electrode or the lower electrode, the plasma luminous intensity, etc. are provided in the substrate processing device 200, and various measurement values ​​are measured during the execution of the process. In addition, in the substrate processing device 200, it is not limited to the above-mentioned measurement values, and appropriate time series data such as images (RGB data) of the substrate (wafer) before and after the process and process logs are collected at any time. The substrate processing device 200 outputs the measurement values, images, time series data, etc. obtained during the execution of the process as observation data to the information processing device 100.

[0023] The information processing apparatus 100 acquires observation data as data related to substrate processing from the substrate processing apparatus 200. The information processing apparatus 100 obtains a predicted value related to substrate processing based on the acquired observation data.

[0024] Conventionally, virtual measurements using observation data have been performed. For example, input signals such as sensor measurements, image data, and time series data have been fed into a machine learning model that matches the input signals. The model then performs calculations based on the machine learning model to determine the necessary predicted values.

[0025] However, conventional machine learning models lacked design considerations for spatial correlation, leading to issues with accuracy and interpretability. For example, if spatial correlation is not accounted for, independent predictions are made for each location. This can lead to significant discrepancies in predicted values ​​even for adjacent locations, potentially leading to spatially distorted predictions. Furthermore, if spatial correlation is not accounted for, it is difficult to understand which parameters are most effective at which locations.

[0026] Therefore, in this embodiment, a model that introduces dimensional mapping is proposed as a prediction model MD2 that takes spatial correlation into account. Dimension mapping means converting the dimensions of feature quantities (variables that serve as clues for prediction) extracted from observational data according to the physical dimensions (object dimensions) that are desired to be calculated as predicted values. Feature quantities are extracted using, for example, a machine learning model (hereinafter referred to as feature quantity extraction model MD1). In embodiment 1, by introducing dimensional mapping into a unimodal network structure, spatial correlation is explicitly taken into account, thereby achieving improvements in accuracy and interpretability.

[0027] Figure 2This is an explanatory diagram illustrating the prediction method in Embodiment 1. The information processing apparatus 100 acquires data related to substrate processing from the substrate processing apparatus 200. The data acquired by the information processing apparatus 100 is arbitrary and includes measurement data output from sensors and the like in the substrate processing apparatus 200, image data obtained by capturing the substrate being processed, and time-series data such as process logs.

[0028] The information processing apparatus 100 uses a feature extraction model MD1 (first learning model) that has been learned to take observation data as input and output features of the observation data to extract features of the observation data acquired from the substrate processing apparatus 200. The features to be extracted are preferably variables that serve as clues for prediction.

[0029] As the feature extraction model MD1, a learning model of machine learning including deep learning can be used. For example, a learning model based on CNN (Convolutional Neural Network), Transformer, RNN (Recurrent Neural Networks), LSTM (Long Short Term Memory), MLP (Multi-Layer Perceptrons), etc. can be used. Alternatively, a learning model other than deep learning, such as an autoregressive model, a moving average model, or an autoregressive moving average model, can be used. The learning model used for the feature extraction model MD1 is appropriately set according to the input observation data and the feature to be extracted.

[0030] The feature extraction model MD1, for example, includes an input layer, one or more intermediate layers, and an output layer, and learns by outputting feature values ​​from the output layer based on observation data input to the input layer. Alternatively, the value output from any of the intermediate layers may be used as the feature value. The feature extraction model MD1 may also have a structure without intermediate layers and only include an input layer and an output layer. In this embodiment, the dimension of the feature value output from the feature extraction model MD1 is assumed to be one-dimensional, but the dimension of the feature value may also be two or more.

[0031] Next, the information processing device 100 converts the dimension of the extracted feature quantity according to the object dimension (the physical dimension that is desired to be calculated as the predicted value) (dimensional mapping). When it is desired to calculate the etching rate, etching shape (opening width or opening depth), film thickness, etc. at each position within the substrate surface as a predicted value, it is sufficient to convert the dimension of the extracted feature quantity into two dimensions. Figure 2In the example, the dimensional mapping from one-dimensional feature quantity to two-dimensional feature quantity is shown. The dimensions before and after the conversion can be arbitrary, and are appropriately set according to the observation data used and the predicted value to be calculated. The object dimension is sometimes expanded or reduced, and sometimes it is equal to the dimension of the feature quantity before the conversion. When the feature quantity output from the feature quantity extraction model MD1 is composed of N (N = N x ×N y ) elements, by reconfiguring (mapping) each element into N x ×N y The matrix can convert one-dimensional feature quantities into two-dimensional feature quantities.

[0032] The information processing apparatus 100 obtains a prediction value related to substrate processing using the prediction model MD2 (second learning model) that has been learned to output a prediction value related to substrate processing using dimensionally mapped feature quantities as input.

[0033] Prediction model MD2 can use a machine learning model, including deep learning. For example, a learning model based on CNN, Transformer, RNN, LSTM, MLP, etc. can be used. Alternatively, a learning model other than deep learning, such as an autoregressive model, a moving average model, or an autoregressive moving average model, can be used. The learning model used for prediction model MD2 is appropriately set based on the object dimension of the input feature value and the predicted value to be calculated.

[0034] In this embodiment, dimension mapping is described as an independent process for the sake of convenience, but it can also be a process executed within the prediction model MD2. Therefore, the prediction model MD2 is also called a dimension mapping model.

[0035] In this embodiment, for convenience, the feature extraction model MD1 and the prediction model MD2 are described as independent learning models, but they can also be constructed as a single learning model. In this case, feature extraction, dimension mapping, and prediction value calculation are performed within the single learning model.

[0036] Figure 3 10 is a block diagram showing the internal structure of the information processing device 100. The information processing device 100 is a dedicated or general-purpose computer including a control unit 101, a storage unit 102, a communication unit 103, an operation unit 104, and a display unit 105, for example.

[0037] The control unit 101 includes a CPU (Central Processing Unit), ROM (Read Only Memory), and RAM (Random Access Memory). The ROM included in the control unit 101 stores control programs and other information that control the operation of the various hardware components included in the information processing device 100. The CPU within the control unit 101 reads and executes the control program stored in the ROM and the computer program (described later) stored in the storage unit 102 to control the operation of the various hardware components, thereby enabling the entire device to function as the information processing device of the present disclosure. The RAM included in the control unit 101 temporarily stores data used in executing calculations.

[0038] In the embodiment, the control unit 101 is configured to include a CPU, ROM, and RAM, but the configuration of the control unit 101 is not limited to the above configuration. For example, the control unit 101 may include one or more control circuits or arithmetic circuits such as a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), a quantum processor, or volatile or non-volatile memory. Furthermore, the control unit 101 may include functions such as a clock that outputs date and time information, a timer that measures the time elapsed from the issuance of a measurement start instruction to the issuance of a measurement end instruction, or a counter that counts a quantity.

[0039] The storage unit 102 includes a storage device such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or an EEPROM (Electronically Erasable Programmable Read Only Memory). The storage unit 102 stores various computer programs executed by the control unit 101 and various data used by the control unit 101.

[0040] The computer programs (program products) stored in the storage unit 102 include a prediction processing program PG1 for causing a computer to execute a process for obtaining a predicted value related to substrate processing based on observation data from the substrate processing apparatus 200. The prediction processing program PG1 may be a single computer program or a program group consisting of multiple computer programs. The prediction processing program PG1 may also be executed collaboratively by multiple computers. Furthermore, the prediction processing program PG1 may partially utilize an existing library.

[0041] The computer program including the prediction processing program PG1 is provided via a non-transitory recording medium RM on which the computer program is recorded in a readable manner. The recording medium RM is a portable memory such as a CD-ROM, a USB memory, an SD (Secure Digital) card, a micro SD card, or a Compact Flash (registered trademark). The control unit 101 uses a reading device (not shown) to read various computer programs from the recording medium RM, and causes the storage unit 102 to store the various computer programs read. In addition, the computer program stored in the storage unit 102 can also be provided through communication. In this case, the control unit 101 obtains the computer program by communicating via the communication unit 103, and causes the storage unit 102 to store the obtained computer program.

[0042] Furthermore, the storage unit 102 stores a feature extraction model MD1 for extracting features from observation data, and a prediction model MD2 for calculating predicted values ​​related to substrate processing based on the features converted to object dimensions. Alternatively, these feature extraction model MD1 and prediction model MD2 may be stored in an external device. In this case, the control unit 101 of the information processing apparatus 100 may access the external device via a communication network, transmit the observation data acquired from the substrate processing apparatus 200 to the external device, and acquire the predicted values ​​obtained as a result of calculations performed by the external device via the communication network.

[0043] The communication unit 103 includes a communication interface for transmitting and receiving various data to and from external devices. A communication interface compliant with a communication standard such as a LAN (Local Area Network) can be used as the communication interface of the communication unit 103. The external devices include the aforementioned substrate processing apparatus 200 and a user terminal (not shown). Upon receiving data to be transmitted from the control unit 101, the communication unit 103 transmits the data to the destination external device. Upon receiving data transmitted from the external device, the communication unit 103 outputs the received data to the control unit 101.

[0044] The operation unit 104 includes operation devices such as a touch panel, keyboard, and switches, and accepts various operations and settings from the user, etc. The control unit 101 performs appropriate control based on various operation information received from the operation unit 104 and stores setting information in the storage unit 102 as needed.

[0045] The display unit 105 includes a display device such as a liquid crystal monitor or an organic EL (Electro-Luminescence) monitor, and displays information to be reported to a user or the like in accordance with an instruction from the control unit 101 .

[0046] The information processing apparatus 100 in this embodiment can be a single computer or a computer system composed of multiple computers and peripheral devices. Furthermore, the information processing apparatus 100 can be a virtualized virtual machine or a cloud. Furthermore, in this embodiment, the information processing apparatus 100 and the substrate processing apparatus 200 are described as separate entities, but the information processing apparatus 100 can also be located within the substrate processing apparatus 200.

[0047] The following describes the operation of the information processing device 100 .

[0048] The information processing apparatus 100 according to the present embodiment generates a prediction model MD2 in a learning phase before starting actual operation of the substrate processing apparatus 200 .

[0049] Figure 4 This is a flowchart showing the steps for generating the prediction model MD2. Before generating the prediction model MD2, the training data required for learning is collected. For example, when the etching shape at each position within the substrate surface is determined as a predicted value based on the plasma emission intensity, the measurement data of the plasma emission intensity measured by an OES (Optical Emission Spectrometer) and the measurement data of the etching shape at each position measured using an optical observation device, an ultrasonic microscope, etc. are collected as training data. The training data is not limited to the measurement data of the plasma emission intensity and the etching shape. The observation data of the value to be predicted and the measured value of the value to be predicted are also collected as training data. The collected training data is stored in the storage unit 102 of the information processing device 100. The feature extraction model MD1 is pre-generated using a well-known algorithm.

[0050] The control unit 101 reads the training data stored in the storage unit 102 (step S101) and selects a set of training data from the read training data (step S102). The control unit 101 inputs the observation data (values ​​to be used for prediction) contained in the selected training data into the feature extraction model MD1, performs calculations based on the feature extraction model MD1, and extracts feature values ​​of the observation data (step S103).

[0051] The control unit 101 converts the dimension of the feature amount extracted from the observation data into the target dimension (step S104). That is, the control unit 101 performs dimension mapping on the dimension of the extracted feature amount according to the physical dimension to be calculated as the predicted value.

[0052] The control unit 101 inputs the feature values ​​converted to the object dimension into prediction model MD2, performs calculations based on prediction model MD2, and thereby obtains a predicted value for each position (step S105). Before learning begins, initial values ​​are set for the model parameters of prediction model MD2. While this flowchart describes the dimensionality mapping process and the calculation process based on prediction model MD2 as separate processes, dimensionality mapping can also be performed during the processing of prediction model MD2.

[0053] The control unit 101 evaluates the predicted value calculated in step S105 (step S106) and determines whether learning is complete (step S107). The predicted value is evaluated using a known loss function. During the process of optimizing (minimizing) the loss function, if the value of the loss function is less than a threshold, the control unit 101 can determine that learning of the prediction model MD2 is complete.

[0054] When determining that the learning is not completed ( S107 : No), the control unit 101 updates the model parameters (weight coefficients and biases between nodes) in the prediction model MD2 (step S108 ), and returns the process to step S102 .

[0055] When it is determined that the learning is complete ( S107 : Yes), a learned model is obtained, and the control unit 101 stores the model as a learned prediction model MD2 in the storage unit 102 (step S109 ).

[0056] In the operation phase after generating the prediction model MD2, the information processing device 100 performs prediction using the prediction model MD2. Figure 5 1 is a flowchart showing a prediction procedure using the prediction model MD2. The control unit 101 of the information processing apparatus 100 acquires observation data for prediction from the substrate processing apparatus 200 via the communication unit 103, for example (step S121).

[0057] The control unit 101 inputs the acquired observation data into the feature extraction model MD1 to execute calculations based on the feature extraction model MD1 , thereby extracting the feature of the observation data (step S122 ).

[0058] The control unit 101 converts the dimension of the feature amount extracted from the observation data into the target dimension (step S123). In other words, the control unit 101 performs dimension mapping on the dimension of the extracted feature amount according to the physical dimension to be calculated as the predicted value.

[0059] The control unit 101 inputs the feature quantity converted into the object dimension into the prediction model MD2, executes calculation based on the prediction model MD2, and thereby obtains a prediction value for each position (step S124).

[0060] The control unit 101 outputs the prediction result based on the prediction model MD2 (step S125 ). The control unit 101 may display the prediction result on the display unit 105 or notify the user terminal or the like via the communication unit 103 .

[0061] Figure 6 This is an explanatory diagram for explaining the performance evaluation of the prediction model MD2. Figure 6 The graphs shown represent the in-plane distribution of the etched shape (opening width) when measured virtually or physically. The horizontal axis of each graph corresponds to a first direction within the substrate plane, and the horizontal axis corresponds to a second direction of the substrate perpendicular to the first direction. The density shown in each graph corresponds to the width of the opening, indicating that the opening width is wider in areas with lower concentrations and narrower in areas with higher concentrations. Figure 6 A represents the prediction result based on the previous method (virtual measurement), Figure 6 B represents the prediction result (virtual measurement) based on the method disclosed in this disclosure, Figure 6 C represents a measured value based on actual measurement.

[0062] In actual measurement, multiple openings are formed on the substrate surface by etching, and the width of each opening is measured using optical observation equipment, ultrasonic microscopes, and other measurement equipment. In virtual measurement, a camera is used to capture images of the substrate surface with the same openings. The resulting images are used as observation data to estimate the opening width. The captured images use RGB color images captured by a wafer optical inspection system.

[0063] The design value of the opening width is set to be constant regardless of the position where the opening is formed. However, when the opening width of the opening formed in the substrate is actually measured, as shown in FIG. Figure 6 As shown in C, an in-plane distribution was confirmed in which the opening width was widest near the center of the substrate surface and became narrower toward the periphery.

[0064] On the other hand, when predicting the opening width using the conventional method (linear regression in this example), Figure 6 As shown in A, although the opening width is widest near the center of the substrate surface and gradually narrows toward the periphery, the area with the same opening width expands in the horizontal direction of the graph, and the prediction result becomes distorted.

[0065] In contrast, when the method disclosed herein (prediction model MD2) is used to predict the opening width, Figure 6 As shown in Figure B, the prediction results are not distorted in specific directions and are evenly distributed along the circumference, close to the actual measurement. The mean square error between the predicted and measured values ​​using the conventional method is approximately 0.8, while the mean square error between the predicted and measured values ​​using the disclosed method is approximately 0.6, significantly improving prediction accuracy.

[0066] exist Figure 6 , the prediction results using the captured image as observation data are shown, but it can be seen that the results of predicting the opening width using the plasma emission intensity and process log as observation data are improved compared to the conventional method by the method disclosed in the present invention.

[0067] As described above, in Embodiment 1, a method for performing virtual measurements using a machine learning model (prediction model MD2) was disclosed, in which spatial correlation was incorporated into the machine learning model using dimensional mapping. The use of spatial correlation facilitates model interpretation and enables predictions to reflect actual spatial distributions. Furthermore, prediction accuracy was significantly improved compared to conventional methods that did not consider spatial correlation.

[0068] (Implementation Method 2)

[0069] In the second embodiment, a configuration will be described in which the importance (also referred to as contribution) of a feature value is calculated for each position and the spatial distribution of the calculated importance is output.

[0070] The information processing device 100 according to Embodiment 2 uses the prediction model MD2 to calculate the importance (contribution) of the feature value at each position. This importance is calculated using well-known methods such as Lime (Local Interpretable Model-Agnostic Explanations), SHAP (SHapley Additive Explanations), and CAM (Class Activation Mapping). Lime and SHAP determine how much the output changes when the input decreases, and the greater the change in output, the higher the importance. CAM calculates importance using backpropagation of errors during learning.

[0071] Figure 7 It is a graph showing the spatial distribution of the importance of each observation data. Figure 7 A is the plasma luminescence intensity (OES), Figure 7 B is the captured image (wafer optical inspection system), Figure 7 C represents the spatial distribution of importance when process logs (P-logs) are applied to the observed data. The horizontal axis of each graph corresponds to a first direction within the substrate surface, and the horizontal axis corresponds to a second direction of the substrate perpendicular to the first direction. The density shown in each graph corresponds to the level of importance. High density areas on the graph represent locations of high importance, and low density areas represent locations of low importance.

[0072] When the plasma emission intensity is used as observation data to predict the aperture width, the importance of the feature quantity based on the plasma emission intensity becomes a spatial distribution in which the importance decreases toward the center of the substrate and increases toward the periphery of the substrate ( Figure 7 A). From this graph, it can be seen that when the plasma luminescence intensity is used as the observation data, the opening width can be well predicted at the periphery of the substrate. The same result is obtained when the process log is used as the observation data ( Figure 7 C).

[0073] On the other hand, when the aperture width is predicted using the image captured by the wafer optical inspection system as observation data, the importance of the feature quantity based on the captured image is low in a region of the substrate periphery (the region corresponding to the upper right corner and the lower left corner of the graph) and high in other regions ( Figure 7 B) As can be seen from this graph, when using the captured image, the opening width can be well predicted in the region excluding a portion of the substrate periphery.

[0074] Since the spatial distribution of importance differs depending on the type of observation data (feature quantity), when generating prediction model MD2, a loss function with weights adjusted for each location can be used for learning. For example, when using plasma emission intensity or process logs as observation data, a loss function that increases the weight of the periphery can be used for learning to generate a prediction model MD2 specifically for the periphery. Alternatively, when using images captured by a wafer optical inspection system as observation data, a loss function that increases the weight of the center can be used for learning to generate a prediction model MD2 specifically for the center.

[0075] Furthermore, this embodiment allows for the determination of the contribution of feature values ​​at each location. This allows, for example, understanding which portion of the substrate the sensor output values ​​in the process log contribute to. This allows for process adjustments to alter the sensor output values, thereby contributing to process improvements. Furthermore, in actual substrate processing, if poor process conditions at the periphery result in poor yields, the aforementioned method can be used to create a prediction model MD2 specifically for the periphery, and the prediction results based on this prediction model can be used to improve the process.

[0076] Figure 8 This is a flowchart showing the procedure of processing executed by the information processing apparatus 100 according to Embodiment 2. The control unit 101 of the information processing apparatus 100 acquires observation data for prediction from the substrate processing apparatus 200 via, for example, the communication unit 103 (step S201 ).

[0077] Based on the acquired observation data, the control unit 101 calculates a predicted value for each location (step S202). The method for calculating the predicted value is the same as in Implementation 1. Specifically, the control unit 101 inputs the acquired observation data into the feature extraction model MD1 to extract features, and then dimensionally maps the extracted features to the object dimensions (the physical dimensions for which the predicted value is desired). Next, the control unit 101 inputs the dimensionally mapped features into the prediction model MD2 and performs calculations to calculate the predicted value for each location.

[0078] The control unit 101 calculates the contribution of the observation data to the calculated prediction value for each position (step S203). The contribution is, for example, a SHAP value that can be calculated using the prediction model MD2. The SHAP value is equivalent to the difference between the prediction value calculated by inputting multiple observation data into the prediction model MD2 and the prediction value calculated by the prediction model MD2 in the absence of one of the multiple observation data. The contribution is not limited to the SHAP value and can be calculated using existing methods such as Lime and CAM.

[0079] The control unit 101 outputs the spatial distribution of contribution (step S204). The control unit 101 creates, for example, Figure 7 A~ Figure 7 The graph shown in C (color contour map) is displayed on the display unit 105. In addition, the control unit 101 may also transmit the created graph to the user terminal.

[0080] The control unit 101 performs control based on the contribution of each position (step S205). The control unit 101 adjusts the parameters of the control target based on the contribution of each position and controls the process according to the adjusted parameters. For example, if it is known that the plasma emission intensity of a specific frequency is highly beneficial near the periphery, the gas flow rate can be adjusted to increase the emission intensity and improve in-plane uniformity. The adjustment amount for the contribution parameter is determined, for example, by a rule base.

[0081] In addition, Figure 8 In the flowchart, it is assumed that after the spatial distribution of the contribution is output in step S204, a control step corresponding to the contribution is executed in step S205, but these steps may be preceded and followed, or only one of the steps may be executed.

[0082] As described above, in the second embodiment, the importance (contribution) of the feature value is calculated for each position and the spatial distribution of the calculated importance is output. This makes it possible to understand which parameter is likely to act at which position, thereby improving and controlling the process.

[0083] (Implementation 3)

[0084] In the third embodiment, a configuration for calculating a predicted value from a plurality of types of observation data will be described.

[0085] Typically, there are several measurement points on a wafer. Rather than calculating each measurement point independently, we extract features or calculate predicted values ​​based on the physical dimensions of the measurement point, enabling the creation of a highly accurate and interpretable model.

[0086] Figure 9 This is an explanatory diagram for explaining the prediction method in Embodiment 3. In Embodiment 3, multimodal virtual measurement taking into account spatial correlation is explained. The information processing device 100 acquires a variety of observation data. Figure 9 In the example, inputs 1 through 3 are observation data input to feature extraction models MD11, MD12, and MD13, respectively. For example, input 1 is the plasma emission intensity from an OES, input 2 is an image captured by a wafer optical inspection system, and input 3 is a process log. The observation data used for prediction is not limited to three types; it can also be two or four or more.

[0087] Feature extraction model MD11 corresponds to feature extraction model MD1 described in Embodiment 1. It is learned so that, when inputted with observation data of input 1, it outputs the feature values ​​of that observation data. Feature extraction models MD12 and MD13 are similarly learned so that, when inputted with observation data of input 1 and input 2, they output their respective feature values. The learned feature extraction models MD11, MD12, and MD13 are stored in storage unit 102 of information processing device 100.

[0088] The information processing device 100 uses the feature extraction models MD11 to MD13 to extract the feature values ​​of inputs 1 to 3, respectively, and converts the dimensions of each extracted feature value into a feature value of the object dimension. The dimension conversion of the feature value uses the dimension mapping described in the first embodiment. When converting the feature value extracted from the feature extraction model MD11 into, for example, N x ×N y In the case of two-dimensional feature quantities, the feature quantities extracted from the feature quantity extraction models MD12 and MD13 are also converted into N x ×N y Two-dimensional feature quantity.

[0089] The information processing device 100 connects the feature quantities after dimension conversion through the connection layer CL. x ×N y In the case of two-dimensional features, just add channels, such as N x ×N y ×C in the channel direction. Here, C is the number of inputs (the number of types of observation data). Figure 9 In the case of C=3.

[0090] The information processing device 100 inputs the feature values ​​connected via the connection layer CL into the prediction model MD20 to obtain a predicted value. The prediction model MD20 corresponds to the prediction model MD2 described in the first embodiment and is learned to output predicted values ​​related to substrate processing based on the input feature values. The types of models that can be used for the prediction model MD20 and the model learning method are the same as those in the first embodiment. The learned prediction model MD20 is stored in the storage unit 102 of the information processing device 100. The information processing device 100 uses the prediction model MD20 stored in the storage unit 102 to calculate the predicted value for each position on the substrate.

[0091] As described above, in Embodiment 3, a method for multimodal virtual measurement using a learning model (prediction model MD20) that incorporates spatial correlation is disclosed. By applying the method disclosed in Embodiment 2 to prediction model MD20, the contribution of feature quantities can be calculated for each modality and each position. This allows understanding of the preferred location within the dimension for each modality, improving interpretability.

[0092] Furthermore, it is possible to clearly identify the preferred locations within each dimension for each modality. For example, by using plasma emission intensity and process logs from an OES to predict the substrate periphery, and images captured by a wafer optical inspection system to predict areas beyond the substrate periphery, prediction accuracy can be improved. Furthermore, it is possible to analyze which modality influences which location, enabling improvements to the model and process.

[0093] (Implementation 4)

[0094] In the fourth embodiment, a configuration for outputting an alarm based on a predicted value will be described.

[0095] Figure 10 This is a flowchart showing the procedure of processing executed by the information processing apparatus 100 according to Embodiment 4. The control unit 101 of the information processing apparatus 100 acquires observation data for prediction from the substrate processing apparatus 200 via, for example, the communication unit 103 (step S401).

[0096] The control unit 101 calculates the predicted value of each position based on the acquired observation data (step S402). The method for calculating the predicted value is the same as that in Implementation 1. That is, the control unit 101 inputs the acquired observation data into the feature extraction model MD1 to extract the feature, and maps the dimension of the extracted feature to the object dimension. Then, the control unit 101 calculates the predicted value of each position by inputting the dimension-mapped feature into the prediction model MD2 and performing operations. In the case where multiple observation data are obtained as observation data for prediction, the control unit 101 can also use the method disclosed in Implementation 3 to calculate the predicted value through the prediction model MD20.

[0097] The control unit 101 determines whether an alarm output is necessary based on the calculated predicted value (step S403). For example, the control unit 101 compares the calculated predicted value with a pre-set threshold value and determines that an alarm output is necessary if the predicted value exceeds the threshold value (or if it is less than the threshold value). Alternatively, the control unit 101 may determine whether the predicted value falls within a pre-set normal range and determine that an alarm output is necessary if the predicted value falls outside the normal range. Furthermore, a threshold value or normal range may be set for each location of the prediction target.

[0098] When it is determined that the alarm output is not necessary (S403: No), the control unit 101 does not output the alarm and ends the processing of this flowchart.

[0099] If it is determined that an alarm is necessary (S403: Yes), the control unit 101 outputs an alarm (step S404). For example, the control unit 101 outputs the alarm by displaying information indicating that the substrate processing is abnormal on the display unit 105. Alternatively, the control unit 101 may notify a user terminal, etc., of the information indicating that the substrate processing is abnormal, using the communication unit 103.

[0100] In this embodiment, prediction models that take spatial correlation into account (prediction models MD2 and MD20) are used for prediction, thereby obtaining a more accurate prediction value. In this embodiment, since such a highly accurate prediction value is compared with a threshold value or a normal range, it is possible to more accurately determine whether an alarm output is necessary.

[0101] (Implementation 5)

[0102] In the fifth embodiment, a configuration will be described in which control during substrate processing is performed based on a predicted value.

[0103] Figure 11 This is a flowchart showing the procedure of processing executed by the information processing apparatus 100 according to Embodiment 5. The control unit 101 of the information processing apparatus 100 acquires observation data for prediction from the substrate processing apparatus 200 via, for example, the communication unit 103 (step S501).

[0104] The control unit 101 calculates the predicted value of each position based on the acquired observation data (step S502). The method for calculating the predicted value is the same as that in embodiment 1. That is, the control unit 101 inputs the acquired observation data into the feature extraction model MD1 to extract the feature, and maps the dimension of the extracted feature to the object dimension. Then, the control unit 101 calculates the predicted value of each position by inputting the dimension-mapped feature into the prediction model MD2 and performing operations. In the case where multiple observation data are obtained as observation data for prediction, the control unit 101 can also use the method disclosed in embodiment 3 to calculate the predicted value through the prediction model MD20.

[0105] The control unit 101 executes control related to substrate processing in the substrate processing apparatus 200 based on the calculated predicted value (step S503). For example, the control unit 101 compares the calculated predicted value with a pre-set reference value and, based on the deviation between the predicted value and the reference value, determines a control value for the substrate processing apparatus 200 (e.g., a control value where the predicted value approaches the reference value). Alternatively, a reference value may be set for each position of the prediction target. The control unit 101 performs control related to substrate processing by outputting a control command including the determined control value to the substrate processing apparatus 200.

[0106] In this embodiment, prediction models (prediction models MD2 and MD20) that take spatial correlation into account are used for prediction, thereby obtaining more accurate prediction values. In this embodiment, control related to substrate processing is performed based on such highly accurate prediction values, thereby improving the process.

[0107] The embodiments disclosed herein are to be considered in all respects as illustrative and non-restrictive. The scope of the present invention is indicated by the claims rather than the above, and is intended to include all modifications within the meaning and scope equivalent to the claims.

[0108] The matters described in each embodiment can be combined with each other. In addition, the independent claims and dependent claims described in the claims can be combined with each other in all combinations, regardless of the reference format. Furthermore, the claims may use a format of describing claims that reference two or more other claims (multiple claim format), but are not limited to this. Multiple claims that reference at least one multiple claim (multiple-reference-multiple claim) may also be described.

[0109] Description of Reference Signs

[0110] 100…information processing device; 101…control unit; 102…storage unit; 103…communication unit; 104…operation unit; 105…display unit; 200…substrate processing device; PG1…prediction processing program; MD1…feature quantity extraction model; MD2…prediction model; RM…recording medium.

Claims

1. A computer program for causing a computer to execute the following processing: Acquiring data related to substrate processing; Using the first learning model, the feature quantity of the acquired data is extracted, wherein, The first learning model is a model that is learned in such a manner as to output a feature value of the data based on the input of the data; Converting the extracted feature quantity into a feature quantity of a set object dimension; and The dimensionally converted feature quantity is input to a second learning model to obtain a prediction value, wherein the second learning model is a model that has been learned so as to output a prediction value related to substrate processing based on the input of the feature quantity having the target dimension.

2. The computer program according to claim 1, wherein Used to cause the above computer to execute the following processing: Outputs data representing the spatial distribution of the feature quantity after dimensionality conversion.

3. The computer program according to claim 1, wherein The second learning model is learned using a loss function in which weights are set for the spatial distribution of the feature values.

4. The computer program according to claim 1, wherein Used to cause the above computer to execute the following processing: Acquiring various data related to the substrate processing; For each of the acquired multiple data, extract a feature value using the first learning model; converting each feature quantity extracted from each of the plurality of data into a feature quantity of the object dimension; and Each feature quantity after the dimension conversion is input into the second learning model to obtain a predicted value.

5. The computer program according to claim 1, wherein Used to cause the above computer to execute the following processing: Calculating the contribution of the feature quantity at each position of the substrate to the predicted value; and Output the calculation results. The computer program according to claim 1 , wherein: Used to cause the above computer to execute the following processing: Calculating the contribution of the above data to each position of the substrate; and The control in the substrate processing is performed based on the calculation result.

7. The computer program according to claim 1, wherein Used to cause the above computer to execute the following processing: An alarm is output based on the predicted value obtained using the second learning model.

8. The computer program according to claim 1, wherein Used to cause the above computer to execute the following processing: Control during the substrate processing is performed based on the predicted value obtained using the second learning model.

9. The computer program according to claim 1, wherein The second learning model is trained using a loss function in which weights are set for the spatial distribution of the feature values.

10. A computer program for causing a computer to execute the following processing: Acquiring data related to substrate processing; Using the first learning model, the feature quantity of the acquired data is extracted, wherein, The first learning model is a model that is learned in such a manner as to output a feature value of the data based on the input of the data; Convert the extracted feature quantity into the feature quantity of the set object dimension; Setting weights in the loss function based on the spatial distribution of the feature quantities after dimensionality conversion; and A second learning model is generated using a loss function to which weights are set, wherein the second learning model is a model that has been learned so as to output a predicted value related to the substrate process based on an input of the feature amount.

11. An information processing method, comprising: performing the following processing by a computer: Acquiring data related to substrate processing; Using the first learning model, the feature quantity of the acquired data is extracted, wherein, The first learning model is a model that is learned in such a manner as to output a feature value of the data based on the input of the data; Converting the extracted feature quantity into a feature quantity of a set object dimension; and The dimensionally converted feature quantity is input to a second learning model to obtain a prediction value, wherein the second learning model is a model that has been learned so as to output a prediction value related to substrate processing based on the input of the feature quantity having the target dimension.

12. An information processing device comprising: an acquisition unit, which acquires data related to substrate processing; The extraction unit extracts the feature quantity of the acquired data using the first learning model, wherein The first learning model is a model that is learned in such a manner as to output a feature value of the data based on the input of the data; A conversion unit converts the extracted feature quantity into a feature quantity of a set object dimension; and The prediction value calculation unit inputs the dimensionally converted feature quantity into a second learning model to obtain a prediction value, wherein the second learning model is a model that has been learned to output a prediction value related to substrate processing based on the input of the feature quantity having the target dimension.

Images