Process result prediction method, information processing apparatus, and program

Abstract

A process result prediction method performed by an information processing apparatus, the method including: extracting a plurality of feature items from a process recipe of a dataset in which the process recipe, a process log of a substrate processing apparatus that has executed a process according to the process recipe, and process result information of the substrate processing apparatus that has executed the process according to the process recipe are associated with each other; receiving, from an operator, selection of a feature item to be used for classification of the data set from the extracted plurality of feature items; and predicting the process result information using a prediction model for each classification machine-learned using a data set matching the feature item selected by the operator.

Description

Process result prediction method, information processing device, and program

[0001] This disclosure relates to a process result prediction method, an information processing device, and a program.

[0002] For example, the semiconductor manufacturing process involves repeatedly depositing films and etching them onto a substrate. In development environments, the search for the optimal process recipe is conducted daily. This search for the optimal process recipe has sometimes been performed using machine learning models that analyze the correspondence between process recipes, process logs, and process results (see, for example, Patent Document 1).

[0003] Japanese Patent Publication No. 2022-102821

[0004] This disclosure provides a technique for improving the prediction accuracy of a predictive model that predicts process result information according to a process recipe.

[0005] One aspect of the present disclosure is a process result prediction method performed by an information processing device, comprising: extracting a plurality of feature items from a process recipe of a dataset associated with a process recipe, a process log of a substrate processing device that has executed a process according to the process recipe, and process result information of a substrate processing device that has executed a process according to the process recipe; receiving a selection from an operator of the feature items to be used for classifying the dataset from the extracted plurality of feature items; and predicting the process result information using a prediction model for each classification that has been machine-trained using the dataset that matches the feature items selected by the operator.

[0006] This disclosure provides a technology to improve the prediction accuracy of a predictive model that predicts process result information according to a process recipe.

[0007] This is a diagram illustrating the configuration of an example process result prediction system according to this embodiment. This is a hardware configuration diagram of an example information processing device. This is a functional block diagram of an example server device that performs the process result prediction method according to this embodiment. This is an image diagram of an example screen that displays multiple extracted feature items and accepts the selection of feature items to be used for classifying the dataset from the operator. This is an image diagram of an example screen that displays multiple extracted feature items and accepts the selection of feature items to be used for classifying the dataset from the operator. This is an image diagram of an example screen that displays multiple extracted feature items and accepts the selection of feature items to be used for classifying the dataset from the operator. This is a flowchart of an example of the processing of the process result prediction system according to this embodiment. This is an explanatory diagram of an example process recipe included in the dataset. This is an explanatory diagram of an example process recipe included in the dataset. This is a diagram of an example of multiple feature items extracted from the process recipe of the dataset. This is a diagram of an example of multiple feature items extracted from the process recipe of the dataset. This is an image diagram of an example screen that displays multiple feature items extracted in step S12 and accepts the selection of feature items to be used for classifying the dataset from the operator. This is an image diagram of an example screen that displays process result information predicted using a prediction model for each classification. This is a flowchart of an example of the process of creating a new classification prediction model using machine learning with outliers. This is an example image of a screen displaying process result information predicted using prediction models for each further subdivided category. This is an example image of a screen displaying process result information predicted using prediction models for each category. This is an example image of a screen displaying process result information for a substrate processing device predicted using prediction models for each category. This is an example image of a screen that searches for and displays a process recipe that satisfies the process result information the operator is aiming for.

[0008] The embodiments for carrying out the present invention will be described below with reference to the drawings.

[0009] <System Configuration> Figure 1 is a configuration diagram of an example of a process result prediction system according to this embodiment. The process result prediction system 1 includes a substrate processing device 10, a measuring device 11, a device controller 12, a server device 14, a database device 15, and a worker terminal 16. The substrate processing device 10, the measuring device 11, and the device controller 12 are installed in the manufacturing plant 2. The server device 14, the database device 15, and the worker terminal 16 may be installed in the manufacturing plant 2 or elsewhere. The substrate processing device 10, the measuring device 11, the device controller 12, the server device 14, the database device 15, and the worker terminal 16 are connected to each other via networks 18 and 20 such as the Internet and LAN (Local Area Network).

[0010] The substrate processing apparatus 10 is a device that performs the manufacturing process of semiconductors and the like. The manufacturing process of semiconductors and the like is carried out by repeatedly performing processes such as film deposition and etching on a substrate to be processed. The substrate processing apparatus 10 is an example of a device that executes a process according to a process recipe and outputs a process log. The process recipe shows the procedure of the process that the substrate processing apparatus 10 will execute. For example, the process recipe is set to divide the process into multiple STEPs (sections). The process log is information such as sensor values ​​that indicate the status of the substrate processing apparatus 10 while it is executing the process according to the recipe, and is sometimes called process history information. The substrate processing apparatus 10 may be a film deposition apparatus, an etching apparatus, a semiconductor manufacturing apparatus, or a heat treatment apparatus.

[0011] The substrate processing apparatus 10 receives control commands from the apparatus controller 12 according to the process recipe and executes the process. As shown in Figure 1, the substrate processing apparatus 10 may have the apparatus controller 12 installed, or it may not necessarily have the apparatus controller 12 installed as long as it is connected in a communicative manner. The apparatus controller 12 causes the substrate processing apparatus 10 to execute the process according to the process recipe by outputting control commands that control the adjustment items (control components) of the substrate processing apparatus 10 according to the process recipe. The apparatus controller 12 also has a human-machine interface function that receives instructions for the substrate processing apparatus 10 from the operator and provides information about the substrate processing apparatus 10 to the operator.

[0012] The measuring device 11 measures the film thickness, resistance, impurity concentration, or leakage current of the substrate to be processed (e.g., a semiconductor wafer) after the process, as process result information of the substrate processing apparatus 10 that has executed the process according to the process recipe.

[0013] The database device 15 stores and manages various information necessary for processing the process result prediction system 1 according to this embodiment. For example, the database device 15 may store and manage process recipes executed by the substrate processing device 10, process logs of the substrate processing device 10 that executed processes according to those process recipes, and process result information of the substrate processing device 10 that executed processes according to those process recipes.

[0014] Furthermore, the database device 15 may store and manage a dataset (hereinafter simply referred to as a dataset) which associates the process recipe executed by the substrate processing device 10, the process log of the substrate processing device 10 that executed the process according to that process recipe, and the process result information of the substrate processing device 10 that executed the process according to that process recipe.

[0015] The server device 14 obtains a dataset from the database device 15. The server device 14 may also obtain from the database device 15 the process recipe executed by the substrate processing device 10, the process log of the substrate processing device 10 that executed the process according to that process recipe, and the process result information of the substrate processing device 10 that executed the process according to that process recipe, and associate them to create a dataset.

[0016] The correspondence between the process recipe executed by the substrate processing apparatus 10, the process log of the substrate processing apparatus 10 that executed the process according to that process recipe, and the process result information of the substrate processing apparatus 10 that executed the process according to that process recipe may be performed by an operator.

[0017] The server device 14 extracts multiple feature items from the process recipe of the dataset, as described below. Feature items are items used to define the process content in the process recipe. The server device 14 accepts from the operator the selection of feature items to be used for classifying the dataset from the extracted multiple feature items, as described below. The server device 14 creates a predictive model for each classification (category) using machine learning with the dataset that matches the feature items selected by the operator, as described below. By accepting from the operator the selection of feature items to be used for classifying the dataset, this embodiment makes it possible to create a predictive model that has been machine-learned on the dataset of the classification intended by the operator. Furthermore, the server device 14 uses the created predictive model for each classification to predict the process result information of the substrate processing apparatus 10 that executes the process according to the process recipe, as described below.

[0018] Furthermore, the worker terminal 16 is a PC (Personal Computer) or smartphone operated by a worker, such as the equipment operator or analysis operator, of the substrate processing device 10 installed in the manufacturing plant 2. The worker terminal 16 may transmit the operator's operations to the server device 14 and receive and display the processing results from the server device 14.

[0019] It should be noted that the process result prediction system 1 shown in Figure 1 is just one example, and there are various system configurations depending on the application and purpose. For example, at least a part of the processing performed by the server device 14 described above may be performed by the device controller 12 or the worker terminal 16. The device controller 12, the server device 14, or the worker terminal 16 are examples of information processing devices that perform the process result prediction method of this embodiment. Furthermore, the substrate processing device 10 equipped with the device controller 12 that performs the processing of the server device 14 described above is an example of a substrate processing device 10 that performs the process result prediction method of this embodiment. The server device 14 may be implemented by multiple information processing devices, or it may be implemented as a cloud computing service.

[0020] The process result prediction system 1 shown in Figure 1 may be implemented using a standalone substrate processing device 10 and a device controller 12. The process result prediction system 1 shown in Figure 1 may have an integrated configuration in which at least a portion of the server device 14, database device 15, and worker terminal 16 are further separated.

[0021] <Hardware Configuration> The device controller 12, server device 14, database device 15, and worker terminal 16 of the process result prediction system 1 in Figure 1 are realized by, for example, the information processing device 500 with the hardware configuration shown in Figure 2. Figure 2 is a hardware configuration diagram of an example of the information processing device.

[0022] The information processing device 500 in Figure 2 includes an input device 501, an output device 502, an external interface 503, a RAM (Random Access Memory) 504, a ROM (Read Only Memory) 505, a CPU (Central Processing Unit) 506, a communication interface 507, and an HDD (Hard Disk Drive) 508, all of which are interconnected via bus B. The input device 501 and output device 502 may be connected and used only when necessary.

[0023] The input device 501 is a keyboard, mouse, touch panel, etc., and is used by the operator to input various operation signals. The output device 502 is a display, etc., and displays the processing results from the information processing device 500.

[0024] The communication interface 507 is an interface that connects the information processing device 500 to the network 18 or 20. The HDD 508 is an example of a non-volatile storage device that stores programs and data.

[0025] The external I / F 503 is an interface to an external device. The information processing device 500 can read and / or write to a recording medium 503a such as an SD (Secure Digital) memory card via the external I / F 503. ROM 505 is an example of a non-volatile semiconductor memory (storage device) in which programs and data are stored. RAM 504 is an example of a volatile semiconductor memory (storage device) that temporarily holds programs and data.

[0026] The CPU 506 is a computing device that controls and implements the functions of the entire information processing device 500 by reading programs and data from storage devices such as ROM 505 and HDD 508 onto RAM 504 and executing processing.

[0027] The device controller 12, server device 14, database device 15, and worker terminal 16 in Figure 1 can implement various functions described later by executing a program on the information processing device 500 with the hardware configuration shown in Figure 2.

[0028] <Functional Configuration> Below, an example is described in which the information processing device 500 that performs the process result prediction method according to this embodiment is a server device 14. The information processing device 500 that performs the process result prediction method according to this embodiment may be a device controller 12 or an operator terminal 16.

[0029] The server device 14 that performs the process result prediction method according to this embodiment is implemented, for example, by the functional blocks shown in Figure 3. Figure 3 is a functional block diagram of an example of the server device 14 that performs the process result prediction method according to this embodiment. Note that the functional block diagram in Figure 3 omits the illustration of components that are not necessary for the explanation of this embodiment.

[0030] The server device 14 in Figure 3 executes a program for the server device 14 and implements a dataset acquisition unit 50, a dataset storage unit 52, an extraction unit 54, a prediction model creation unit 56, a reception unit 58, a prediction unit 60, a display unit 62, a process recipe acquisition unit 64, a process result information acquisition unit 66, and a search unit 68.

[0031] The dataset acquisition unit 50 acquires a dataset and stores it in the dataset storage unit 52. The dataset may be acquired from the database device 15, or it may be created by associating the process recipe executed by the substrate processing device 10, the process log of the substrate processing device 10 that executed the process according to that process recipe, and the process result information of the substrate processing device 10 that executed the process according to that process recipe, all of which were acquired from the database device 15.

[0032] The extraction unit 54 extracts multiple feature items from the process recipe of the dataset stored in the dataset storage unit 52. For example, the extraction unit 54 recognizes a STEP in which gas is flowing to the substrate to be processed as a film deposition STEP from the process recipe, and extracts multiple feature items from the film deposition STEP. The feature items extracted from the process recipe include, for example, the type of gas, the order of gas release, the number of STEPs, and the plasma.

[0033] The display unit 62 displays the multiple feature items extracted by the extraction unit 54, for example, as shown in Figures 4A to 4C. Figures 4A to 4C are illustrative images of an example screen that displays the multiple extracted feature items and accepts the selection of feature items to be used for classifying the dataset from the operator.

[0034] Figure 4A is a screen image before accepting the selection of feature items used for classifying a dataset from an operator. Figure 4B is a screen image after accepting the selection of gas types (gas A, gas B, gas C, and gas D) as feature items used for classifying a dataset from an operator. Figure 4C is a screen image after further accepting the selection of gas release orders (A / C / B / C, A / C / A / C / B / C, and A / D / C / B / C) as feature items used for classifying a dataset from an operator in the state of Figure 4B.

[0035] The reception unit 58 accepts the selection of feature items used for classifying a dataset from an operator on the screens shown in FIGS. 4A to 4C, for example. The feature items selected by the operator are notified from the reception unit 58 to the prediction model creation unit 56 via the extraction unit 54.

[0036] The prediction model creation unit 56 creates a model formula of a prediction model using main adjustment items (also called main process knobs) that affect the process result, and machine-learns the model formula using a dataset that matches the feature items selected by the operator to create a prediction model for each classification. The creation of the prediction model by the prediction model creation unit 56 is performed each time the selection of feature items used for classifying a dataset is accepted from an operator.

[0037] The prediction unit 60 predicts the process result information of the substrate processing apparatus 10 that has executed a process according to a process recipe using the prediction model for each classification created by the prediction model creation unit 56, and causes the display unit 62 to display it. The process recipe acquisition unit 64 acquires a process recipe for causing the prediction unit 60 to predict process result information, and notifies the prediction unit 60.

[0038] The search unit 搜索部68 searches for a process recipe that satisfies the process result information targeted by the operator by using the process result information predicted by the prediction unit 60, and causes the display unit 62 to display it. The search unit 搜索部68 searches for a process recipe that satisfies the process result information targeted by the operator by inverse problem analysis using the prediction model for each classification. The process result information acquisition unit 66 acquires the process result information targeted by the operator, and notifies the search unit 搜索部68.

[0039] In the process result prediction system 1 according to the present embodiment, in order to improve the prediction accuracy of the prediction model for predicting process result information, as a dataset used for machine learning, it is necessary to classify datasets of approximate process recipes (for example, the same gas type, the same gas emission order, and the same number of STEPs).

[0040] The classification of the dataset can be mechanically performed using a clustering method such as the k - means method, for example. However, in a clustering method such as the k - means method, the classification of the dataset may be performed against the intention of the operator, and the prediction accuracy of the prediction model trained with that dataset may be significantly reduced. Also, the k - means method has problems such as the accuracy varying depending on the way the initial value of the centroid is taken and the problem that the operator must set the number of clusters.

[0041] Therefore, in the process result prediction system 1 according to the present embodiment, by accepting the selection of feature items used for classifying the dataset from the operator, it is possible to suppress the classification of the dataset against the intention of the operator. Also, in the process result prediction system 1 according to the present embodiment, by classifying the dataset reflecting the intention of the operator, a prediction model of the classification intended by the operator can be created with high prediction accuracy.

[0042] <Processing> FIG. 5 is a flowchart of an example of the processing of the process result prediction system according to the present embodiment.

[0043] In step S10, the dataset acquisition unit 50 of the server device 14 acquires a dataset and stores it in the dataset storage unit 52. The dataset stored in the dataset storage unit 52 in step S10 includes, for example, the process recipes shown in FIGS. 6A and 6B.

[0044] FIGS. 6A and 6B are explanatory diagrams of an example of the process recipe included in the dataset. As shown in FIGS. 6A and 6B, for the process recipe, information such as temperature, pressure, gas, and number of repetitions is set for each STEP.

[0045] FIG. 6A shows an example of a process recipe where the number of STEPs is "4" and the gas types are "Gas A, Gas B, and Gas C". FIG. 6B shows an example of a process recipe where the number of STEPs is "5" and the gas types are "Gas A, Gas B, Gas C, and Gas D".

[0046] Gas A is, for example, TiCl4. Gas B is, for example, NH3. Gas C is, for example, H2. Gas D is, for example, SiH4. For example, Gas A may be WCl5, WF6, TaCl5, or AlCl3. Gas B may be N ２ gas, N ２ H ４ gas, or N ２ H ２ gas. Gas C may be diborane (B ２ H ６ ) gas, phosphine (PH ３ ) gas, dichlorosilane (SiH ２ Cl ２ ) gas, N ２ gas, or Ar gas. Gas D may be SiH ４ gas, Si ２ H ６ gas, BH ３ gas, or B ２ H ６ gas.

[0047] In step S12, the extraction unit 54 reads the dataset from the dataset storage unit 52 and extracts a plurality of feature items from the process recipe of the dataset. FIGS. 7 and 8 are diagrams of an example of a plurality of feature items extracted from the process recipe of the dataset. The plurality of feature items in FIGS. 7 and 8 have a hierarchical configuration.

[0048] Note that the plurality of feature items extracted from the process recipe of the dataset may have different contents as shown in FIGS. 7 and 8. Also, the names of the plurality of feature items extracted from the process recipe of the dataset may be different as shown in "Gas type" in FIG. 7 and "Gas / raw material type" in FIG. 8.

[0049] Here, we will explain several feature items in Figure 7. Several feature items in Figure 7 indicate that the process recipe of the acquired dataset is set to gas A, gas B, gas C, or gas D. Several feature items in Figure 7 also indicate that the process recipe of the acquired dataset is set to gas emission order A / B / C, A / B / C / D, A / C / B / C, A / D / C / B / C, A / C / A / C / B / C, and A / D / A / C / B / C). Several feature items in Figure 7 also indicate that the process recipe of the acquired dataset is set to STEP number "4" or "5". Several feature items in Figure 7 indicate that the process recipe of the acquired dataset is set to plasma "DC Power" or "RF Power". Several feature items in Figure 7 indicate that the plasma "DC Power" or "RF Power" in the process recipe of the acquired dataset is set to "0 [W]" or "100 [W]".

[0050] In step S14, the display unit 62 displays a screen 1000, such as the one shown in Figure 9, which accepts the operator's selection of feature items. Figure 9 is an illustrative image of an example screen that displays multiple feature items extracted in step S12 and accepts the operator's selection of feature items to be used for classifying the dataset.

[0051] In Figure 9, screen 1000 displays the multiple feature items extracted in step S12 in the display area 1002. The operator can select the feature items to be used for classifying the dataset by operating the checkboxes displayed in the display area 1002.

[0052] In step S16, the reception unit 58 determines whether or not it has received a selection of feature items from the operator. If it has not received a selection of feature items from the operator, it proceeds from step S16 to step S24, where the reception unit 58 determines whether or not it has received an instruction from the operator to finish creating the prediction model. If it has not received an instruction from the operator to finish creating the prediction model, the reception unit 58 returns to the process in step S16. Once it receives an instruction from the operator to finish creating the prediction model, the process shown in the flowchart in Figure 5 is completed.

[0053] In step S16, if the reception unit 58 determines that it has received the selection of feature items from the operator, it proceeds to the process in step S18. In step S18, the prediction model creation unit 56 is notified of the feature items selected by the operator. The prediction model creation unit 56 creates a model equation for the prediction model using adjustment items that affect the process results.

[0054] Adjustment items that affect the process results may be set in advance or selected using machine learning or the like. The adjustment items that affect the process results may be items from the data included in the process recipe or process log (such as adjustment values ​​for the substrate processing device 10). The adjustment values ​​for the substrate processing device 10 include gas flow rate, temperature, pressure, STEP time, and the number of repetitions of a predetermined STEP.

[0055] In step S20, the prediction model creation unit 56 uses a dataset that matches the feature items selected by the operator to machine-learn the model equation of the prediction model created in step S18, thereby creating a prediction model for each classification. In this embodiment, the model equation of the prediction model can be machine-learned using a dataset classified to reflect the operator's intentions. The machine learning in step S20 is a process of adjusting the coefficients of the model equation of the prediction model so that the process result information predicted by the prediction model based on the process recipe included in the dataset that matches the feature items selected by the operator approximates the process result information included in the dataset.

[0056] In step S22, the prediction unit 60 predicts process result information using the classification-specific prediction model created in step S20. The display unit 62 displays the process result information predicted by the prediction unit 60, for example, as shown in screen 1000 of Figure 10, and then returns to the process in step S14.

[0057] Figure 10 is an illustrative diagram of an example screen that displays process result information predicted using a prediction model for each classification. In the display area 1002 of screen 1000 in Figure 10, the checkboxes for the feature items selected by the operator in step S16 are checked. Furthermore, screen 1000 in Figure 10 shows an example of displaying the process result information predicted using a prediction model for each classification in a graph.

[0058] In the screen 1000 of Figure 10, process result information predicted using a prediction model for each classification is shown as an example of film thickness. The graph shows the relationship between the predicted value of film thickness predicted using the prediction model for each classification and the actual measured value of film thickness measured by the measuring device 11. Below the graph, the feature items selected by the operator and the created prediction models for each classification are displayed.

[0059] In Figure 10, screen 1000 displays graphs corresponding to the prediction models "model1-1" and "model1-2" for each classification, but it is also acceptable to display only the graph corresponding to the prediction model "model1-1" for each classification. The graph corresponding to the prediction model "model1-2" for each classification may also be a graph corresponding to a new classification prediction model that has been machine-learned using outliers.

[0060] Figure 11 is a flowchart of an example of the process for creating a new classification prediction model using machine learning with outliers. The process in the flowchart of Figure 11 is executed, for example, in step S20 of Figure 5.

[0061] In step S30, the prediction model creation unit 56 acquires a dataset that matches the feature items selected by the operator. In step S32, the prediction model creation unit 56 uses the dataset acquired in step S32 to machine-learn the model formula of the prediction model and creates a prediction model for each classification.

[0062] In step S34, the prediction model creation unit 56 determines whether or not there are outliers in the process result information predicted using the prediction models created for each classification. If there are no outliers, the prediction model creation unit 56 terminates the process shown in Figure 11. If there are outliers, the prediction model creation unit 56 proceeds to the process in step S36, uses the outlier dataset to machine-learn the model equation of the prediction model, and creates a new prediction model for a new classification.

[0063] In the screen 1000 of Figure 10, the operator can further select feature items to be used for classifying the dataset by operating the checkboxes displayed in the display area 1002.

[0064] In step S16, if the reception unit 58 determines that it has received the selection of feature items from the operator, it proceeds to the processing in steps S18 to S22. The prediction unit 60 predicts process result information using the prediction model for each classification created in step S20. The display unit 62 displays the process result information predicted by the prediction unit 60, for example, as shown in screen 1000 in Figure 12, and then returns to the processing in step S14.

[0065] Figure 12 is an illustrative image of an example screen that displays process result information predicted using prediction models for each further subdivided classification. Display field 1002 in Figure 12 has the checkbox for the feature item "Gas release order," which was additionally selected by the operator, checked in display field 1002 in Figure 10. Also, screen 1000 in Figure 12 shows an example of displaying process result information predicted using the classification-specific prediction models "model1-1," "model1-2," and "model1-3" in a graph.

[0066] In the screen 1000 of Figure 12, process result information predicted using a prediction model for each classification is shown as an example of film thickness. The graph shows the relationship between the predicted value of film thickness predicted using the prediction model for each classification and the actual measured value of film thickness measured by the measuring device 11. Below the graph, the feature items selected by the operator and the created prediction models for each classification are displayed.

[0067] Furthermore, in the screen 1000 of Figure 12, a graph corresponding to the predictive model "model 1-1," which was trained using machine learning on a dataset containing a process recipe with a gas release sequence of "A / C / B / C," is displayed. Also in the screen 1000 of Figure 12, a graph corresponding to the predictive model "model 1-2," which was trained using machine learning on a dataset containing a process recipe with a gas release sequence of "A / C / A / C / B / C," is displayed. In addition, in the screen 1000 of Figure 12, a graph corresponding to the predictive model "model 1-3," which was trained using machine learning on a dataset containing a process recipe with a gas release sequence of "A / D / A / B / C," is displayed.

[0068] Note that the graphs on screen 1000 shown in Figures 10 and 12 are examples. For example, the graph displayed on screen 1000 may be a graph showing the relationship between the predicted resistance value predicted using a prediction model for each classification and the actual measured value of the film thickness measured by the measuring device 11, as shown in Figure 13. Also, below the graph, the feature items selected by the operator and the prediction models for each classification that were created are displayed. In the graph of Figure 13, the relationship between film thickness and resistance is represented by a hyperbola.

[0069] The operator can create a highly accurate predictive model for the intended classification by repeatedly changing the selection of feature items in the display area 1002 of Figure 10, Figure 12, or Figure 13.

[0070] Furthermore, the operator may use a prediction model for the intended classification to predict the process result information of the substrate processing apparatus 10 that has executed the process according to the specified process recipe, and display it, for example, as shown in the screen 1100 of Figure 14. Figure 14 is an illustrative diagram of an example of a screen that displays the process result information of the substrate processing apparatus predicted using a prediction model for each classification. The screen 1100 of Figure 14 displays the process result information 1102 of the substrate processing apparatus predicted using a prediction model for each classification, and a graph 1104 showing the trend of the process result information and the predicted value.

[0071] Alternatively, the operator may use a predictive model for the intended classification to search for process recipes that satisfy the target process result information, and display them, for example, as shown in screen 1200 of Figure 15. Figure 15 is an illustrative image of an example screen that searches for and displays process recipes that satisfy the target process result information set by the operator. The process result information acquisition unit 66 of the server device 14 acquires the process result information set by the operator.

[0072] The search unit 68 of the server device 14 searches for a process recipe that satisfies the process result information targeted by the operator, using the process result information predicted by the prediction unit 60. The search unit 68 can search for a process recipe that satisfies the process result information targeted by the operator by, for example, inverse problem analysis using a prediction model for each category. The server's display unit 62 displays the process recipe 1202 found by the search unit 68, for example, as shown in screen 1200 of Figure 15. Alternatively, the search unit 68 may search for the optimal process recipe from a graph showing the relationship between the process result information predicted using a prediction model for each category and the items included in the process recipe (number of repetitions, pressure, etc.).

[0073] Although the present invention has been described above based on examples, the present invention is not limited to the above examples, and various modifications are possible within the scope of the claims. This application claims priority to Basic Application No. 2024-190687 filed with the Japan Patent Office on October 30, 2024, the entire contents of which are incorporated herein by reference.

[0074] 1 Process Result Prediction System 10 Substrate Processing Device 12 Device Controller 14 Server Device 15 Database Device 16 Operator Terminal 18, 20 Network 50 Dataset Acquisition Unit 54 Extraction Unit 56 Prediction Model Creation Unit 58 Reception Unit 60 Prediction Unit 62 Display Unit 64 Process Recipe Acquisition Unit 66 Process Result Information Acquisition Unit 68 Search Unit

Claims

1. A process result prediction method performed by an information processing device, comprising: extracting a plurality of feature items from a process recipe of a dataset associated with a process recipe, a process log of a substrate processing device that executed a process according to the process recipe, and process result information of a substrate processing device that executed a process according to the process recipe; receiving a selection from an operator of the feature items to be used for classifying the dataset from the extracted plurality of feature items; and predicting the process result information using a prediction model for each classification that has been machine-trained using the dataset that matches the feature items selected by the operator.

2. A process result prediction method according to claim 1, further comprising: receiving selections of feature items from an operator; and further predicting process result information using a further subdivided prediction model for each classification, which has been machine-trained using the dataset matching the feature items further selected by the operator.

3. A method for predicting process results according to claim 1 or 2, further comprising: creating a model equation for a prediction model for each classification using adjustment items that affect the process results; and creating a prediction model for each classification by machine learning the model equation using the dataset that matches the feature items selected by the operator.

4. The process result prediction method according to claim 3, further comprising: using outliers included in the process result information predicted by the prediction model to machine-learn the model formula and create a new classification prediction model.

5. The process result prediction method according to claim 1 or 2, further comprising displaying the process result information predicted by the prediction model based on the characteristic items selected by the operator in a graph.

6. The process result prediction method according to claim 5, further comprising displaying the process result information predicted by the prediction model in a graph of a different type of graph.

7. The process result prediction method according to claim 1 or 2, wherein the characteristic item includes the gas release sequence of the gas used in the process.

8. The process result prediction method according to claim 1 or 2, further comprising using the prediction model for each classification to search for the process recipe that satisfies the process result information targeted by the worker.

9. The process result prediction method according to claim 1 or 2, wherein the process result information is the processing result on the substrate to be processed.

10. An information processing device comprising: an extraction unit that extracts a plurality of feature items from a process recipe in a dataset associated with a process recipe, a process log of a substrate processing device that has executed a process according to the process recipe, and process result information of a substrate processing device that has executed a process according to the process recipe; a reception unit that receives from an operator the selection of feature items to be used for classifying the dataset from the plurality of extracted feature items; and a prediction unit that predicts the process result information using a prediction model for each classification that has been machine-trained using the dataset that matches the feature items selected by the operator.

11. A program that causes an information processing device to execute an extraction procedure for extracting multiple feature items from a process recipe in a dataset which is associated with a process recipe, a process log of a substrate processing device that has executed a process according to the process recipe, and process result information of a substrate processing device that has executed a process according to the process recipe; an acceptance procedure for receiving a selection from an operator of the multiple feature items extracted to be used for classifying the dataset; and a prediction procedure for predicting the process result information using a prediction model for each classification that has been machine-trained using the dataset which matches the feature items selected by the operator.

Images