Search method, search device, computer, semiconductor device manufacturing system, search method, and analysis method

Abstract

The purpose of the present invention is to provide a technology capable of making a search more optimal.　One of search methods according to the present invention is a search method for inferring, by way of machine learning, a case in which a recipe defined by a recipe file is to be executed by using a processing device of a semiconductor substrate, the search method being characterized by: in the machine learning, using a machine learning model that receives numerical data describing the recipe as an input, outputs inference data indicating the inferred result of processing of the semiconductor substrate, and is trained by using training data including the numerical data and data indicating the execution result obtained by executing the recipe; and, in generating the training data, integrating the numerical data and the data indicating the execution result into a data set, and further performing conversion processing for converting the structure of the recipe file into a structure corresponding to the machine learning model.

Description

Search method, search device, computer, semiconductor device manufacturing system, search method and analysis method

[0001] The present invention relates to a search method, a search apparatus, a computer, a semiconductor device manufacturing system, a search method, and an analysis method.

[0002] AI (artificial intelligence) is being used in a variety of fields to perform machine learning and autonomously find data rules and patterns from given data.

[0003] For example, in the field of semiconductor manufacturing, machine learning is used to search for recipe parameters (hereinafter referred to as recipe search) in general semiconductor processing equipment (hereinafter also referred to as "processing equipment") (e.g., plasma etching equipment, film formation equipment, CMP (Chemical Mechanical Polishing) equipment, etc.).

[0004] For example, Patent Document 1 discloses a searching device in which a parameter compression unit compresses first input parameter values ​​so that they can be restored by a parameter restoration unit, and generates first compressed input parameter values ​​with a reduced number of control parameters; a model learning unit learns a predictive model using learning data that is a pair of the first compressed input parameter values ​​and first output parameter values ​​that are processing results obtained by providing the first input parameter values ​​to a processing device as multiple control parameters; and a processing condition searching unit uses the predictive model to estimate second compressed input parameter values ​​that correspond to a target output parameter value.

[0005] International Publication No. 2021 / 111511

[0006] To operate a search device, the procedure involves creating a dataset (preprocessing), building a model, and applying the model (extracting a recipe). The technology disclosed in Patent Document 1 leaves room for improvement when it comes to preprocessing data and selecting meaningful data as features (explanatory variables). The present invention aims to provide a technology that can further optimize searches.

[0007] In order to solve the above-mentioned problems, one representative search method of the present invention is a search method that uses machine learning to estimate when a recipe defined by a recipe file is executed using a semiconductor substrate processing apparatus, wherein the machine learning uses a machine learning model in which numerical data describing the recipe is input and estimated data indicating the estimated results of the semiconductor substrate processing is output, and learning is performed using learning data including the numerical data and data indicating the execution results of the recipe, and when generating the learning data, a data set including the numerical data and the data indicating the execution results is integrated, and a conversion process is further performed to convert the structure of the recipe file into a structure corresponding to the machine learning model.

[0008] According to the present invention, it is possible to make the search more optimal. Problems, configurations and effects other than those described above will become apparent from the following description of the preferred embodiments of the present invention.

[0009] FIG. 1 is a diagram showing an example of a recipe for an etching apparatus. FIG. 2 is a diagram showing an example of a modeling target used in machine learning recipe search for an etching apparatus. FIG. 3 is a diagram showing an example of a machine learning recipe search method. FIG. 4 is a diagram showing data used in machine learning and processing performed on the data. FIG. 5 is a diagram showing a system according to a first embodiment to which a recipe search method is applied. FIG. 6 is a diagram showing an example of the configuration of an AP server of a machine learning system. FIG. 7 is a diagram showing an example of a recipe. FIG. 8 is a diagram showing an actual processing procedure of the search method according to the first embodiment. FIG. 9 is a flowchart ( FIG. 9( a) ) showing details of data structure conversion (step S102 in FIG. 8 ) of the search method according to the first embodiment, and a flowchart ( FIG. 9( b) ) showing the process of registering and updating a data structure conversion method performed before reading definition data (step S101 in FIG. 8 ). FIG. 10 is a diagram showing a first type of data structure conversion. FIG. 11 is a diagram showing a second type of data structure conversion. FIG. 12 is a diagram showing a third type of data structure conversion. FIG. 13 is a diagram showing a fourth type of data structure conversion. FIG. 14 is a diagram showing a fifth type of data structure conversion. FIG. 15 is a diagram showing a sixth type of data structure conversion. FIG. 16 is a diagram showing an example of data structure conversion. FIG. 17 is a diagram showing a second embodiment of a system to which a recipe search method is applied. FIG. 18 is a diagram showing a pre-processing procedure of the search method of the second embodiment. FIG. 19 is a diagram showing an example of a recipe file configuration of the second embodiment. FIG. 20 is a diagram showing a third embodiment of a system to which a recipe search method is applied. FIG. 21 is a diagram showing a pre-processing procedure of the search method of the third embodiment. FIG. 22 is a diagram showing an example of a recipe file configuration of the third embodiment. FIG. 23 is a diagram showing a fourth embodiment of a system to which a recipe search method is applied. FIG. 24 is a diagram showing an example of a recipe file configuration of the fourth embodiment. FIG. 25 is a diagram showing an actual processing procedure of the search method of the fifth embodiment. FIG. 26 is a diagram showing an example of a recipe file configuration of the fifth embodiment.

[0010] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Note that the present invention is not limited to this embodiment. In addition, in the description of the drawings, the same parts are designated by the same reference numerals.

[0011] In this disclosure, a recipe refers to a set of data describing setting values ​​for operating a processing device. A processing device is composed of components that perform the processing function, and a recipe holds setting values ​​for operating each component. AI refers to information processing technologies such as programs and systems that operate in a manner similar to human thought processes. Machine learning refers to AI that realizes the equivalent of human learning. It discovers patterns and rules from input data (learning data, training data) and, when applied to new data, enables prediction and classification of the new data. Features refer to data to be input into a machine learning model. The term "features" also includes parameters indicating the output of a gas supply source, raw material supply source, heat source, electron beam source, electromagnetic wave source, light source, power source, sound source, etc. In the following description, a personal computer (PC) is used for explanation, but is not limited to a PC terminal. Virtual domains, servers, and mobile terminals may also be used.

[0012] [Prior Art] (Example of Recipe for Etching Apparatus) An example of a recipe for a processing apparatus and an example of a model are shown with reference to Figures 1 and 2. Figure 1 is a diagram showing an example of a recipe for an etching apparatus. Figure 2 is a diagram showing an example of a model used for machine learning recipe search for an etching apparatus.

[0013] When training (learning) an AI for recipe search, a data set consisting of recipe parameters and data indicating the results of executing the recipe is input to the AI ​​as training data. Recipe 1 managed in a certain etching processing device is composed of multiple CSV files as shown in FIG. 1(a). In other words, recipe 1 is defined by multiple recipe files. Each recipe file specifies the operations to be performed by the etching processing device, and includes data such as the degree of vacuum, exhaust volume, type of gas used, and amount introduced into the chamber.

[0014] These CSV files are structured according to the hardware configuration of the etching processing apparatus. Figure 1(b) shows an example of the types of gases included in Recipe 1 and the amounts introduced into the chamber. Number 10, enclosed by a dashed line, indicates that Ar gas is introduced into gas pipe 1 (referred to as "gas line" in the figure). The flow rate is 100 mL / min in step 1 and 100 mL / min in step 2. Number 13 indicates that Ar gas is introduced into gas pipe 4. The flow rate is 5 mL / min in step 1 and 0 mL / min in step 2. In this way, gases supplied to the chamber from different supply pipes are treated as separate parameters in the recipe, even if they are the same gas type, as shown in Figure 1(b).

[0015] When generating a model using the parameters specified in the recipe as is for machine learning, if the number of parameters is unnecessarily increased to treat the same Ar gas as a different parameter, the learning efficiency cannot be improved and the predictability of the model cannot be expected to improve. The data set used for machine learning needs to be structured from the perspective of plasma control. For example, even if the same type of gas is supplied to the chamber from different supply pipes, it may be preferable to treat it as a common parameter when introducing it in the same step 1.

[0016] (Example of a model used in machine learning) An example of a modeling target in machine learning is shown with reference to Fig. 2. Fig. 2 is a diagram showing an example of a modeling target used in a machine learning recipe search for an etching apparatus.

[0017] The model shown in Figure 2 shows the reaction system that occurs on a semiconductor substrate in an etching system. The semiconductor substrate (wafer) is placed on a stage inside the etching system. The stage is equipped with heating and cooling functions, and the heat generated by these functions is applied to the wafer. Electric power is also supplied to the stage, which generates a bias potential for the plasma generated in the etching system. Inside the etching system, the wafer is affected by ions, radicals, atoms and molecules, and electromagnetic waves. The wafer surface is processed by these effects.

[0018] Although an example of the modeling target is shown here, the present disclosure is not limited to this. The model is determined based on the performance of the computer processing device performing the machine learning, the type and number of recipes that serve as learning data, the accuracy required for recipe search, and the like. Furthermore, while heat, power, ions, radicals, atoms and molecules, and electromagnetic waves are shown as parameters of the modeling target, the present disclosure is not limited to this. For example, it is also possible to use factors related to the inflow of various particles, such as the amount, frequency, and angle of incidence of ions, factors related to surface reactions such as reaction and diffusion, and factors related to the outflow, such as the adsorption and desorption of molecules and atoms, as parameters. Furthermore, a model may be constructed from a microscopic perspective, or macroscopic parameters such as pressure and temperature may be used instead.

[0019] (Relationship between the modeled object and the recipe (data set) acquired in the processing apparatus) Each component in the processing apparatus operates based on a set value defined in the data group. For example, if the processing apparatus is an etching apparatus, the components include a high-frequency power supply that generates electromagnetic waves, a gas flow meter that introduces gases for generating atoms, molecules, ions, and radicals, a power supply that applies electrostatic force to the stage, a heater and cooler that heat the stage, and a vacuum exhaust device that adjusts the pressure in the chamber where wafers are processed.

[0020] For a processing device to operate normally, each data item in the data set describing the setting values ​​must be uniquely linked to a component of the processing device. In other words, the data set describing the setting values ​​stores each setting value in a form linked to a name that can uniquely identify the processing device.

[0021] When considering handling such data in machine learning, the input variables are items that can be uniquely identified within the processing device, but in this case, several issues arise.

[0022] The first issue is that, because the object of modeling in machine learning is a reaction system occurring within a processing chamber, it is preferable that the explanatory variables be variables related to the reaction system. On the other hand, if data obtained from a processing device is used as is, the input variables will become control variables of the processing device. In this case, what is modeled is the processing device, not the reaction system within the processing chamber, which goes against the original purpose of learning about the reaction system. Furthermore, because the model is based on the control variables of the processing device, it becomes difficult to interpret the model generated by machine learning (hereinafter also referred to as a "machine learning model").

[0023] The second issue is that if the variables in the processing device are treated as input variables for the machine learning model, the accuracy may generally be lower than if the variables in the reaction system are treated as input variables. This is because it is necessary to learn conversion rules for converting the variables in the processing device into variables in the reaction system, and when comparing the same training data, the training data is insufficient by that amount.

[0024] The third issue is the possibility of unnecessarily increasing the dimensionality of the training data. The control variables of the processing equipment do not necessarily correspond one-to-one to the variables related to the reaction system within the processing equipment. For example, in the example shown in Figure 1, the same type of gas pipes are not necessarily concentrated in one location within the processing equipment, but may be separated into multiple locations from a control perspective, each connected to a processing chamber. In this case, the inflow rate of each gas pipe must be controlled for processing equipment control. However, by the time the inflowing gases reach the semiconductor substrate, they are already mixed, and information about which pipe they arrived from is lost. Therefore, they should be treated as a single variable in the machine learning model of the reaction system.

[0025] To solve the above problem, it is necessary to treat the data group describing the setting values ​​in the processing device as machine learning learning data, rather than treating it as is, by performing a data structure transformation using arithmetic operations, statistical processing, and similar arithmetic calculations to convert it into variables that are in line with the reaction system occurring within the processing device, before treating it as machine learning learning data.

[0026] An example of data structure conversion will be described based on the previously mentioned example of multiple connected gas pipes of the same type shown in Figure 1. Gas pipes 1 and 4 are connected to a processing apparatus, and when a semiconductor substrate is processed in the processing apparatus, it is considered impossible to distinguish which gas pipe the Ar gas introduced from gas pipe 1 and gas pipe 4 came from when it reaches the semiconductor substrate. In this case, the flow rates of the Ar gas introduced from gas pipe 1 and the Ar gas introduced from gas pipe 4 are summed, and this sum is used as the flow rate of the introduced gas in the learning data.

[0027] Note that gas pipes 2 and 3 shown in No. 11 and No. 12 supply a gas other than Ar gas. The flow rates of these gas pipes are used as the learning data as they are if multiple pipes of the same gas are not connected. If multiple pipes of the same gas are connected, the sum of their flow rates is calculated as in the above example of Ar gas, and this is used as the learning data as the flow rate of the introduced gas.

[0028] There are many other examples of such sources, including raw material supply sources, heat sources, electron beam sources, electromagnetic wave sources, light sources, power sources, sound sources, etc. Furthermore, the way in which they are handled does not simply involve treating the total amount, but also includes expressions using the four basic arithmetic operations and the like, such as expressions that represent the difference between the amount actually introduced into the processing chamber, such as the relationship between incident waves and reflected waves from an electromagnetic wave source, expressions that represent the product or integral of the flow rate of the raw material fluid and the processing time, and expressions that represent the quotient or differential, such as the change in heater power for temperature control in the processing chamber.

[0029] In this way, the variables in the processing device are converted into appropriate variables and their set values ​​by preprocessing, and the converted data structure is output as learning data.

[0030] First Embodiment (Application of Search System to Processing Equipment) A machine learning processing procedure will be described with reference to FIG. 3 . FIG. 3 is a diagram illustrating an example of a recipe search method using machine learning. The search method uses machine learning to estimate the execution of a recipe defined by a recipe file using a semiconductor substrate processing equipment. For example, the processing result predicted to be obtained if the input recipe were executed is estimated. A machine learning model used in machine learning receives numerical data describing a recipe and outputs estimated data indicating the estimated results when the recipe is executed. Learning is performed using training data including the numerical data and data indicating the execution results of the recipe. In the present disclosure, a reaction system occurring within the processing equipment is modeled using machine learning based on data obtained from the processing equipment, and a set of data describing the setting values ​​during processing is optimized to achieve the desired processing results. In FIG. 3 , steps S1 to S5 represent the process of training the machine learning model, and steps S6 and S7 represent the process of searching for a recipe using the machine learning model.

[0031] First, data describing information required for executing processing in the processing equipment is acquired from the processing equipment for semiconductor substrates (step S1).

[0032] Specifically, the acquired data describes various setting values ​​required for execution of the processing device, restrictions that must be met during execution, and the like. The acquired data is classified into a learning data group and a restriction information group (step S2). In addition to the numerical data described in the recipe, the data includes data indicating the execution results when the recipe is executed and target values ​​for semiconductor substrate processing. For example, in the case of an etching device, the data indicating the execution results includes data indicating the shape of the semiconductor substrate after processing when the processing device processes the semiconductor substrate according to the recipe, data indicating the amount of processing that has changed as a result of the processing, and the like. Furthermore, the target values ​​for semiconductor substrate processing indicate the target state of the semiconductor substrate that is expected to be achieved when processing is performed according to the recipe. For example, in the case of an etching device, this is information indicating the surface shape profile, such as the depth and width of the trench to be formed. The data storage format is set for each processing device and is not limited to a specific storage format.

[0033] Next, when generating learning data, a data set including numerical data and data indicating execution results is integrated, and the structure of the recipe file is converted into a structure corresponding to the machine learning model. Details of the data will be described later. Each piece of data is converted into a data structure that conforms to the reaction system (Step S3). This is what is known as data structure preprocessing. Data acquisition and preprocessing are repeated until all data are ready (Step S4).

[0034] Next, a model to be used for machine learning is trained (step S5). The data that has undergone data structure conversion is used as learning data for the machine learning model. The machine learning model uses parameters in the reaction system as input variables and dimensions of the semiconductor substrate after processing by the processing device as output variables. In other words, the dimensions of the semiconductor substrate after processing by the processing device are estimated data that indicate the results of estimating the processing of the semiconductor substrate.

[0035] Then, using the trained machine learning model, the input variables are optimized so that the output predicted post-processing dimensions satisfy the constraint information (step S6). The data input to the machine learning model at this time is recipe data (hereinafter also referred to as "trial data") when executing the search method of the present disclosure. Care is also taken to ensure that the trial data also satisfies the constraint information.

[0036] After the optimization is completed, the optimized input variables are converted back into processing unit variables to generate an optimized recipe (step S7).

[0037] (Data Group Used in Machine Learning) Data used in machine learning will be described with reference to Fig. 4. Fig. 4 is a diagram showing data used in machine learning and processing performed on the data.

[0038] It includes data acquired from the recipe file, user-side restriction information 11, device-side restriction information 12, device information 13, and other information 14 (corresponding to steps S1 and S2 in FIG. 3).

[0039] Data acquired from recipe files, equipment, databases, etc. can be classified into a data group describing setting values ​​and a data group describing restriction information. The former corresponds to various setting values ​​required for the operation of the processing equipment, while the latter corresponds to data describing restrictions that must be satisfied when the processing equipment is in operation and target values ​​for processing semiconductor substrates.

[0040] The device information 13 is data describing setting values. The data describing setting values ​​holds setting values ​​for operating each component in order to operate the processing device. Specific examples include, but are not limited to, gas flow rates and temperatures. In principle, each component in the processing device operates based on the device information 13.

[0041] On the other hand, the user-side restriction information 11 and the device-side restriction information 12 are classified into data groups describing restriction information. The restriction information is further classified into two types: hardware restrictions and software restrictions.

[0042] Specifically, hardware limitations include requirements for the normal operation of the processing device, such as upper and lower limits of setting values ​​and input limitations that are determined dependently on specific setting values. Furthermore, the upper and lower limits and input limitations are not limited to being specified by specific values ​​or thresholds, but may also be specified by specific ranges. In the present disclosure, the device-side limitation information 12 corresponds to information indicating hardware limitations. The device-side limitation information 12 includes information that specifies the range within which the processing device will operate normally when the processing device operates based on the device information 13.

[0043] 1(b), gas pipes No. 10 to No. 13 are each provided with upper and lower limit values ​​for the set value of the flow rate. Furthermore, gas pipe 1 No. 10 and gas pipe 4 No. 13 share Ar gas, and when the flow rate of Ar gas supplied by these gas pipes is used as a parameter of the machine learning model, the sum of the upper limit values ​​of gas pipe 1 and gas pipe 4 is specified as the upper limit value. Note that the upper and lower limit values ​​specified here are not limited to the upper and lower limit values ​​as the limit values ​​of the apparatus specifications for each gas pipe, but can also be upper and lower limit values ​​that can ensure a predetermined level of accuracy in recipe search.

[0044] Specific software limitations include post-processing dimensions, yield, and operation rate that must be satisfied in terms of production management. These values ​​may be specified not only by specific values ​​but also by ranges, thresholds, etc. Post-processing dimensions, which are target values ​​for processing semiconductor substrates, are also included as software limitations. In this disclosure, user-side limitation information 11 corresponds to information indicating software limitations.

[0045] The other information 14 includes data indicating the execution results when the recipe is executed, as well as literature values, processing results obtained from other devices, past processing information obtained from a database, virtual data obtained by virtual measurement or a physical model, input / output information or inference results of a separately created trained machine learning model, and the like.

[0046] The classified information undergoes data structure conversion (step S20) and is divided into constraint conditions 16 and learning data 17 (corresponding to step S3 in FIG. 3). The constraint conditions 16 include target values ​​for semiconductor substrate processing included in user-side limit information 11, upper and lower limit values ​​of parameters included in equipment-side limit information 12, and the like.

[0047] Next, the learning data 17 is input to the machine learning model 18, and training of the machine learning model is performed (step S21) (corresponding to step S5 in FIG. 3).

[0048] Next, an optimized recipe is searched for (step S22). For example, input data satisfying the constraint information is generated by a program and input to the trained model 20, and the obtained output result is determined to be close to the target value for semiconductor substrate processing. The result is reflected in the next input data that satisfies the constraint information. This method is repeated to optimize the recipe. The optimized recipe is output and provided, for example, to the user (step S23).

[0049] (Examples of Application to Systems) Hereinafter, systems to which the recipe search method is applied will be described as Examples 1 to 5.

[0050] (First embodiment) (Configuration of first embodiment) The first embodiment will be described with reference to Figs. 5 to 14. Fig. 5 is a diagram showing a system according to a first embodiment to which a recipe search method is applied. The first embodiment assumes, for example, a case where a processing device is used alone. The search system 100 includes a user PC 110 １ From 110 ＮThe search system 100 includes a network (N is a positive integer), a machine learning system 120, a manufacturing data system 130, an edge PC 140, and a processing device 150. A local area network (hereinafter also referred to as "LAN") is formed in the search system 100. In the following description, it is assumed that the processing device 150 is an etching processing device, but the present disclosure is not limited to this. The present disclosure is also applicable to devices other than etching devices. The processing device 150 is defined as an etching processing device in a broad sense, including not only an etching processing device in the narrow sense for performing plasma etching, etc., but also devices related to etching processes, such as semiconductor substrate inspection devices and measurement devices. The search system 100 can also be described as a semiconductor device manufacturing system equipped with a platform to which semiconductor manufacturing devices are connected via a network and on which an application for machine learning is implemented to estimate when a recipe defined by a recipe file is executed using a semiconductor substrate processing device.

[0051] User PC 110 １ From 110 Ｎ accepts requests from users who use the search system 100 and presents the results of searches requested by the users to the users. In the following description, when referring to a user PC without specifying it, the user PC 110 is referred to as the user PC. ｎ (n is a positive integer equal to or less than N) is used. As will be described later, ｎ is a computer capable of communicating with a search device (AP server) that estimates the process when a recipe defined by a recipe file is executed using a semiconductor substrate processing device using machine learning corresponding to a model of the process performed by the processing device.

[0052] The machine learning system 120 performs machine learning and searches for recipes. The machine learning system 120 includes a web server 121, an AP (application) server 122, and a DB (database) server 123. The web server 121 is connected to the user PC 110. ｎ and sends the result according to the request to the user PC 110. ｎThe request is transmitted to the AP server 122. If the request includes dynamic processing, such as searching for a recipe, the Web server 121 transmits the request to the AP server 122. The AP server 122 receives the request transmitted from the Web server 121, performs a search in response to the request, and transmits the searched recipe to the Web server 121. Specifically, the AP server 122 is equipped with a learning device that receives numerical data defining the recipe as input and outputs estimated data that estimates the execution result of the recipe. The DB server 123 receives the request transmitted from the AP server 122, stores the data in a database in response to the request, and extracts necessary data and transmits it to the AP server 122. The DB server 123 stores, for example, the recipe used to train the machine learning model and data such as the surface shape and processing amount of the semiconductor substrate that are the execution result of the recipe.

[0053] The manufacturing data system 130 collects learning data and equipment information and transmits them to the machine learning system 120. The manufacturing data system 130 includes a web server 131, an application server 132, and a database server 133. The web server 131 receives requests sent from the application server 122 of the machine learning system 120 and transmits results corresponding to the requests to the application server 122. If the request includes dynamic processing, the web server 131 transmits the request to the application server 132. The application server 132 receives the requests sent from the web server 131, processes the requests, and transmits the results of the processing to the web server 131. The database server 133 receives requests sent from the application server 132, stores the data in a database in accordance with the requests, and extracts necessary data and transmits it to the application server 132.

[0054] The edge PC 140 processes data output from the processing device 150. The processing device 150 outputs data including the executed recipe and the measurement results of the semiconductor substrate resulting from the execution of the recipe. The edge PC 140 performs optimization processing on the data output from the processing device 150, thereby suppressing communication delays when transmitting the data to the manufacturing data system 130.

[0055] A user of the search system 100 uses a user PC 110 ｎ It is possible to search for recipes using the machine learning system 120 directly or via the machine learning system 120. Note that although an example in which the edge PC 140 and the processing device 150 are connected to the LAN of the search system 100 has been shown, the present disclosure can also be applied to cases in which the edge PC 140a and the processing device 150a are not included in the LAN. Details will be described later.

[0056] (Configuration of the AP Server of the Machine Learning System) Figure 6 is a diagram showing an example of the configuration of the AP server 122 of the machine learning system 120. The AP server 122 can also be considered a search device that estimates the execution of a recipe defined by a recipe file using a semiconductor substrate processing device through machine learning corresponding to a model of the processing performed by the processing device. The AP server 122 receives as input numerical data describing the recipe and outputs estimated data indicating the estimated results of the semiconductor substrate processing by the processing device 150. The AP server 122 includes a computer implemented with a learner that performs learning using learning data including the numerical data and data indicating the execution results, and a display that displays a user interface for setting target values ​​for the processing. When generating the learning data, the computer integrates a data set including the numerical data and data indicating the execution results, and further converts the structure of the recipe file into a structure corresponding to the model.

[0057] The AP server 122 includes a CPU (Central Processing Unit) 1220, a memory 1221, a storage 1222, and an interface 1223. The CPU 1220 functions as a computing device. The memory 1221 stores programs for performing various functions. The memory 1221 includes, for example, a data input / output block 1224, a data structure conversion block 1225, an ML model optimization block 1226, and a recipe generation block 1227. As will be described in detail later, the data input / output block 1224 has a program for acquiring data (step S100) and acquiring input variables used in step S104 of FIG. 8. The data structure conversion block 1225 has a program for reading definition data (step S101) and converting data structures (step S102) of FIG. 8. The ML model optimization block 1226 has a program for optimizing input variables of the machine learning model (step S104). The recipe generation block 1227 has a program for performing inverse conversion (step S105) of input variables into processing device variables. The storage 1222 is used to store data when, for example, transaction processing is performed by the CPU 1220. The storage 1222 is, for example, a hard disk drive (HDD).

[0058] The interface 1223 accepts requests sent from outside the AP server 122 and transmits the results of the requests to the outside. The input / output device 207 is, for example, a device operated by a user of the search system 100. The user can also operate the machine learning system 120 using a user interface displayed on a display of the input / output device 207. The interface 1223 enables communication between the AP server 122 and the Web server 121, between the AP server 122 and the DB server 123, and between the AP server 122 and the Web server 131 of the manufacturing data system 130.

[0059] Note that the above description shows an example of the configuration of the AP server 122, and the present disclosure is not limited to this. For example, if large-scale calculations are required, the AP server 122 can also be configured to include a GPU (Graphical Processing Unit) in addition to the CPU 1220.

[0060] (Example of Recipe) FIG. 7 is a diagram showing an example of a recipe. The recipe 1501 shown in FIG. 7(a) is executed by, for example, the processing device 150. The recipe 1501 shows the types of gases and electrode conditions used in processing steps 1 to 3 in the plasma processing. In processing step 1, Ar gas is used at 0 mL / min and 100 mL / min, NF ３ The type and flow rate of the gas are set, such as 10 mL / min for Ar gas and 20 mL / min for HBr gas. Here, two flow rates are shown for Ar gas because it is controlled by two outputs in the processing device 150. The high frequency power applied to the wafer is set to 10 W, and the electrode temperature is set to 10°C. In processing step 2, Ar gas is set to 20 mL / min and 0 mL / min, and NF ３ In process step 3, the flow rates of Ar gas and NF gas are set to 10 mL / min and 0 mL / min, respectively. ３ The gas flow rate is set to 20 mL / min, and the HBr gas flow rate is set to 0 mL / min. The high frequency power applied to the wafer is set to 20 W, and the electrode temperature is set to 10°C.

[0061] FIG. 7(b) is a diagram showing the upper and lower limit values ​​for control of each item indicated in the recipe. The "lower control limit value" and "upper control limit value" are information that specifies the upper and lower limit values ​​of items included in the recipe when performing data structure conversion and recipe search, which will be described later. Furthermore, "zero input possible" is information that indicates whether the gas piping can be closed and the flow rate can be controlled to 0. The flow rate of Ar gas in gas line 1 is limited to 5 mL / min or more and 50 mL / min or less, or 0 mL / min. The NF in gas line 2 is limited to 5 mL / min or more and 50 mL / min or less. ３The gas flow rates are limited to 0 mL / min or more and 50 mL / min or less, or 0 mL / min. The flow rate of HBr gas in gas line 3 is limited to 0 mL / min or more and 50 mL / min or less, or 0 mL / min. The flow rate of Ar gas in gas line 4 is limited to 50 mL / min or more and 200 mL / min or less, or 0 mL / min. Here, upper and lower limit values ​​for the gas flow rates are indicated, but settable ranges are also specified for other items included in the recipe.

[0062] (Method of Example 1) A search method in Example 1 will be described with reference to FIGS.

[0063] FIG. 8 is a diagram showing the actual processing procedure of the search method of the first embodiment. First, data describing information required for executing processing in the processing device 150 is acquired from the processing device 150 (step S100). Numerical data described in the recipe and related information (such as the recipe file name and information indicating the execution results of the recipe) are acquired. The AP server 122 of the machine learning system 120 acquires this data via the edge PC 140 and the manufacturing data system 130. When collecting data sent from the processing device 150 to the DB server 133, the AP server 122 acquires the data via the DB server 133, the AP server 132, and the Web server 131.

[0064] Next, the AP server 122 reads the definition data (step S101). The definition data is data that specifies the target and method of data structure conversion. The definition data is stored, for example, in the DB server 123. The registration and update of the definition data will be described later. The AP server 122 inquires about the target and method of data structure conversion for the acquired data.

[0065] Next, the AP server 122 performs data structure conversion on the acquired data based on the target and method of data structure conversion indicated in the definition data (step S102).

[0066] Next, the AP server 122 trains the machine learning model using the data whose data structure has been converted (step S103).

[0067] Next, the user inputs the recipe to be executed and target values ​​for the process, and searches for a recipe. The variables (input data) input to the machine learning model are optimized for the machine learning model. In addition, the input and output of the machine learning model are set to satisfy the restriction information classified from the input data (step S104).

[0068] The AP server 122 reverse-converts the input variables into variables of the processing device (recipe generation) (step S105). In other words, the searched recipe is expressed in the format of the input variables of the machine learning model. By performing reverse conversion of the data structure conversion on the searched recipe, a recipe converted into a format that describes the recipe of the processing device 150 is generated.

[0069] 9A and 9B are flowcharts showing details of the data structure conversion (step S102 in FIG. 8) in the search method of Example 1 (FIG. 9A), and flowcharts showing the process of registering and updating the data structure conversion method that is performed before reading the definition data (step S101 in FIG. 8) (FIG. 9B).

[0070] The data structure conversion (step S102 in FIG. 8) will be described with reference to FIG. 9(a). The DB server 133 stores the data from the processing device 150, and the AP server 122 reads the data from the DB server 133 (step S1020). For each piece of data that has been read, the AP server 122 determines whether the data requires data structure conversion (step S1021). The AP server 122 performs data structure conversion on all data that requires data structure conversion (step S1022).

[0071] With reference to FIG. 9B, the registration and update of the data structure conversion method defined in the definition data (step S101 in FIG. 8) will be described.

[0072] The user operates the user PC 110. ｎAlternatively, a data structure conversion method is registered and updated based on the machine learning model and the numerical data described in the recipe via the interface 1223 of the AP server 122. Here, the data structure conversion method is, for example, one of the first to sixth types described below. The type is registered by associating the type of data with the corresponding type.

[0073] The AP server 122 registers the target and method of data structure conversion (step S1001 in FIG. 9B). Step S1001 is performed for all data (features) that can be obtained from the recipe. The registered data and method may be stored, for example, in a storage device (memory 1221) of the AP server 122, or may be stored in the DB server 123.

[0074] (Types of Data Structure Conversion) Data structure conversion will be described with reference to FIGS. 10 to 15. A recipe file has a tabular data structure including the numerical data and labels attached to the numerical data. Through data structure conversion, the numerical data indicated in the recipe is converted into parameters and numerical values ​​suitable for recipe search using a machine learning model. The data structure conversion process includes at least one of adding or deleting numerical data, performing arithmetic operations on values, grouping, dividing, and replacing.

[0075] FIG. 10 illustrates a first type of data structure transformation. FIG. 10(a) shows a schematic diagram of the data structure transformation, and FIG. 10(b) shows a specific example of numerical data acquired from a recipe. In the first type, there is no direct input to the data structure transformation, but feature values ​​are output through indirect input. For example, a case where a correction for each device is substituted into the numerical data as a correction term can be considered. Data 51 before the data structure transformation does not include numerical data. Data 52 after the data structure transformation is numerical data added without input. As a specific example, as shown in FIG. 10(b), before the data structure transformation, no numerical data is acquired for a specific feature 1, and no numerical data is included in rows #1 to #10. After the data structure transformation, numerical data is added to the specific feature 1, and the values ​​20, 30, 100, 70, 30, 90, 10, 30, 80, and 90 are substituted in rows #1 to #10, respectively. These values ​​can also be added, for example, to correction terms that have been previously registered.

[0076] FIG. 11 illustrates a second type of data structure transformation. FIG. 11(a) schematically illustrates the data structure transformation, and FIG. 11(b) illustrates a specific example of numerical data obtained from a recipe. In the second type, there is input to the data structure transformation, but the numerical data is deleted for some reason. For example, this may be the case when the recipe contains numerical data unrelated to the modeling target. Data 53 before the data structure transformation contains numerical data, but data 54 after the data structure transformation has the numerical data deleted. As a specific example, as shown in FIG. 11(b), before the data structure transformation, predetermined feature 1 contains the values ​​20, 30, 100, 70, 30, 90, 10, 30, 80, and 90 in rows #1 to #10, respectively. After the data structure transformation, predetermined feature 1 has been deleted, and the numerical values ​​have been deleted in rows #1 to #10. Feature 1 was not necessary for training the machine learning model, so it is deleted by the data structure transformation.

[0077] FIG. 12 illustrates a third type of data structure conversion. FIG. 12(a) shows a schematic diagram of the data structure conversion, and FIG. 12(b) shows a specific example of numerical data acquired from a recipe. In the third type, numerical data before and after data structure conversion are in one-to-one correspondence. For example, this can be used when changing the name of a feature or converting the units of numerical data. Data 56 before data structure conversion is named Feature A, while data 56 after data structure conversion is named Feature 1. As a specific example, as shown in FIG. 12(b), before data structure conversion, feature A includes numerical data of 20, 30, 100, 70, 30, 90, 10, 30, 80, and 90 in rows #1 to #10, respectively. After the data structure conversion, feature 1 contains the following numerical data in rows #1 to #10: 0.02, 0.03, 0.1, 0.07, 0.03, 0.09, 0.01, 0.03, 0.08, and 0.09, respectively. The unit of feature a is 1000 times the unit of feature 1. For unit conversion, the numerical data of feature a is multiplied by a constant 0.001.

[0078] FIG. 13 illustrates a fourth type of data structure transformation. FIG. 13(a) shows a schematic diagram of the data structure transformation, and FIG. 13(b) shows a specific example of numerical data acquired from a recipe. In this fourth type, two or more pieces of numerical data before the data structure transformation are aggregated into one piece of numerical data after the data structure transformation (so-called numerical data grouping). For example, a case can be considered in which a feature, which is managed as multiple variables in the processing device but is preferably captured as a single variable from the perspective of the reaction system, is handled. Data 57a and data 57b before the data structure transformation are two pieces of data, feature a and feature b, respectively. On the other hand, data 58 after the data structure transformation is a single piece of data, feature 1. As a specific example, as shown in FIG. 13(b), before the data structure transformation, feature a includes numerical data of 20, 30, 100, 70, 30, 90, 10, 30, 80, and 90 in rows #1 to #10, respectively. Furthermore, feature b contains numerical data of 10, 30, 90, 60, 30, 70, 40, 30, 60, and 70 in rows #1 to #10, respectively. After data structure conversion, feature 1 contains numerical data of 30, 60, 190, 130, 60, 160, 50, 60, 140, and 160 in rows #1 to #10, respectively. Feature a and feature b have the same function in the reaction system and can be treated as synonyms. For this reason, feature a and feature b are the sum of the same components and are treated as feature 1.

[0079] FIG. 14 illustrates a fifth type of data structure conversion. FIG. 14(a) shows a schematic diagram of the data structure conversion, and FIG. 14(b) shows a specific example of numerical data acquired from a recipe. In the fifth type, data before the data structure conversion was in a storage format including delimiters, but after the data structure conversion, the data is separated into tabular data. For example, a recipe file may be handled in which multiple numerical data are separated by commas or the like in one row. Data 59 before the data structure conversion is text data containing one delimiter per row. On the other hand, data 60a and 60b after the data structure conversion are tabular data each containing one numerical value per row. As a specific example, as shown in FIG. 14B , before the data structure conversion, feature a includes text data in rows #1 to #10, each containing two numbers separated by a comma, such as "70,100," "10,30," "50,80," "40,70," "70,50," "10,50," "40,50," "40,60," "20,60," and "10,80." After the data structure conversion, feature 1 includes numeric data of 70, 10, 50, 40, 70, 10, 40, 40, 20, and 10 in rows #1 to #10, respectively. Furthermore, feature 2 includes numeric data of 100, 30, 80, 70, 50, 50, 50, 60, 60, and 80 in rows #1 to #10, respectively. Since feature a is text data separated by commas, it is separated by commas and the text data is converted into numerical data, and feature 1 and feature 2 are generated in a table format.

[0080] FIG. 15 illustrates a sixth type of data structure transformation. FIG. 15(a) illustrates a schematic diagram of the data structure transformation, and FIG. 15(b) illustrates a specific example of numerical data acquired from a recipe. In the sixth type, two or more pieces of data before the data structure transformation are combined into two or more pieces of numerical data after the data structure transformation, and the numerical data are replaced based on a predetermined rule, such as a correspondence table. For example, past data is referenced and the numerical data is replaced with features obtained by matching the past data with the current data. Data 61a and 61b before the data structure transformation are combined and replaced based on a predetermined rule, and data 62a and 62b are generated after the data transformation. As a specific example, FIG. 15(b) illustrates a correspondence table of features before the data structure transformation. As shown in correspondence table 61X, feature x corresponds to numerical data of 20, 30, 100, 70, 30, 90, 10, 30, 80, and 90 in rows #1 to #10, respectively. As shown in correspondence table 61A, when feature quantity x is 10, 20, 30, 40, 50, 60, 70, 80, 90, and 100, feature quantity a corresponds to the numerical data of 40, 10, 30, 0, 20, 10, 60, 60, 70, and 90, respectively. As shown in correspondence table 61B, when feature quantity x is 10, 20, 30, 40, 50, 60, 70, 80, 90, and 100, feature quantity b corresponds to the numerical data of 50, 10, 30, 0, 0, 40, 20, 90, 20, and 90, respectively. In correspondence table 61X, feature quantity x in row #1 is 20. When feature quantity x=20, feature quantity a is 10 (second row of correspondence table 61A) and feature quantity b is 10 (second row of correspondence table 61B). Correspondence table 62 shows the data generated after the data structure conversion. As shown in correspondence table 62, in row #1, feature quantity a (10) is shown as feature quantity 1, and feature quantity b (10) is shown as feature quantity 2. Similarly, in correspondence table 61X, feature quantity x in row #2 is 30. When feature quantity x = 30, feature quantity a is 30 (row 3 of correspondence table 61A), and feature quantity b is 30 (row 3 of correspondence table 61B). In row #2 of correspondence table 62, feature quantity a (30) is shown as feature quantity 1, and feature quantity b (30) is shown as feature quantity 2.When this conversion process is repeated for correspondence tables 61X, 61A, and 61B, feature quantity 1 corresponds to 10, 30, 90, 60, 30, 70, 40, 30, 60, and 70 in rows #1 to #10, respectively, as shown in correspondence table 62, and feature quantity 2 corresponds to 10, 30, 90, 20, 30, 20, 50, 30, 90, and 20 in rows #1 to #10, respectively. In this way, numerical data is combined and replaced through data structure conversion.

[0081] A specific example of the sixth type is the relationship between the input power to a heater in a vacuum processing apparatus and the temperature of the inner wall of the vacuum processing apparatus. The temperature of the inner wall during operation of the vacuum processing apparatus is one of the factors that affect the processing results. During operation, the interior of the vacuum processing apparatus is sealed to maintain a vacuum state. To know the temperature of the inner wall of the vacuum processing apparatus, a measuring instrument must be attached to the inner wall. However, it may be difficult to operate the vacuum processing apparatus with the measuring instrument attached to the inner wall when the vacuum processing apparatus is in operation, for example, due to exposure of the interior of the vacuum processing apparatus to plasma discharge. Therefore, the relationship between the input power to the heater and the inner wall temperature is acquired in advance without plasma discharge. For example, if the input power is the feature x, the actual measured temperature (actual measured temperature value) of the upper part of the inner wall of the vacuum processing apparatus when the first heater is in operation is the feature a, and the actual measured temperature of the lower part of the inner wall of the vacuum processing apparatus when the second heater is in operation is the feature b, then the relationship between the input power of the first heater and the actual measured temperature of the inner wall is shown in correspondence table 61A, and the relationship between the input temperature of the second heater and the inner wall temperature is shown in correspondence table 61B. Furthermore, the input power to the first heater and the second heater during actual operation of the vacuum processing apparatus in which plasma discharge has been performed is measured, for example, in chronological order, and values ​​are obtained as shown in correspondence table 61X. By performing data structure conversion, for example, feature 1 can be generated as an estimated temperature (estimated temperature value) of the upper part of the inner wall of the vacuum processing apparatus, and feature 2 can be generated as an estimated temperature of the lower part of the inner wall of the vacuum processing apparatus. Feature 1 and feature 2 can be applied as display values ​​of measuring instruments. By performing data structure conversion of numerical data as shown in types 1 to 6, recipe files are combined when generating training data.

[0082] (Example of Data Structure Conversion) Fig. 16 is a diagram showing an example of data structure conversion. Fig. 16 shows a case where data structure conversion of the fourth type of data summation is performed on the recipe 1501 shown in Fig. 7(a). As shown in Fig. 7(a), Ar gas is managed as two outputs in the processing device 150. From the viewpoint of the reaction system, it is desirable to treat it as a single variable, so the duplicated item Ar is combined into one item and the flow rate values ​​are summed.

[0083] (Functions and Effects of Example 1) When the numerical data described in a recipe is directly input to a machine learning model, if the machine learning model represents a reaction system within a processing device, it is difficult to efficiently train the model and improve the accuracy of recipe search. In Example 1, the data structure of the information acquired from the recipe is converted to match the machine learning model, thereby improving the efficiency of learning and the accuracy of recipe search.

[0084] (Modification of First Embodiment) When a recipe search is performed for the edge PC 140a and the processing device 150a, which are so-called offline devices that are not included in the LAN, the learning data is transmitted to the user PC 110 via the external recording medium 160. ｎ The user PC 110 ｎ sends the training data to the machine learning system 120, where machine learning is performed.

[0085] (Example 2) Example 2 will be described with reference to Figures 17 to 19. Example 2 assumes, for example, a case in which there are multiple processing devices, the data specifications in the processing devices are unified, and the processing devices are connected to an external LAN. Example 2 differs from Example 1 in that it uses training data acquired from multiple processing devices. In the following description, components that are the same as or equivalent to those in Example 1 described above will be assigned the same reference numerals, and their description will be simplified or omitted.

[0086] 17 is a diagram showing a system according to a second embodiment to which a recipe search method is applied. The search system 200 includes LAN A and LAN B. A machine learning system 220 and a user PC 210 in LAN B １ From 210 Ｎ are the machine learning system 120 and the user PC 110 in the first embodiment, respectively. １ From 110 Ｎ The manufacturing data system 230 in Lan A corresponds to the manufacturing data system 130 in the first embodiment.

[0087] In Lan A, Edge PC 240 １ From 240 Ｑ (Q is a positive integer) corresponds to the edge PC 140 in the first embodiment. １ From 250 Ｑ Each of these corresponds to the processing device 150 in the first embodiment. In the following description, when referring to an edge PC and a processing device without specifying, the edge PC 240 ｑ (q is a positive integer equal to or less than Q) and a processing unit 250 ｑ is used.

[0088] The manufacturing data system 230 in Lan A and the machine learning system 220 in Lan B are linked via an API (Application Programming Interface). The AP server 222 can use services provided by the Web server 231.

[0089] (Method of the Second Embodiment) A search method in the second embodiment will be described.

[0090] The actual processing procedure is the same as that of the first embodiment shown in FIG. 8, and therefore will be described with reference to FIG. 8. ｑ From the processing device 250 ｑThe AP server 222 of the machine learning system 220 acquires data describing information required for executing the process in the edge PC 240 (corresponding to step S100 in FIG. 8). The AP server 222 acquires the numerical data described in the recipe and related information (such as the file name of the recipe and information indicating the execution result of the recipe). ｑ These data are acquired via the manufacturing data system 230. ｑ When collecting data sent from the database server 233 to the web server 231, the application server 222 acquires the data via the database server 233, the application server 232, and the web server 231. １ From 250 Ｑ It is possible to obtain numerical data and related information.

[0091] Next, the AP server 222 reads the definition data (corresponding to step S101 in FIG. 8). The AP server 222 inquires about the target and method of data structure conversion for the acquired data.

[0092] Next, the AP server 222 performs data structure conversion on the acquired data based on the target and method of data structure conversion indicated in the definition data (corresponding to step S102 in FIG. 8). The data structure conversion is performed according to the procedure of the flowchart showing the details of data structure conversion shown in FIG. 9(a).

[0093] Next, the AP server 222 trains the machine learning model using the data whose data structure has been converted (corresponding to step S103 in FIG. 8).

[0094] Next, the user inputs the recipe and target values ​​for the process to be performed and searches for a recipe. The variables (input data) input to the machine learning model are optimized for the machine learning model. The input and output of the machine learning model are also adjusted to satisfy the restriction information classified from the input data (corresponding to step S104 in FIG. 8 ).

[0095] The AP server 222 reversely converts the input variables into variables of the processing device (recipe generation) (corresponding to step S105 in FIG. 8). In other words, the searched recipe is expressed in the form of input variables of the machine learning model. By performing reverse conversion of the data structure conversion on the searched recipe, the processing device 250 ｑ A recipe is generated that is converted into a format that describes the recipe.

[0096] 18 is a diagram showing a pre-processing procedure of the search method of the second embodiment. FIG. 18 is a flowchart showing the pre-processing procedure, which is performed before reading the definition data (step S101 in FIG. 8), and the process of registering and updating the data structure conversion method. The user operates the user PC 210. ｎ Alternatively, the data structure conversion method is registered and updated via the interface of the AP server 222 based on the machine learning model and the numerical data described in the recipe.

[0097] The data structure conversion method is registered and updated by the processing unit 250. １ From 250 Ｑ This is carried out for all target processing equipment.

[0098] The AP server 222 registers and updates the data structure conversion method (step S200). The registered method may be stored in the storage device of the AP server 222, or may be stored in the DB server 223.

[0099] (Configuration Example of Recipe File in Second Embodiment) FIG. 19 is a diagram showing a configuration example of a recipe file in the second embodiment.

[0100] The recipe information 2400 is data acquired from the LAN A, for example, the edge PC 240 ｑ The recipe information 2400 can be acquired from the DB server 233. The recipe information 2400 is a recipe with a process ID of A001. The recipe information 2400 specifies pressure in Pa units, with the pressure in process step 1 set to 1.0 Pa, the pressure in process step 2 set to 0.8 Pa, and the pressure in process step 3 set to 0.8 Pa.

[0101] The recipe information 2230 is data acquired from the LAN B, and can be acquired from, for example, the DB server 223. The recipe information 2230 is stored in the user PC 210. １ ~210 Ｎ The recipe information 2230 is a recipe with a process ID of B001. The recipe information 2230 specifies the pressure in mTorr, with the pressure set to 7.5 mTorr in process step 1, 6 mTorr in process step 2, and 4.5 mTorr in process step 3.

[0102] Merged data 2220 shows a list of items included in recipe information 2400 and recipe information 2230. Both recipe information 2400 and recipe information 2230 indicate processes for which pressure is set, so "Pressure" is written in the "Item Name." Additionally, both Pa and mTorr are written in the "Units," with numerical values ​​listed for each process ID. The table shows duplicated pressure settings in different units, indicating that the settings are not normalized.

[0103] Data after data structure conversion 2221 shows the result when data structure conversion is performed on post-merged data 2220. Only Pa is entered in "unit." In post-merged data 2220, B001 in processing step 1 was 7.5 mTorr, but in data after data structure conversion 2221, this has been converted to 1.0 Pa. Here, a third type of data structure conversion is performed, and specifically, the numeric value in mTorr units is multiplied by a constant 1 / 7.5 to convert units. Similar data structure conversions are also performed for processing steps 2 and 3.

[0104] In the second embodiment, the machine learning model can be trained using data acquired from multiple processing devices, thereby improving the efficiency of learning and the accuracy of recipe search. In addition, since data from processing devices located on a LAN different from the LAN on which the machine learning system 220 is located can be utilized, it is possible to build a search system between different business locations or between a company's system and a system of another company, for example.

[0105] (Example 3) Example 3 will be described with reference to Figures 20 to 22. Example 3 is assumed to include, for example, literature values, databases, and other data other than semiconductor manufacturing equipment in the dataset of the machine learning model. Example 3 differs from Example 1 in that the dataset of the machine learning model also includes data acquired from literature, databases, and other data other than semiconductor manufacturing equipment. In the following description, components that are the same as or equivalent to those in Examples 1 and 2 described above are denoted by the same reference numerals, and their description will be simplified or omitted.

[0106] 20 is a diagram showing a system according to a third embodiment to which a recipe search method is applied. The search system 300 includes LANA, LANB, and LANC. A machine learning system 320 and a user PC 310 in LANC. １ From 310 Ｎ are the machine learning system 120 and the user PC 110 in the first embodiment, respectively. １ From 110 Ｎ The manufacturing data system 330 in LanB corresponds to the manufacturing data system 130 in the first embodiment.

[0107] In Lan B, Edge PC 340 １ From 340 Ｑ (Q is a positive integer) corresponds to the edge PC 140 in the first embodiment. １ From 350 Ｑ Each of these corresponds to the processing device 150 in the first embodiment. In the following description, when referring to an edge PC and a processing device without specifying, the edge PC 340 ｑ (q is a positive integer equal to or less than Q) and a processing unit 350ｑ is used.

[0108] The literature management system 360 is a system that manages literature information. The web server 361 accepts requests from the LANC AP server 322. The AP server 362 generates data based on the processing request from the web server 361. If necessary, the AP server 362 requests data from the database server 363. The database server 363 stores information such as specifications, patent documents, academic literature, and research records. In response to a data request from the AP server 362, the database server 363 provides data to the AP server 362.

[0109] The machine learning system 370 includes a processing unit 350 １ From 350 Ｍ The Web server 371 receives a request from the AP server 322 of the LanC. The AP server 372 generates data based on the processing request from the Web server 371. If necessary, the AP server 372 requests data from the DB server 373. The DB server 373 receives data from the processing device 350. １ From 350 Ｍ The DB server 373 stores information (such as a trained machine learning model) related to machine learning performed based on data acquired from the AP server 372. The DB server 373 provides data to the AP server 372 in response to a data request from the AP server 372.

[0110] The AP servers 322 and 372 have the same configuration as the AP server 122 of the first embodiment. The AP servers 322 and 372 are capable of training machine learning models and searching for data.

[0111] The manufacturing data system 330 in Lan B, the literature management system 360 and machine learning system 370 in Lan A, and the machine learning system 320 in Lan C are linked via an API (Application Programming Interface). The AP server 322 can use services provided by the web servers 331, 361, and 371.

[0112] (Method of the Third Embodiment) A search method in the third embodiment will be described.

[0113] The actual processing procedure is the same as that of the first embodiment shown in FIG. 8, and therefore will be described with reference to FIG. 8. ｑ From the processing device 350 ｑ The AP server 322 of the machine learning system 320 acquires data describing information required for executing the process in the edge PC 340 (corresponding to step S100 in FIG. 8). The AP server 322 acquires the numerical data described in the recipe and related information (such as the file name of the recipe and information indicating the execution result of the recipe). ｑ These data are acquired via the manufacturing data system 330. ｑ When collecting data sent from the database server 333 to the web server 331, the application server 322 acquires the data via the database server 333, the application server 332, and the web server 331. １ From 350 Ｑ The AP server 322 can acquire the above-mentioned numerical data and related information. The AP server 322 also makes requests to the Web server 371 and the Web server 361. The Web server 371 generates information in response to the request in the AP server 372, or acquires information from the DB server 373 and sends it to the AP server 322. The Web server 361 also acquires information from the DB server 363 via the AP server 362 and sends it to the AP server 322.

[0114] Next, the AP server 322 reads the definition data (corresponding to step S101 in FIG. 8). The AP server 322 inquires about the target and method of data structure conversion for the acquired data.

[0115] Next, the AP server 322 performs data structure conversion on the acquired data based on the target and method of data structure conversion indicated in the definition data (corresponding to step S102 in FIG. 8). The data structure conversion is performed according to the procedure of the flowchart showing the details of data structure conversion shown in FIG. 9(a).

[0116] Next, the AP server 322 trains the machine learning model using the data whose data structure has been converted (corresponding to step S103 in FIG. 8).

[0117] Next, the user inputs the recipe and target values ​​for the process to be performed and searches for a recipe. The variables (input data) input to the machine learning model are optimized for the machine learning model. The input and output of the machine learning model are also adjusted to satisfy the restriction information classified from the input data (corresponding to step S104 in FIG. 8 ).

[0118] The AP server 322 reversely converts the input variables into variables of the processing device (recipe generation) (corresponding to step S105 in FIG. 8). In other words, the searched recipe is expressed in the form of input variables of the machine learning model. By performing reverse conversion of the data structure conversion on the searched recipe, the processing device 350 ｑ A recipe is generated that is converted into a format that describes the recipe.

[0119] Fig. 21 is a diagram showing a pre-processing procedure in the search method of Example 3. Fig. 21 is a flowchart showing the process of registering and updating a data structure conversion method that is performed before reading definition data (step S101 in Fig. 8).

[0120] The user operates the user PC 310. ｎ Alternatively, the data structure conversion method is registered and updated based on the machine learning model and the numerical data described in the recipe via the interface of the AP server 322. The registration and update of the data structure conversion method is performed by the processing device 350. １ From 350 Ｑ The data structure conversion is also registered and updated for the document data stored in the document management system 360 and the machine learning model in the machine learning system 370.

[0121] The AP server 322 registers the target and method of data structure conversion. Steps S301 to S303 are performed by the processing device 350. １ From 350 ＱSteps S301 to S303 are also performed on all parameters acquired from the literature management system 360 and the machine learning system 370.

[0122] The AP server 322 determines whether the data structure conversion method for the data to be converted is unknown (step S301). If the data structure conversion method is unknown (Yes in step S301), the user determines the data to be converted and the data structure conversion method (step S302). For example, the AP server 322 presents the user with candidate data structure conversion methods to select from. Once the data structure conversion method has been determined, the AP server 322 registers the data to be converted and the data structure conversion method (step S303). Even for data for which the data structure conversion method is known in advance (No in step S301), the data structure conversion mode and the method are registered (step S303).

[0123] For example, if the method of data structure conversion is unknown, the AP server 322 may use the AP server interface or the user PC 310 ｎ It is also possible to notify the user that the target data and the data structure conversion method are unknown via the notification. For data for which the data structure conversion method is unknown, the AP server 322 can automatically regard the data structure conversion method for another data equivalent to the relevant data as the data structure conversion method and register it.

[0124] The registered data and methods may be stored in the storage device of the AP server 322 or may be stored in the DB server 323 .

[0125] (Configuration Example of Recipe File in Third Embodiment) FIG. 22 is a diagram showing a configuration example of a recipe file in the third embodiment.

[0126] The recipe information 3630 is data acquired from Lan A, and can be acquired from, for example, DB server 363. Note that the recipe information 3630 can also be acquired from DB server 373 or DB server 333. The recipe information 3630 is a recipe with a process ID of A001. The recipe information 3630 defines an item "Ar," and is set to 100 mL / min in process step 1, 20 mL / min in process step 2, and 10 mL / min in process step 3.

[0127] The recipe information 3400 is data acquired from the LANB, for example, the edge PC 340 ｑ The recipe information 3400 can be acquired from the database server 333. The recipe information 3400 is a recipe with a process ID of B001. The recipe information 3400 defines an item "MFC1 (AR)" and sets the flow rate as 110 mL / min in process step 1, 10 mL / min in process step 2, and 20 mL / min in process step 3.

[0128] Merged data 3220 shows a list of items included in recipe information 3630 and recipe information 3400. Since process ID A001 and process ID B001 both indicate processes related to the supply of Ar gas, the item names are listed as "Ar" and "MFC1 (AR)." The table shows numerical values ​​corresponding to different item names for each process step.

[0129] Data after data structure conversion 3221 shows the result when data structure conversion is performed on merged data 3220. Only "Ar" is entered as the "item name." Although recipe information 3630 and recipe information 3400 have different "item names," they both indicate the supply amount of Ar gas and share the same units. Therefore, data after data structure conversion 3221 has undergone the third type of data structure conversion, and recipe information 3630 and recipe information 3400 share the same "Ar" as the "item name," with the value entered as is.

[0130] (Operations and Effects of the Third Embodiment) Multiple Processing Devices 350 １ From 350 Ｑ The data may include equipment for which no equipment alliance has been established. Furthermore, data specifications may not be standardized in the literature management system 360, or literature values ​​and databases may contain data other than semiconductor manufacturing equipment. It is anticipated that the acquired data may include not only original data such as recipes, but also data with standardized or dimensionless numerical values, data that presents only relationships, data partially provided from a database server via an API, and so on. In such cases, in Example 3, it is possible to independently define a method for data structure conversion of the target data. Because data structure conversion can be independently defined, machine learning models can be efficiently trained.

[0131] (Example 4) Example 4 will be described with reference to Figures 23 and 24. Example 4 assumes a case in which, for example, data is shared between customers who have concluded an alliance contract based on the contents of the contract. Example 4 differs from Example 1 in that a data set for a machine learning model is acquired between different networks. In the following description, components that are the same as or equivalent to those in Examples 1 to 3 described above are denoted by the same reference numerals, and their description will be simplified or omitted.

[0132] (Configuration of Example 4) FIG. 23 is a diagram showing Example 4 of a system to which a recipe search method is applied. The search system 400 includes Lans A to C. Example 4 shows a case in which Lan C includes a machine learning system 420, Lan A includes a manufacturing data system 430, and Lan B includes a manufacturing data system 460. This is equivalent to Example 2 in which a Lan including a manufacturing data system is further added. Note that Example 4 shows a case in which there are two Lans including manufacturing data systems, but the present disclosure can also be used in cases in which there are two or more Lans including manufacturing data systems.

[0133] Machine learning system 420 and user PC 410 in LANC １ From 410 ＮThe machine learning system 120 and the user PC 110 in the first embodiment １ From 110 Ｎ The configuration of the machine learning system 420 (Web server 421, AP server 422, DB server 423) corresponds to the configuration of the machine learning system 120 in the first embodiment (Web server 121, AP server 122, DB server 123).

[0134] In Lan A, Edge PC 440 １ From 440 Ｐ (P is a positive integer) corresponds to the edge PC 140 in the first embodiment. １ From 450 Ｐ Each of these corresponds to the processing device 150 in the first embodiment. In the following description, when referring to an edge PC and a processing device without specifying, the edge PC 440 ｐ (p is a positive integer equal to or less than P) and a processing unit 450 ｐ In addition, in LanB, the edge PC 470 １ From 470 Ｑ (Q is a positive integer) corresponds to the edge PC 140 in the first embodiment. １ From 480 Ｑ Each of these corresponds to the processing device 150 in the first embodiment. In the following description, when referring to an edge PC and a processing device without specifying, the edge PC 470 ｑ (q is a positive integer equal to or less than Q) and a processing unit 480 ｑ is used.

[0135] (Method of the Fourth Embodiment) A search method in the fourth embodiment will be described.

[0136] The actual processing procedure is the same as that of the first embodiment shown in FIG. 8, and therefore will be described with reference to FIG. 8. ｐ and processing unit 480 ｑ From the processing device 450 ｐ and processing unit 480 ｑThe AP server 422 of the machine learning system 420 acquires data describing information required for executing the process in the edge PC 440 (corresponding to step S100 in FIG. 8). The AP server 422 acquires the numerical data described in the recipe and related information (such as the file name of the recipe and information indicating the execution result of the recipe). ｐ and via the manufacturing data system 430 or the edge PC 470 ｑ These data are acquired via the manufacturing data system 460. ｐ When collecting data sent from the database server 433 to the web server 431, the application server 422 acquires the data via the database server 433, the application server 432, and the web server 431. １ From 450 Ｐ The numerical data and related information can be obtained by the processing unit 480. ｑ When collecting data sent from the database server 460 to the web server 461, the application server 422 acquires the data via the database server 463, the application server 462, and the web server 461. １ From 480 Ｑ It is possible to obtain numerical data and related information.

[0137] Next, the AP server 422 reads the definition data (corresponding to step S101 in FIG. 8). The AP server 422 inquires about the target and method of data structure conversion for the acquired data.

[0138] Next, the AP server 422 performs data structure conversion on the acquired data based on the target and method of data structure conversion indicated in the definition data (corresponding to step S102 in FIG. 8). The data structure conversion is performed according to the procedure of the flowchart showing the details of data structure conversion shown in FIG. 9(a).

[0139] Next, the AP server 422 trains the machine learning model using the data whose data structure has been converted (corresponding to step S103 in FIG. 8).

[0140] Next, the user inputs the recipe and target values ​​for the process to be performed and searches for a recipe. The variables (input data) input to the machine learning model are optimized for the machine learning model. The input and output of the machine learning model are also adjusted to satisfy the restriction information classified from the input data (corresponding to step S104 in FIG. 8 ).

[0141] The AP server 422 reversely converts the input variables into variables of the processing device (recipe generation) (corresponding to step S105 in FIG. 8). In other words, the searched recipe is expressed in the form of input variables of the machine learning model. By performing reverse conversion of the data structure conversion on the searched recipe, the processing device 450 ｐ a format or processor 480 for describing the recipe of ｑ A recipe is generated that is converted into a format that describes the recipe.

[0142] The pre-processing procedure in the search method of the fourth embodiment is the same as the pre-processing procedure in the search method of the third embodiment shown in FIG.

[0143] The user operates the user PC 410. ｎ Alternatively, the data structure conversion method is registered and updated based on the machine learning model and the numerical data described in the recipe via the interface of the AP server 422. The registration and update of the data structure conversion method is performed by the processing device 450. １ From 450 Ｐ and processing unit 480 １ From 480 Ｑ This is carried out for all target processing equipment.

[0144] 21, the AP server 422 registers the target and method of data structure conversion. Steps S301 to S303 are performed by the processing device 450. １ From 450 Ｑ and processing unit 480 １ From 480 Ｑ This is performed on all data (features) that can be obtained from the recipes used in the

[0145] The AP server 422 determines whether the data structure conversion method for the data to be converted is unknown (step S301). If the data structure conversion method is unknown (Yes in step S301), the user determines the data to be converted and the data structure conversion method (step S302). For example, the AP server 422 presents the user with candidate data structure conversion methods to select from. Once the data structure conversion method has been determined, the AP server 422 registers the data to be converted and the data structure conversion method (step S303). Even for data for which the data structure conversion method is known in advance (No in step S301), the data structure conversion mode and the method are registered (step S303).

[0146] The registered data and methods may be stored in the storage device of the AP server 422 or may be stored in the DB server 423 .

[0147] (Configuration Example of Recipe File in Fourth Embodiment) FIG. 24 is a diagram showing a configuration example of a recipe file in the fourth embodiment.

[0148] The recipe information 4400 is data acquired from the LAN A, for example, the edge PC 440 ｐ The recipe information 4400 can be acquired from the DB server 433. The recipe information 4400 is a recipe with a process ID of A001. The recipe information 4400 defines the item "Ar" and is standardized, which is a scaling method in which the mean is 0 and the variance is 1, and is set to -1.197 in process step 1, 0.473 in process step 2, and 1.475 in process step 3.

[0149] The recipe information 4630 is data acquired from the LANB, for example, the edge PC 470 ｑThe recipe information 4630 can be acquired from the DB server 463. The recipe information 4630 is a recipe with a process ID of B001. The recipe information 4630 defines the item "Ar" and is standardized, which is a scaling method in which the mean is 0 and the variance is 1, and is set to -0.841 in process step 1, 0.060 in process step 2, and 0.661 in process step 3.

[0150] The merged data 4220 shows a list of items included in the recipe information 4400 and the recipe information 4630. Since both relate to gas supply, the category is listed as "Gas." The item name is "Ar." As shown in the LAN item, a numerical value is shown for each LAN from which the recipe information was acquired.

[0151] Data after data structure conversion 4221 shows the result when data structure conversion is performed on the merged data 4220. Because the mean and standard deviation used in standardization, a scaling method that sets the mean to 0 and the variance to 1, for feature quantities are different, classifications set for each Lan, "Method A" and "Method B," are provided to represent each. Furthermore, new names are set for the feature quantities in this way, and a third type of data structure conversion has been performed.

[0152] (Functions and Effects of Example 4) Example 4 is assumed to be a case in which, for example, data is shared between customers who have concluded an alliance contract based on the contents of the contract. More specifically, for example, a case in which data to which a filter, mask, or anonymization based on the contract has been applied (not original data) is available is assumed. In such a case, Example 4 allows data to be shared between customers based on the alliance contract, making it possible to efficiently train a machine learning model. Note that Example 4 differs from Example 2 in that, while Example 2 is designed with a single customer in mind, Example 4 allows necessary information to be shared between multiple customers who have concluded an alliance based on the terms of each contract.

[0153] (Example 5) Example 5 will be described with reference to FIGS. 25 and 26 . Example 5 is an example utilizing virtual data, i.e., virtual measurement, digital twin, reuse of machine learning models, and physical models. While Examples 1 to 3 use real (or previously existing) data such as equipment specifications and operation records, Example 5 utilizes virtually generated data, i.e., data that does not actually exist (or previously existed). Specific examples of "virtually generated" include simple interpolation using a single value or average value, interpolation using an approximate formula based on other real data, inference results or simulation results using another machine learning model, etc. Furthermore, while Example 4 focuses on who the data provider is, Example 5 focuses on how the provided data is generated. Therefore, Example 4 and this example may be combined (e.g., sharing of trained models or physical models based on an alliance between customers). In the following description, components that are the same or equivalent to those in Examples 1 to 4 described above are designated by the same reference numerals, and their description will be simplified or omitted.

[0154] (Configuration of the Fifth Embodiment) The configuration of the fifth embodiment is a configuration in which virtual data is utilized in the configuration of the fourth embodiment shown in FIG.

[0155] (Method of Example 5) A search method in Example 5 will be described with reference to FIGS. 25 and 26. FIG.

[0156] Fig. 25 is a diagram showing the actual processing procedure of the search method of Example 5. In Fig. 25, steps S100, S101, and S102 to S105 are the same as the steps included in Examples 1, 2, 3, and 4. The difference is that virtual data generation (step S500) is included. As described above, specific examples of virtual data generation include simple interpolation using a single value or average value, interpolation using an approximate formula based on other real data, inference results using another machine learning model, simulation results, etc.

[0157] As an example, a digital twin that realizes a common configuration with the manufacturing data system 430 can be provided in Lan C. Instead of the machine learning system 420 acquiring recipe information in Lan A or Lan B, the digital twin can generate information equivalent to the recipe information. Furthermore, the AP server 422 of the machine learning system 420 can perform interpolation, inference, or simulation on the acquired recipe information to supplement virtual data.

[0158] The generated virtual data is used to train the machine learning model (step S103).

[0159] The details of the pre-processing procedure and data structure conversion are the same as those in the fourth embodiment.

[0160] (Configuration Example of Recipe File in Fifth Embodiment) FIG. 26 is a diagram showing a configuration example of a recipe file in the fifth embodiment.

[0161] The recipe information 4700a and the acquired data 4700b are both data acquired from Lan B. The recipe information 4700a is data acquired from the device a, and the recipe information 4700b is data acquired from the device b. The devices a and b are, for example, the DB server 463, the edge PC 470, １ From 470 Ｑ , processing unit 480 １ ~480 Ｑ The recipe information 4700a is a plurality of recipes including process IDs B001 and B002. The recipes are in the "gas" category and specify the item "Ar." The recipe with process ID B001 is set to 90 mL / min in process step 1 and 30 mL / min in process step 2. The recipe with process ID B002 is set to 100 mL / min in process step 1 and 20 mL / min in process step 2.

[0162] The acquired data 4700b shows estimated data obtained from a semiconductor substrate inspection device or measurement device, which shows the estimated results of processing when a recipe defined by a recipe file is executed using a semiconductor substrate processing device. The acquired data 4700b shows that, for process ID B001, measurement location A is 100 nm and measurement location B is 130 nm. For process ID B002, measurement location A is 120 nm and measurement location B is 110 nm.

[0163] The merged data 5220 shows a list of items included in the recipe information 4700a and the acquired data 4700b. For each process ID, data related to gases in the recipe information 4700a and data related to measurement locations in the acquired data 4700b are listed.

[0164] The machine learning model 520 is identified with an ID of A001. The machine learning model 520 is trained using data obtained from LanA.

[0165] The data after data structure conversion 5221 shows the result when data structure conversion is performed on the merged data 5220. Compared to the merged data 5220, the data after data structure conversion 5221 has a measurement location C (estimated value) added. For example, the machine learning model 520 is trained on the measurement location C measured on Lan A, and estimates the measurement value of the measurement location C for the data indicated in the recipe information 4700a acquired from Lan B. A new feature, the measurement location C (estimated value), has been added, and a first type of data structure conversion has been performed.

[0166] (Functions and Effects of Example 5) In Example 5, it is expected that by using virtual data, virtual measurements can be performed when a physical model is executed in a processing device, the model can be reused (including a machine learning system), and the model can be linked with a physical model. Since the results of both the real environment and the virtual environment can be used for machine learning, it is possible to efficiently train a machine learning model.

[0167] [Second Embodiment] The second embodiment differs from the first embodiment in that a recipe search is performed without performing estimation by machine learning. In the following description, components that are the same as or equivalent to those in the first embodiment described above are denoted by the same reference numerals, and their description will be simplified or omitted.

[0168] In order to obtain information for optimizing a recipe, which is a set of data describing the settings during processing so that the desired processing results are obtained based on data obtained from the processing device, we consider searching for a recipe that satisfies specified search conditions from a set of past recipes.

[0169] First, a group of past recipes is acquired. Possible methods of acquisition include acquiring the recipes from a storage medium connected to the processing device itself, or acquiring the recipes from an external system that stores information received from the processing device. The processing device is composed of components that perform the processing device's functions, and the recipe group holds setting values ​​for operating each component of the processing device. Specifically, examples of setting values ​​include, but are not limited to, gas flow rates and temperatures.

[0170] For a processing device to operate normally, each setting value described in a recipe must be uniquely associated with the target processing device component. In other words, the recipe stores each setting value associated with a name that can uniquely identify each processing device component.

[0171] When searching for a desired recipe from a group of recipes, if such data is to be targeted, the setting items used as search conditions will be items that can uniquely identify each component part of the processing device, but in this case, several issues arise.

[0172] The first challenge is that, because the purpose of the search is to optimize the recipe so that the desired processing results are achieved, it is desirable to specify the setting items to be searched for that are in line with the reaction system occurring within the processing chamber. However, if the data obtained from the processing equipment is used as is, the setting items to be searched for will only be the control items of the processing equipment.

[0173] The second problem is that when the configuration of a processing apparatus is partially changed, it may become impossible to search for the required recipe. For example, to improve the performance of the apparatus, consider replacing a temperature control component that controls the overall temperature with a temperature control component that can precisely control local temperatures. In this case, even though the locations within the processing apparatus that are subject to temperature control remain the same, the setting items have changed before and after the change. As a result, the setting items that can be specified as search targets change before and after the configuration change of the processing apparatus, making it difficult to perform consistent searches before and after the configuration change of the processing apparatus.

[0174] A third problem is the difficulty of performing searches across multiple processing equipment. Different equipment often has different components. For example, even if the settings are for the exact same type of gas, the settings used in the recipe may differ due to factors such as different connected piping. In this case, it is extremely difficult to perform a consistent search across multiple processing equipment by specifying some search criteria. Meanwhile, there is a strong need for optimization that utilizes past processing results between similar processing equipment, and there is a strong desire to create a method that can utilize even a portion of past information.

[0175] As a means for solving the above problem, a search method of a second embodiment is a search method for searching for a desired recipe from multiple recipes defined by a recipe file, and when generating data to be searched, a conversion process is performed to convert some or all of the items in the recipe file into items common to all of the data to be searched. The conversion process includes at least one of adding, deleting, performing arithmetic operations on values, grouping, dividing, and replacing the numeric data. An analysis method of the second embodiment is an analysis method for searching for a desired recipe from multiple recipes defined by a recipe file and analyzing the data by comparing it with processing results linked to the recipe. When generating data to be searched, a conversion process is performed to convert some or all of the items in the recipe file into items common to all of the data to be searched.

[0176] Specifically, rather than treating the data group describing the setting values ​​in the processing device as the data to be searched, a data structure conversion is performed using arithmetic operations, statistical processing, and similar arithmetic calculations to convert the data into variables that are in line with the reaction system occurring within the processing device.

[0177] An explanation will be given based on the above-mentioned example in which multiple gas pipes of the same type are connected. Assume that gas pipes 1, 2, and 3 are connected to a processing apparatus, and gas pipes 1 and 2 contain the same type of gas, while gas pipe 3 contains a different type of gas. In this case, when processing a semiconductor substrate in the processing apparatus, if it is considered impossible to distinguish which gas pipe the gas introduced from gas pipe 1 and gas pipe 2 came from when it reaches the semiconductor substrate, the flow rates of the gas introduced from gas pipe 1 and the gas introduced from gas pipe 2 are summed, and this sum is used as the flow rate of the introduced gas and is regarded as the data to be searched. On the other hand, since there is no other gas pipe connected to gas pipe 3 that introduces the same type of gas as the gas connected to gas pipe 3, the flow rate of gas pipe 3 is used as the flow rate of that gas and is regarded as the data to be searched.

[0178] There are many other examples of such sources, including raw material supply sources, heat sources, electron beam sources, electromagnetic wave sources, light sources, power sources, sound sources, etc. Furthermore, the way in which they are handled does not simply involve treating the total amount, but also includes expressions using the four basic arithmetic operations and the like, such as expressions that represent the difference between the amount actually introduced into the processing chamber, such as the relationship between incident waves and reflected waves from an electromagnetic wave source, expressions that represent the product or integral of the flow rate of the raw material fluid and the processing time, and expressions that represent the quotient or differential, such as the change in heater power for temperature control in the processing chamber.

[0179] In this way, the variables in the processing device are preprocessed to convert the data structure into appropriate variables and their set values, and the converted data is output as data to be searched.

[0180] By using the output search target data, a desired search can be performed to extract the required recipe. For example, a group of recipes can be searched for that differ only in the pressure setting value inside the processing chamber, and the other setting items are set to the same values.

[0181] With these recipes and their corresponding process results, it is possible to analyze the responsiveness of the process results to specific setpoints. For example, in the example above, by creating a graph with the pressure setpoint on the horizontal axis and the process result value on the vertical axis, the responsiveness of the process results to pressure can be visualized. The analysis results obtained in this way can be used as information to optimize the recipe to achieve the desired process results.

[0182] The search method and analysis method of the second embodiment can be implemented, for example, by applying a computer system including a CPU and memory instead of the machine learning system 120 in Example 1 of FIG. 5 . In this case, a program for causing the CPU to execute the search method and analysis method is stored in the memory. The computer system can search for recipes from the manufacturing data system 130. The search method and analysis method of the second embodiment are not limited to being applied to Example 1, and can be applied instead of the machine learning systems in other Examples.

[0183] Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the gist of the present invention.

[0184] The following are some possible aspects of the present invention, although the present invention is not limited to these. (Aspect 1) A search method for estimating, through machine learning, when a recipe defined by a recipe file is executed using a semiconductor substrate processing apparatus, wherein the machine learning uses a machine learning model that inputs numerical data describing the recipe, outputs estimated data indicating the estimated results of the semiconductor substrate processing, and performs learning using training data including the numerical data and data indicating the execution results of the recipe, and when generating the training data, integrates a data set including the numerical data and data indicating the execution results, and further performs a conversion process to convert the structure of the recipe file into a structure compatible with the machine learning model. (Aspect 2) A search method according to Aspect 1, wherein the conversion process includes at least one of adding, deleting, performing arithmetic operations on the numerical data, grouping, dividing, and replacing. (Aspect 3) A searching device that estimates processing when a recipe defined by a recipe file is executed using a semiconductor substrate processing device by machine learning corresponding to a model of processing performed by the processing device, comprising: a computer implemented with a learner that receives as input numerical data describing the recipe and target values ​​for the processing, outputs estimated data indicating estimated results of semiconductor substrate processing by the processing device, and performs learning using learning data including the numerical data and data indicating execution results of executing the recipe, and a display that displays a user interface for setting the target values, wherein when generating the learning data, the computer integrates data sets including the numerical data and data indicating the execution results, and further performs a conversion process that converts the structure of the recipe file into a structure corresponding to the model. (Aspect 4) The searching device according to Aspect 3, wherein the conversion process includes at least one of adding, deleting, performing arithmetic operations on the numerical data, grouping, dividing, and replacing.(Aspect 5) The searching device according to Aspect 3 or 4, wherein the computer combines the recipe files when generating the learning data. (Aspect 6) The searching device according to any one of Aspects 3 to 5, wherein the searching device comprises a learning device to which the target value is input and the numerical data is output. (Aspect 7) A computer capable of communicating with a searching device that estimates processing when a recipe defined by a recipe file is executed using a semiconductor substrate processing device by machine learning corresponding to a model of processing performed by the processing device, wherein the searching device comprises a learning device that inputs numerical data describing the recipe and outputs estimated data indicating an estimated result of semiconductor substrate processing by the processing device, and that performs learning using learning data including the numerical data and data indicating an execution result of executing the recipe, and wherein the computer comprises a processor that integrates a data set including the numerical data and data indicating the execution result, and further performs a conversion process that converts the structure of the recipe file into a structure corresponding to the model. (Aspect 8) The searching device according to any one of Aspects 3 to 6, wherein the numerical data describing the recipe when searching for an optimal recipe is generated so as to satisfy control information including requirements for operating the processing device. (Aspect 9) The searching device according to any one of Aspects 3 to 6 and 8, wherein the recipe file has a tabular data structure including the numerical data and labels attached to the numerical data. (Aspect 10) The searching device according to any one of Aspects 3 to 6 and 8 and 9, wherein the searching device receives a recipe file used for a processing device different from the processing device via an API-linked server, and performs data conversion to merge the recipe file with the recipe file of the processing device.(Aspect 11) A semiconductor device manufacturing system including a platform on which semiconductor manufacturing equipment is connected via a network and on which an application for estimating, by machine learning, processing when a recipe defined by a recipe file is executed using a semiconductor substrate processing equipment is implemented, wherein the machine learning uses a machine learning model in which numerical data describing the recipe is input and estimated data indicating an estimated result of the semiconductor substrate processing is output, and learning is performed using training data including the numerical data and data indicating an execution result of executing the recipe, wherein, when the training data is generated, a data set including the numerical data and data indicating the execution result is integrated, and further, a conversion process is performed to convert the structure of the recipe file into a structure corresponding to the machine learning model. (Aspect 12) A search method for searching for a desired recipe from a plurality of recipes defined by recipe files, wherein, when generating data to be searched, a conversion process is performed to convert some or all of the items in the recipe file into items common to all of the data to be searched. (Aspect 13) The search method according to Aspect 12, wherein the conversion process includes at least one of adding, deleting, performing arithmetic operations on values, grouping, dividing, and replacing the data. (Aspect 14) An analysis method for searching for a desired recipe from among a plurality of recipes defined by a recipe file and analyzing the recipe against processing results linked to the recipe, wherein, when generating data to be searched, a conversion process is performed to convert some or all of the items in the recipe file into items that are common to all of the data to be searched.

[0185] 11: User-side restriction information, 12: Device-side restriction information, 13: Device information, 14: Other information, 16: Constraint conditions, 17: Learning data, 18: Machine learning model, 20: Model, 22: Verification data, 51-59, 60a, 60b, 61a, 61b, 62a, 62b: Data, 100, 200, 300, 400: Search system, 110 １ ~110Ｎ , 210 １ ~210 Ｎ , 310 １ ~310 Ｎ , 410 １ ~410 Ｎ : User PC, 120, 220, 320, 370, 420: Machine learning system, 121, 131, 221, 231, 321, 331, 361, 371, 421, 431, 461: Web server, 122, 132, 222, 232, 322, 332, 362, 372, 422, 432, 462: AP (application) server, 123, 133, 223, 233, 323, 333, 363, 373, 423, 433, 463: DB (database) server, 130, 230, 330, 430, 460: Manufacturing data system, 140, 140a, 240 １ ~240 Ｑ , 340 １ ~340 Ｑ , 440 １ ~440 Ｑ , 470 １ ~470 Ｑ : Edge PC, 150, 150a, 250 １ ~250 Ｑ , 350 １ ~350 Ｑ , 450 １ ~450 Ｐ , 480 １ ~480 Ｑ : Processing device, 160: External recording medium, 1220: CPU, 1221: Memory, 1222: Storage, 1223: Interface, 207: Input / output device, 1224: Data input / output block, 1225: Data structure conversion block, 1226: ML model optimization block, 1227: Recipe generation block, 230, 330, 460: Manufacturing data system, 360: Literature management system, 2230, 2400, 3230, 3400, 3630, 4400, 4630, 4700a: Recipe information 4700b: Acquired data 2220, 3220, 4220, 5220: Merged data 2221, 3221, 4221, 5221: Data after data structure conversion

Claims

1. A search method for estimating, by machine learning, when a recipe defined by a recipe file is to be executed using a semiconductor substrate processing apparatus, the search method using a machine learning model in which numerical data describing the recipe is input, estimated data indicating the estimated results of the semiconductor substrate processing is output, and learning is performed using learning data including the numerical data and data indicating the results of executing the recipe, and when generating the learning data, a data set including the numerical data and the data indicating the results is integrated, and a conversion process is further performed to convert the structure of the recipe file into a structure corresponding to the machine learning model.

2. A searching method according to claim 1, wherein the conversion process includes at least one of the processes of adding, deleting, performing arithmetic operations on the numerical data, grouping, dividing, and replacing.

3. A search device that estimates the result of a recipe defined by a recipe file being executed using a semiconductor substrate processing device through machine learning corresponding to a model of the processing performed by the processing device, comprising: a computer having a learning device implemented therein that receives as input numerical data describing the recipe and a target value for the processing, outputs estimated data indicating the estimated result of the processing of the semiconductor substrate by the processing device, and performs learning using learning data including the numerical data and data indicating the execution result of executing the recipe; and a display that displays a user interface for setting the target value, wherein when generating the learning data, the computer integrates a data set including the numerical data and the data indicating the execution result, and further performs a conversion process to convert the structure of the recipe file into a structure corresponding to the model.

4. A search device according to claim 3, wherein the conversion process includes at least one of the processes of adding, deleting, performing arithmetic operations on the numerical data, grouping, dividing, and replacing.

5. A searching device according to claim 3, wherein said computer combines said recipe files when generating said learning data.

6. A searching device according to claim 3, further comprising a learning device to which the target value is input and to which the numerical data is output.

7. A computer capable of communicating with a search device that estimates the execution of a recipe defined by a recipe file using a semiconductor substrate processing device through machine learning corresponding to a model of processing performed by the processing device, wherein the search device is provided with a learning device that receives numerical data describing the recipe, outputs estimated data indicating the estimated results of semiconductor substrate processing by the processing device, and performs learning using learning data including the numerical data and data indicating the execution results of executing the recipe, and the computer is provided with a processor that integrates a data set including the numerical data and the data indicating the execution results, and further performs a conversion process to convert the structure of the recipe file into a structure corresponding to the model.

8. A search device according to claim 3, characterized in that the numerical data describing the recipe when searching for an optimum recipe is generated so as to satisfy control information including requirements for operating the processing device.

9. A searching device according to claim 3, wherein the recipe file has a tabular data structure including the numerical data and labels attached to the numerical data.

10. A search device according to claim 3, which receives a recipe file used in a processing device different from the processing device via an API-linked server, and executes data conversion to merge the recipe file with the recipe file of the processing device.

11. A semiconductor device manufacturing system having a platform to which semiconductor manufacturing equipment is connected via a network and which is implemented with an application for estimating by machine learning when a recipe defined by a recipe file is executed using a semiconductor substrate processing equipment, wherein the machine learning uses a machine learning model in which numerical data describing the recipe is input and estimated data indicating an estimated result of the processing of the semiconductor substrate is output, and learning is performed using learning data including the numerical data and data indicating the execution result of executing the recipe, wherein when the learning data is generated, a data set including the numerical data and the data indicating the execution result is integrated, and further a conversion process is performed to convert the structure of the recipe file into a structure corresponding to the machine learning model.

12. A search method for searching for a desired recipe from among multiple recipes defined by a recipe file, characterized in that when generating the data to be searched, a conversion process is performed to convert some or all of the items in the recipe file into items that are common to all of the data to be searched.

13. A search method according to claim 12, wherein the conversion process includes at least one of the processes of adding, deleting, performing arithmetic operations on values, grouping, dividing, and replacing the data.

14. An analysis method for searching for a desired recipe from among multiple recipes defined by a recipe file and analyzing the recipe against the processing results linked to the recipe, characterized in that when generating the data to be searched, a conversion process is performed to convert some or all of the items in the recipe file into items that are common to all of the data to be searched.

Images