Semiconductor device manufacturing system, processing result estimation method, search method, and server

Abstract

The present invention aims to provide a technique that enables the estimation of processing results by inputting parameters into a model corresponding to each step, even when the processing conditions differ in the number of steps or step structure from the processing conditions used as training data. [Solution] The semiconductor device manufacturing system of the present invention includes a learning step in which the model parameters of the integrated model are learned using at least one step model, an integrated model which is a model that integrates the step models, and training data; an estimation step in which information on processing results under desired processing conditions is estimated using the learned integrated model; and an output step in which the estimated processing results are output. The integrated model outputs information on processing results using the parameter values ​​of the steps corresponding to the step models as input values, and the previously acquired information on processing results includes information on processing results from two or more steps.

Description

[Technical Field]

[0001] The present invention relates to a semiconductor device manufacturing system, a processing result estimation method, a search method, and a server for use in a semiconductor process. [Background Art]

[0002] In a semiconductor process, desired processing of a semiconductor device is performed by setting appropriate processing conditions obtained through process development in a processing apparatus. Here, the term "processing" refers to processing such as film formation processing, lithography processing, and etching processing performed on a sample. The processing conditions include at least one or more items of control parameters for the processing apparatus.

[0003] In recent years, with the introduction of new materials constituting devices and the expansion of the control range of apparatuses accompanying the complication of device structures, many new items have been added to processing conditions.

[0004] In order to fully bring out the performance of a processing apparatus, optimization of processing conditions is essential. To achieve this, it is important to understand the relationship between processing condition parameters and processing results, but the complication of processing conditions and device structures requires advanced knowledge for this. For this reason, the shortage of experts capable of optimizing processing conditions at process development sites has become a problem.

[0005] As a solution, a method of constructing a machine learning model that estimates processing results from processing conditions is attracting attention. Specifically, this involves creating training data from a set of experimental data where processing conditions are input and processing results are output, and then training a regression model. Patent Document 1 is an example of such an invention. Patent Document 1 aims to "support the search for equipment conditions of a manufacturing apparatus, and in particular, support the construction of a dataset for generating a model used for the search," and discloses the following as a computer system and a method for supporting the search for equipment conditions: "A computer system that supports the search for equipment conditions for controlling a manufacturing apparatus that performs processing accepts input training data consisting of a combination of input parameter values, which are equipment conditions, and output parameter values, which are the results of the processing, analyzes the structure of the processing manufacturing process in the reference training data and the structure of the processing manufacturing process in other training data, performs a classification process to identify training data that can be aggregated into the same dataset as the reference training data, generates a dataset based on the results of the classification process, performs a model training process using the dataset, searches for input parameter values ​​based on the model and target values ​​of the output parameters, and presents an interface to display the search results." [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2023-68859 [Overview of the project] [Problems that the invention aims to solve]

[0007] However, semiconductor processing conditions are typically composed of a series of multiple steps, and in process development, the processing conditions that achieve the target processing are found through trial and error by changing the number and configuration of steps. Therefore, experimental data sets contain a mixture of processing conditions with different numbers and configurations of steps. General regression models such as linear regression and kernel regression cannot be used for training unless the data size of the input processing conditions matches. When using these regression models, only data with the same number and configuration of steps are extracted from the experimental data set and used for training. If it is not possible to extract multiple data sets with the same data size from the experimental data set, there may be insufficient training data that can be used for training, resulting in insufficient prediction accuracy (estimation accuracy). Furthermore, if the input-output relationship learned from experimental data is represented by a single regression model, it becomes difficult to extract only the learning results for some steps from the regression model, making it difficult to ensure sufficient reusability of the model.

[0008] Patent Document 1 states that "a classification process is performed to identify training data that can be aggregated into the same dataset as the reference training data, a dataset is generated by aggregating the training data based on the results of the classification process, and a model learning process is performed using the dataset to predict the value of the output parameter from the value of the input parameter." Therefore, the aforementioned problem may arise with Patent Document 1, but this is not sufficiently discussed in Patent Document 1. Therefore, the present invention aims to provide a technique that enables the estimation of processing results by inputting parameters into a model corresponding to each step, even when the processing conditions differ in the number of steps or step structure from the processing conditions used as training data. [Means for solving the problem]

[0009] To solve the above problems, one representative semiconductor device manufacturing system of the present invention is a semiconductor device manufacturing system comprising a platform on which an application is implemented that takes processing conditions of a semiconductor manufacturing apparatus consisting of at least one step as input and outputs information indicating an estimated value of the processing result according to the processing conditions, wherein the application executes a learning step in which the model parameters of the integrated model are learned using at least one step model which is a model which takes the values ​​of the parameters defined in the step as input values, an integrated model which is a model which integrates the step models, and training data, an estimation step in which information on the processing result according to the desired processing conditions is estimated using the learned integrated model, and an output step in which the estimated processing result is output, wherein the integrated model outputs information on the processing result using the values ​​of the parameters of the step corresponding to the step model as input values, and the previously acquired information on the processing result includes information on the processing result from two or more of the steps. [Effects of the Invention]

[0010] According to the present invention, even if the processing conditions differ in the number of steps or step structure from the processing conditions used as training data, it becomes possible to estimate the processing result by inputting parameters into the model corresponding to each step. Other issues, configurations, and effects not mentioned above will be clarified by the description of the embodiments for carrying out the invention below. [Brief explanation of the drawing]

[0011] [Figure 1] Figure 1 is a block diagram showing an example of the configuration of a semiconductor device manufacturing system in Example 1. [Figure 2] Figure 2 shows an example of the processing conditions in Example 1. [Figure 3] Figure 3 shows another example of the processing conditions in Example 1. [Figure 4] Figure 4 shows an example of the integrated model in Example 1. [Figure 5] FIG. 5 is a diagram showing an example of an integration method in a case where the number of steps of processing conditions differs from the number of steps of an integrated model in Example 1. [Figure 6] FIG. 6 is a diagram showing still another example of the integrated model in Example 1. [Figure 7] FIG. 7 is a diagram showing a flowchart of processing performed in the semiconductor device manufacturing system of Example 1. [Figure 8] FIG. 8 is a block diagram showing an example of a configuration of a semiconductor device manufacturing system in Example 2. [Figure 9] FIG. 9 is a diagram showing an example of processing conditions in Example 2. [Figure 10] FIG. 10 is a diagram showing another example of processing conditions in Example 2. [Figure 11] FIG. 11 is a diagram showing an example of preprocessing of processing conditions in Example 2. [Figure 12] FIG. 12 is a diagram showing another example of preprocessing of processing conditions in Example 2. [Figure 13] FIG. 13 is a diagram showing an example of an integrated model in Example 2. [Figure 14] FIG. 14 is a diagram showing another example of an integrated model in Example 2. [Figure 15] FIG. 15 is a diagram showing a flowchart of processing performed in the semiconductor device manufacturing system of Example 2. [Figure 16] FIG. 16 is a block diagram showing an example of a configuration of a semiconductor device manufacturing system in Example 3. [Figure 17] FIG. 17 is a diagram showing an example of a target process in Example 3. [Figure 18] FIG. 18 is a diagram showing an example of a reference process in Example 3. [Figure 19] FIG. 19 is a diagram showing an example of processing conditions for a target process in Example 3. [Figure 20] FIG. 20 is a diagram showing an example of processing conditions for a reference process in Example 3. [Figure 21] Figure 21 is a flowchart showing the process performed by the semiconductor device manufacturing system of Example 3. [Figure 22] Figure 22 shows an example of a GUI that presents the estimation results to the user in the estimation result output unit. [Figure 23] Figure 23 shows an example of a GUI for users to assign labels in the labeling section. [Figure 24] Figure 24 shows another example of a GUI for the user to assign labels in the labeling unit 1102. [Figure 25] Figure 25 shows an example of a GUI for searching and selecting a reference step model to reference from the reference model database. [Figure 26] Figure 26 shows another example of a GUI for searching and selecting a reference step model to reference from a reference model database. [Figure 27] Figure 27 shows an example of the hardware configuration of the semiconductor device manufacturing system of Example 3. [Modes for carrying out the invention]

[0012] Embodiments of the present invention will be described below with reference to the drawings. However, the present invention is not limited to the embodiments described below. It will be readily apparent to those skilled in the art that the specific configuration can be modified without departing from the spirit or intent of the present invention. The positions, sizes, and shapes of the components shown in the drawings and other illustrations in this specification may not represent their actual positions, sizes, and shapes in order to facilitate understanding of the invention. Therefore, the present invention is not limited to the positions, sizes, and shapes disclosed in the drawings and other illustrations.

[0013] In this disclosure, a processing condition step means a processing condition associated with each divided processing step when a series of processing steps based on processing conditions are performed, by dividing the series of processing steps chronologically. For example, a processing condition step refers to a setting condition of a processing device that is switched chronologically when a series of processing steps based on processing conditions are performed by the processing device. [Examples]

[0014] As Example 1, an example of a semiconductor device manufacturing system that estimates processing results such as critical dimension (CD) and etching depth in a plasma etching process will be described. The semiconductor device manufacturing system is capable of performing a processing result estimation method to estimate the processing results. However, this disclosure is not limited to the type and content of the processing. This disclosure can be applied to semiconductor device manufacturing equipment, but semiconductor device manufacturing equipment includes not only plasma etching equipment but also, for example, lithography equipment, film deposition equipment, pattern processing equipment, ion implantation equipment, heating equipment, and cleaning equipment. Lithography equipment includes exposure equipment, electron beam lithography equipment, and X-ray lithography equipment. Film deposition equipment includes CVD (Chemical Vapor Deposition), PVD (Physical Vapor Deposition), evaporation equipment, sputtering equipment, and thermal oxidation equipment. Pattern processing equipment includes wet processing equipment, dry processing equipment, electron beam processing equipment, and laser processing equipment. Ion implantation equipment includes plasma doping equipment and ion beam doping equipment. Heating equipment includes resistance heating equipment, lamp heating equipment, and laser heating equipment. Cleaning equipment includes liquid cleaning equipment and ultrasonic cleaning equipment.

[0015] (System configuration and processing) Figure 1 is a block diagram showing an example of the configuration of a semiconductor device manufacturing system 1 in Embodiment 1. The semiconductor device manufacturing system 1 includes a platform on which an application is implemented that takes the processing conditions of a semiconductor manufacturing apparatus, which is configured by at least one step, as input and outputs information showing an estimated value of the processing result based on the processing conditions. Specifically, the semiconductor device manufacturing system 1 includes an experimental database 101, a step model 102, an integrated method setting unit 103, an integrated model 104, a preprocessing unit 105, a learning unit 106, an estimation unit 107, and an estimation result output unit 108. Each component included in the semiconductor device manufacturing system 1 is connected to each other directly or via a network. Each component may be implemented by hardware or as software (an application). When each component is implemented as an application, the application may be implemented on a platform that integrates software and hardware, and the semiconductor device manufacturing system 1 may be configured to include this platform. The functionality of each component is realized by the execution of the application on the platform. Alternatively, the semiconductor device manufacturing system 1 may be a server on which the application is implemented and executed. The semiconductor device manufacturing system 1 may include a semiconductor device manufacturing apparatus, and information used in the manufacturing apparatus may be stored in the experimental database 101. The semiconductor device manufacturing system 1 estimates processing results derived from processing conditions specified by the user, for example. The processing conditions specified by the user may be referred to as the first processing conditions, and the estimated processing results as the first processing results. Furthermore, the processing conditions and processing results stored in the experimental database 101, which will be described later, may be referred to as the second processing conditions and the second processing results, and explained separately.

[0016] The experimental database 101 stores data including information on a second processing condition, which is a processing condition acquired in advance, and information on the second processing result, which is processed based on the second processing condition. In addition, the information on the processing result acquired in advance in the experimental database 101 includes information on the processing result from two or more steps. The experimental database 101 also includes information on the processing result obtained by measurement after two or more steps have been processed consecutively. Specifically, the experimental database 101 is a database that stores data indicating the processing conditions and the processing results obtained by performing evaluations such as measurements on the processed object (semiconductor wafer in this disclosure) processed based on those processing conditions. Generally, processing conditions consist of several steps, but the processing conditions included in the experimental database 101 are not limited to having a common number of steps and step configuration, and may be processing conditions with different numbers of steps and step configurations. Furthermore, each step of the processing conditions may include information on the material of the object to be processed.

[0017] Figure 2 shows an example of the processing conditions in Example 1. Processing conditions 201 consist of three steps, from Step 1 to Step S3. Processing conditions 201 include the following items: "Material to be processed", "CF4 [ml / min]", "Cl2 [ml / min]", "Ar [ml / min]", and "Processing time [s]". The item "Material to be processed" is information about the material to be processed, indicating the material of the semiconductor wafer to be subjected to plasma etching, and shows Si, which is the material of the silicon substrate, and SiO2, which is the material of the film formed on the silicon substrate. The items "CF4 [ml / min]", "Cl2 [ml / min]", and "Ar [ml / min]" indicate the flow rate of the gas introduced into the plasma processing apparatus, respectively. The item "Processing time [s]" indicates the time from when the gas supply is started until it is stopped. As will be described later, the values ​​of these items are treated as parameters when pre-processing is performed. Note that for simplicity, the unit of the item "CF4 [ml / min]" is omitted and shown as "CF4". The same applies to the other items.

[0018] Figure 3 shows another example of the processing conditions in Example 1. Processing conditions 202 consist of four steps, from step 1 to step 4. Although processing conditions 201 in Figure 2 and processing conditions 202 in Figure 3 have different numbers of steps, both are stored in the experimental database 101.

[0019] For objects subjected to plasma etching based on the processing conditions, measurements such as CD and Etch depth are performed. The processing results, which are data showing the measured results, are linked to processing conditions 201 and 202 respectively and stored in the experimental database 101. Processing results may be set for each processing condition, or for each step of the processing conditions. Furthermore, it is not limited to cases where processing results are set for all steps included in the processing conditions; there may be steps for which processing results are not set. In addition, processing results may include multiple items, such as CD and Etch depth.

[0020] The following explanation describes a case where the processing conditions include five items, but this disclosure is not limited to this case. Other items may be included in the processing conditions, and different items may be included in processing other than plasma etching. Furthermore, the processing results can also be set according to the type of processing performed, similar to the processing conditions. In addition, the data stored in the experimental database 101 may be experimental data obtained through experiments, or data obtained through plasma etching.

[0021] Refer to Figure 1 for explanation. The preprocessing unit 105 creates training data by performing preprocessing on the data stored in the experimental database 101. Training data is generated by performing preprocessing on data that includes information on processing conditions and information on processing results that have been learned in advance (training data generation step). The preprocessing methods include selecting parameters included in the processing conditions and scaling the values ​​of parameters included in the processing conditions. The former includes processes such as removing all parameters from a set of parameters that have a perfect correlation across all processing conditions, or removing parameters whose values ​​remain constant across all processing conditions. The latter includes processes such as scaling each parameter so that the mean is 0 and the variance is 1 across all processing conditions, or scaling so that the Max is 1 and the Min is 0. The preprocessing unit 105 performs preprocessing on multiple processing conditions with different step numbers and step structures to create training data.

[0022] The step model 102 is a model (hereinafter referred to as the "step model") that takes the values ​​of parameters defined for each step as input values. The step model 102 may be a single step model or multiple step models (one or more step models). The step model 102 is learned by the learning unit 106, as described later. The step model 102 includes the values ​​of the processing condition items for each step as input and the processing result items as output. As shown in Example 2, the step model may also include information about the material to be processed at each step as input values.

[0023] The integration method setting unit 103 sets or changes the integration method in the integrated model 104 described later. The integration method setting unit 103 sets the integration method when integrating the step models 102 to generate a single integrated model 104. The integration method setting unit 103 sets or changes the integration method of the step models (integration method setting change step).

[0024] The integrated model 104 is a model that integrates one or more step models and is created according to the integration method set in the integration method setting unit 103. The integrated model 104 takes the parameter values ​​of the steps corresponding to the step models as input values ​​and outputs processing result information. For example, the integrated model 104 takes the parameter values ​​of a series of steps corresponding to each of one or more step models (step model 102) as input values ​​and outputs processing result information when processing a semiconductor wafer based on the series of steps. The integration method will be described later.

[0025] The learning unit 106 learns the model parameters of the integrated model using at least one step model, an integrated model which is a model that combines the step models, and training data (learning step). For example, the learning unit 106 learns the model parameters of the integrated model using training data so that the output of the integrated model 104 matches the processing result. For example, if each of the step models 102 included in the integrated model 104 is a neural network, the learning unit 106 learns the model parameters of the integrated model 104 by backpropagation. Even if the model is not a neural network model, if it is a parametric model, learning can be performed by numerically calculating the model parameters that minimize the generalization error estimated using training data, for example.

[0026] The estimation unit 107 uses the trained integrated model 104 to estimate information about the processing result under the desired processing conditions (first processing conditions) (estimation step). Specifically, the estimation unit 107 can, for example, accept the user's desired processing conditions, and by inputting the user's desired processing conditions into the trained integrated model 104, it estimates the processing result that would be achieved if the plasma etching process were performed according to the desired processing conditions.

[0027] The estimation result output unit 108 outputs the processing result (first processing result) estimated by the estimation unit 107 (output step) and presents it to the user of the semiconductor device manufacturing system 1.

[0028] (Examples of integration methods) The integration method in Example 1 will be explained with reference to Figures 4 to 6. The integration method setting change step generates an additional model which is a different model from the step model. The integration method is one in which the output of the step model is used as the input to the additional model, and the model parameters of the integration model are learned so that the output of the additional model becomes information of the processing result. Figure 4 is a diagram showing an example of the integration model 104 in Example 1. In the integration method shown in Figure 4, the integration method setting unit 103 generates an additional model 310 which is a different model from one or more step models, and sets an integration method in which the output of the step model is used as the input to the additional model 310. The integration model 104 includes step models 303, 305, 307, and 309 connected in parallel, as well as the additional model 310. Step models 303, 305, 307, and 309 correspond to step model 102 in Figure 1.

[0029] In the integrated model 104, the step model 303 is input with the parameters 302 of step 1 of the processing condition 301 (for example, the values ​​for the items "CF4", "CL2", "Ar", and "processing time"; the same applies to the parameters of the other steps). Similarly, the step models 305, 307, and 309 are input with the parameters 304 of step 2, 306 of step 3, and 308 of step 4, respectively. The additional model 310 is input with the output of each step model.

[0030] The learning unit 106 learns the model parameters of step models 303, 305, 307, and 309, as well as the additional model 310, so that the output of the additional model 310 matches the processing result associated with the processing condition 301. Model parameters include, for example, fitting parameters such as weights for the input data and biases for adjusting the predicted values ​​in the case of a linear regression model, and parameters specific to the nonlinear function such as coefficients, cardinal and exponential for each term of the nonlinear function in the case of a nonlinear regression model. In the case of a neural network model, these parameters include the learning rate, number of epochs, and number of units in the hidden layer.

[0031] (Other examples of integration methods) Furthermore, Figure 5 shows the input-to-output process when a processing condition with a different number of steps than the processing condition 301 is input to the integrated model 311. Figure 5 is a diagram showing an example of the integration method when the number of steps in the processing condition and the number of steps in the integrated model differ in Example 1. The integrated model 104 includes step models 403, 405, and 407, as well as an additional model 409. Also, step model 102 includes step models 403, 405, 407, and 408. In addition, steps 1 to 3 of the processing condition 301 in Figure 5 have the same values ​​(parameters) as steps 1 to 3 of the processing condition 301 in Figure 4.

[0032] Processing condition 401 consists of three steps. In this case, step model 403 is input with parameter 402 from step 1 of processing condition 401. Similarly, step models 405 and 407 are input with parameter 404 from step 2 and parameter 406 from step 3, respectively. On the other hand, since processing condition 401 does not have a fourth step, nothing is input to step model 408. Additional model 409 is input with the output of each step model. If the input size of additional model 409 is not variable, a dummy output, such as 0, is output from step model 408 and included in the input of additional model 409.

[0033] (Another example of an integration method) Figure 6 shows yet another example of the integration method in Example 1. The integration method combines a step model corresponding to one step with another step model by using the output of the step model corresponding to one step as the input of the step model corresponding to the other step, where one step is the step preceding the other step. Figure 6 shows yet another example of the integration model in Example 1. The step model integration method shown in Figure 6 combines step models by using the output of a step model associated with a certain step as the input of a step model associated with the next step from among one or more step models. Integration model 104 includes step models 503, 505, and 507 connected in series. Step model 102 also includes step models 503, 505, and 507.

[0034] In the integrated model 104, the parameter 502 of step 1 of processing condition 501 is input to step model 503. Next, the output of step model 503 and the parameter 504 of step 2 are input to step model 505. Similarly, the output of step model 505 and the parameter 506 of step 3 are input to step model 507. The learning unit 106 learns the model parameters of step models 503, 505, and 507 so that the output of step model 507 matches the processing result associated with processing condition 501.

[0035] In Example 1, the integration method setting unit 103 is not a mandatory component. For example, if the integration method is predetermined, or if an integration model generated outside the semiconductor device manufacturing system 1 already exists, the integration method setting unit 103 does not need to be included in the semiconductor device manufacturing system 1.

[0036] (Flowchart of processes in a semiconductor device manufacturing system) Figure 7 is a flowchart showing the processing performed in the semiconductor device manufacturing system of Example 1. In step S1, the semiconductor device manufacturing system 1 stores the processing results and processing conditions. For example, processing conditions obtained through experiments and processing results, which are the results of measuring the processed object processed based on the processing conditions, are stored in the experimental database 101.

[0037] Next, in step S2, the semiconductor device manufacturing system 1 performs preprocessing. The preprocessing unit 105 generates training data by performing preprocessing on the stored data.

[0038] Next, in step S3, the semiconductor device manufacturing system 1 learns the model parameters of the integrated model. The integrated model 104 is a model that integrates one or more step models included in step model 102, which are linked to each step of the processing conditions stored in step S1. The learning unit 106 learns the model parameters of the integrated model using the training data generated in step S2.

[0039] Next, in step S4, the semiconductor device manufacturing system 1 estimates the processing result. The estimation unit 107 uses the trained integrated model 104 learned in step S3 to estimate the processing result when the target is processed based on the user's desired processing conditions.

[0040] Next, in step S5, the semiconductor device manufacturing system 1 outputs the estimation result. The estimation result output unit 108 outputs the estimation result from step S4 as a processing result linked to the desired processing conditions.

[0041] (Effects / Actions) In this disclosure, multiple processing conditions with different numbers of steps and step structures are used as training data to train the model. Furthermore, the integrated model used to estimate the processing results can be configured to match the input processing conditions. Therefore, even if the user's desired processing conditions differ in number of steps and step structure from the processing conditions used as training data, parameters can be input to the integrated model corresponding to each step included in the user's desired processing conditions, and the processing results can be estimated. [Examples]

[0042] (System configuration and processing) A second embodiment of a semiconductor device manufacturing system for estimating processing results in plasma etching is described. A labeling step is further performed by the application in which labels are assigned to the steps, and each step model corresponding to a step with the same label is the same step model. The semiconductor device manufacturing system of Embodiment 2 includes a labeling unit 602 that assigns labels to each step of the second processing conditions (labeling step), and differs from the semiconductor device manufacturing system of Embodiment 1 in that steps with the same label among the steps of the second processing conditions are associated with the same step model. Embodiment 2 describes a semiconductor device manufacturing system in which labels are assigned to the steps of the processing conditions according to information about the material to be processed, and steps with the same label share the same step model, thereby reducing the number of step models and facilitating learning. In the following description, the same or equivalent components as in Embodiment 1 above are denoted by the same reference numerals, and their descriptions may be simplified or omitted.

[0043] Figure 8 is a block diagram showing an example of the configuration of the semiconductor device manufacturing system 10 in Example 2. The semiconductor device manufacturing system 10 in Example 2 includes an experimental database 601, a labeling unit 602, a step model 603, an integrated method setting unit 604, an integrated model 605, a preprocessing unit 606, a learning unit 607, an estimation unit 608, and an estimation result output unit 609. Here, the experimental database 601, step model 603, integrated method setting unit 604, integrated model 605, preprocessing unit 606, learning unit 607, estimation unit 608, and estimation result output unit 609 correspond to the experimental database 101, step model 102, integrated method setting unit 103, integrated model 104, preprocessing unit 105, learning unit 106, estimation unit 107, and estimation result output unit 108 in the semiconductor device manufacturing system 1 of Figure 1, respectively.

[0044] The labeling unit 602 assigns a label to each step of the processing conditions stored in the experimental database 601. The preprocessing unit 606 performs preprocessing for each step of the processing conditions, and the processing content of the preprocessing is defined for each label assigned to the step.

[0045] (label) Labels are assigned to steps based on information about the material being processed in each step. Figure 9 shows an example of processing conditions in Example 2. Processing conditions 701 are processing conditions stored in the experimental database 601. In Example 2, unlike Example 1, the labeling unit 602 assigns two types of labels, A or B, depending on the material being processed. In processing conditions 701, steps 1 and 4 both use SiO2 as the material being processed and are assigned label A. Steps 2 and 3 both use Si as the material being processed and are assigned label B. As will be described later, steps assigned the same label share the same step model (they are linked to the same step model).

[0046] Figure 10 shows another example of the processing conditions in Example 2. In processing conditions 702, steps 1, 3, and 5 all use SiO2 and are labeled A. Steps 2 and 4 both use Si as the material to be processed and are labeled B.

[0047] Note that the method of assigning labels does not necessarily have to correspond to the material being processed. For example, three types of labels may be assigned depending on whether the step is an etching step, a deposition step, or another step. In this case, it is expected that the accuracy of the model estimation will improve by including information on the material being processed as described in processing conditions 701 and 702 as input to the step model. This is because by including information on the material being processed as input to the step model, the step model can reflect the fact that even for steps with the same label, the response of the processing result will differ if the material being processed is different. One way to include information on the material being processed as input to the step model is to replace the name of the material being processed with a numerical sequence of 0s or 1s using the One-hot encoding method and include it in the input parameters. Alternatively, the same step model may be shared for all steps, and the labels may be converted to numerical data using the One-hot encoding method and used as input to switch the step model based on the label to which the step corresponds.

[0048] (Label-based preprocessing) An example of pre-processing in the pre-processing unit 606 will be explained with reference to Figure 11. Figure 11 is a diagram showing an example of pre-processing conditions in Example 2. Table 801 summarizes the steps labeled B and A in processing conditions 701 in Figure 9 and processing conditions 702 in Figure 10. Specifically, Table 801 includes steps 2 and 3 of processing condition 701 and steps 2 and 4 of processing condition 702.

[0049] In the preprocessing in Example 2, parameter selection and scaling processes are performed on the processing conditions. In the parameter selection process, parameters that are not necessary for training the regression model are removed. Specifically, for CF4, the value is 0 at each step of processing condition 701 and also at each step of processing condition 702. As a result, parameters whose values ​​do not change, such as CF4 in Table 801, do not contribute to training the regression model and are therefore removed from the processing condition data. Also, if there is a perfect correlation between multiple parameters, one parameter is kept and the other parameters are removed. For example, in Table 801, there is a perfect correlation of 1:2 between Cl2 and Ar, so only the parameter for Cl2 is kept and the parameter for Ar is removed.

[0050] Next, parameter scaling is performed. In the scaling process, the values ​​of each parameter are standardized so that the mean is 0 and the variance is 1.

[0051] The results of applying the two pretreatments described above to Table 801 are shown in Table 803.

[0052] Figure 12 shows another example of the pretreatment conditions for the processing conditions in Example 2. Table 802 includes steps 1 and 4 of processing condition 701 and steps 1, 3 and 5 of processing condition 702.

[0053] In the parameter selection process for processing conditions, the value of Cl2 is 0 in both steps of processing condition 701 and processing condition 702, so the Cl2 parameter is removed. Next, the parameter scaling process is performed. The results of applying the parameter selection process and the parameter scaling process to Table 802 are shown in Table 804.

[0054] In the learning unit 607, the data that has undergone these processes is used as training data. Although the preprocessing described here applies to processing conditions 801 and 802, in reality, the same processing is performed on all data stored in the experimental database 101.

[0055] (Examples of integration methods) The integration method setting unit 604 sets the integration method when integrating the step models 603 to generate a single integrated model. An example of the integration method in Embodiment 2 is shown in Figure 13. Figure 13 is a diagram showing an example of the integrated model 605 in Embodiment 2. The integrated model 605 includes two step models 903, two step models 905, and an additional model 908, and the outputs of the two step models 903 and two step models 905, which are arranged in parallel, are input to the additional model 908. Also, step models 903 and 905 correspond to step model 603 in Figure 8.

[0056] Specifically, step model 903 is the model associated with label A, and step model 905 is the model associated with label B. The model parameters of the two step models 903 are common. Similarly, the model parameters of the two step models 905 are common. Since the labels for steps 1 and 4 of processing condition 901 are A, their parameters 902 and 907 are input to step model 903. Also, since the labels for steps 2 and 3 of processing condition 901 are B, their parameters 904 and 906 are input to step model 905. The outputs of step models 903 and 905 are input to additional model 908. The learning unit 607 learns the model parameters of step models 903, 905 and additional model 908 so that the output of additional model 908 matches the processing result associated with processing condition 901. Note that step model 903, which corresponds to two steps, will continue to be a model using common model parameters even after learning. The same applies to step model 905.

[0057] (Other examples of integration methods) Another example of the integration method in Example 2 is shown in Figure 14. Figure 14 is a diagram showing another example of the integration model in Example 2. Integration model 605 includes step model 1006 and step model 1007. Step model 603 also includes step models 1006 and 1007.

[0058] Step model 1006 is a model associated with label A, and step model 1007 is a model associated with label B. Processing condition 1001 consists of four steps. First, since the label for step 1 of processing condition 1001 is A, its parameter 1002 is input to step model 1006, and step model 1006 outputs output 1008.

[0059] Next, since the label for step 2 of processing condition 1001 is B, its parameter 1003 is input to step model 1007. At the same time, the output 1008 of step model 1006 is also input to step model 1007. As a result, step model 1007 outputs output 1009.

[0060] Next, since the label for step 3 of processing condition 1001 is also B, its parameter 1004 is input to step model 1007. At the same time, the output 1009 of step model 1007 is also input to step model 1007. As a result, step model 1007 outputs output 1010.

[0061] Finally, since the label for step 4 of processing condition 1001 is A, its parameter 1005 is input to step model 1006. Simultaneously, the output 1010 of step model 1007 is also input to step model 1006. As a result, step model 1006 outputs output 1011. The learning unit 607 learns the model parameters of step models 1006 and 1007 so that output 1011 matches the processing result associated with processing condition 1001. One example of a method that uses the output of one model as input to another model is a recurrent neural network.

[0062] (Flowchart of processes in a semiconductor device manufacturing system) Figure 15 is a flowchart showing the processes performed in the semiconductor device manufacturing system of Example 2. Step S11 in the flowchart of Figure 15 is a process newly added in Example 2. Steps S10 and S12 through S15 correspond to steps S1 through S5 in the flowchart of Example 1 shown in Figure 7, respectively.

[0063] First, as step S10, the semiconductor device manufacturing system 10 stores the processing conditions and processing results. The processing conditions and processing results are stored in the experimental database 601.

[0064] Next, in step S11, the semiconductor device manufacturing system 10 assigns labels to the processing conditions. The labeling unit 602 assigns a label to each step of the processing conditions. The labeled processing conditions are stored in the experimental database 601. It is also possible for the user to set labels via the labeling unit 602.

[0065] Next, in step S12, the semiconductor device manufacturing system 10 performs preprocessing. The preprocessing unit 105 generates training data by performing preprocessing on the stored data. Preprocessing includes, for example, parameter acquisition and selection processing and parameter scaling processing.

[0066] Next, in step S13, the semiconductor device manufacturing system 10 learns the model parameters of the integrated model. In the integrated model 605, the step parameters that have been assigned a common label are input into the common step model.

[0067] Next, in step S14, the semiconductor device manufacturing system 10 estimates the processing result. Then, in step S15, the semiconductor device manufacturing system 10 outputs the estimation result.

[0068] (Effects / Actions) In Example 2, each step of the processing conditions is labeled, and a step model is created according to the label. Compared to Example 1, where a step model is created for each step, steps with a common label share a single step model, thus reducing the number of models included in the integrated model. Therefore, it becomes easier to train the model parameters of the integrated model, or in other words, the model parameters of the step models included in the integrated model. [Examples]

[0069] (System configuration and processing) A third embodiment of a semiconductor device manufacturing system for estimating processing results in plasma etching is described. The model parameters of a step model in the target processing step of a step are learned by referring to the model parameters of a reference step model, and the reference step model is a step model that has been learned in a processing step different from the target processing step of a step. The model parameters of a step model are learned by referring to the model parameters of a reference step model that has been assigned the same label as the label of a step, and the reference step model is a step model that has been learned in a processing step different from the target processing step. The semiconductor device manufacturing system of Embodiment 3 further has a reference step model database that stores a reference step model, which is a model that has been learned in a reference process that is a different process from the target process corresponding to the second processing condition, and differs from Embodiments 1 and 2 in that the model parameters of the step model in the target process are determined by referring to the model parameters of the reference step model. Embodiment 3 describes a semiconductor device manufacturing system that efficiently learns the step model of a target process by referring to the model parameters of a step model learned in a process similar to the target process. In the following description, the same or equivalent components as those in Embodiments 1 and 2 above are denoted by the same reference numerals, and their descriptions may be simplified or omitted.

[0070] Figure 16 is a block diagram showing an example of the configuration of the semiconductor device manufacturing system 20 in Example 3. The semiconductor device manufacturing system 20 in Example 3 includes an experimental database 1101, a labeling unit 1102, a step model 1103, an integrated method setting unit 1104, an integrated model 1105, a preprocessing unit 1106, a reference model database 1107, a learning unit 1108, an estimation unit 1109, and an estimation result output unit 1110. Here, the experimental database 1101, labeling unit 1102, step model 1103, integration method setting unit 1104, integration model 1105, preprocessing unit 1106, learning unit 1108, estimation unit 1109, and estimation result output unit 1110 correspond, respectively, to the experimental database 601, labeling unit 602, step model 603, integration method setting unit 604, integration model 605, preprocessing unit 606, learning unit 607, estimation unit 608, and estimation result output unit 609 in the semiconductor device manufacturing system 10 shown in Figure 8.

[0071] The reference model database 1107 stores step models and additional models that have been trained using training data from other processing steps that differ from the processing conditions stored in the experimental database 1101. The step models and additional models stored in the reference model database 1107 are referred to as trained reference step models and trained reference additional models, respectively, to distinguish them from the step models included in step model 1103 and the additional models included in integrated model 1105. Furthermore, as will be described later, trained reference step models and trained reference additional models are linked to labels assigned to the training data used for training.

[0072] (Target process and reference process) The target process and the reference process will be explained with reference to Figures 17 and 18. Figure 17 is a diagram showing an example of the target process in Example 3. In Example 3, the target process 1201 in Figure 17 is trench etching of a structure in which SiO2 is stacked on top of Si. Etching is performed according to the shape of the mask placed on the SiO2, removing SiO2 and Si, and forming a trench.

[0073] On the other hand, Figure 18 shows an example of the reference step in Example 3. Reference step 1202 in Figure 18 is trench etching of a structure in which SiN is stacked on top of Si. Etching is performed according to the shape of the mask placed on the SiN, removing SiN and Si and forming a trench.

[0074] The reference model database 1107 stores the step model and additional models learned in reference process 1202 shown in Figure 18. Although there is a difference between the target process 1201 and the reference process 1202 in whether the material on top of the Si is SiO2 or SiN, the etching process of the Si underneath is similar.

[0075] The processing conditions for target process 1201 and reference process 1202 will be explained with reference to Figures 19 and 20. Figure 19 is a diagram showing an example of the processing conditions for target process 1201 in Example 3. Processing condition 1301 in Figure 19 is the processing condition corresponding to target process 1201 in Figure 17, and is a processing condition stored in the experimental database 1101. In processing condition 1301, step 1 is assigned label A, and steps 2 and 3 are assigned label B. Of the etching processes, the process of removing SiO2 on Si corresponds to step 1, and the process of removing Si after removing SiO2 corresponds to steps 2 and 3.

[0076] On the other hand, Figure 20 shows an example of the processing conditions for reference step 1202 in Example 3. Processing condition 1302 in Figure 20 is an example of processing conditions used to train the reference step model stored in the reference model database 1107. In processing condition 1302, step 1 is assigned reference label C, and steps 2 and 3 are assigned reference label B. Among the etching processes, step 1 corresponds to the process of removing SiN on Si, and steps 2 and 3 correspond to the process of removing Si after removing SiN. Note that, for convenience, the labels in the reference process are referred to as reference labels to distinguish them from the labels of the target process. In both the target process and the reference process, the labels are assigned according to the material being processed.

[0077] Steps 2 and 3, to which label B is assigned under processing condition 1301, and steps 2 and 3, to which label B is assigned under processing condition 1302, both correspond to trench etching of Si, and it is thought that etching proceeds by a similar mechanism in plasma etching processes based on these steps. Therefore, in Example 3, a trained reference step model linked to label B, stored in the reference model database 1107, is utilized. Specifically, the learning unit 1108 refers to the model parameters of the trained reference step model linked to label B and determines the model parameters of the step model 1103 linked to label B in the target process 1201. As a result, if the reference step model has been trained with a large amount of training data, even if the amount of training data in the target process 1201 is small and it is difficult to train the step model 1103, it is expected that the step model 1103 can be trained with high accuracy by referring to the model parameters of the trained reference step model.

[0078] (Learning model parameters using a reference step model) As an example of how to utilize a reference step model, when the model parameters of the step model are learned in the learning unit 1108, it is conceivable to restrict changes from the model parameters of the reference step model. A penalty function is defined for the integrated model that includes a term indicating the degree of deviation between the model parameters of the step model and the model parameters of the reference step model, and the model parameters of the integrated model that minimize the value of the penalty function are searched for.

[0079] Specifically, as a penalty function, a loss function is defined such that, for example, a penalty is imposed the further the model parameters deviate from the values ​​of the reference step model. The learning unit 1108 searches for the model parameters of the step model 1103 for the target process that minimize the loss function. Here, the form of the loss function can be expressed, for example, as the sum of a term representing the difference between the output of the integrated model 1105 and the processing results stored in the experimental database 1101, and a penalty term p that indicates the penalty mentioned above. This makes it possible to limit the deviation from the model parameters of the reference step model.

[0080] Equation (1) shows an example of a loss function. In equation (1), the term that shows the output of the integrated model 1105 is f(x i |w)(x i ) represents the parameters of the i-th processing condition (where i is a positive integer), and w represents the model parameters. Also, the term showing the processing results stored in the experimental database 1101 is y i (y i This shows the processing result associated with the i-th processing condition. The penalty term is p(w). Note that in equation (1), the first term is shown as the sum of squared errors.

[0081]

number

[0082] Another example of how to utilize a reference step model is transfer learning. Here, assuming that both the step model of the target process and the step model of the reference process are neural network models, we will explain how to utilize a reference step model using transfer learning. The step model is a neural network, and the model parameters of the integrated model are learned by backpropagation. When the step model references the model parameters of the reference step model, the values ​​of the model parameters of the reference step model are set as the initial values ​​of the model parameters of the step model. This is set so that the learning rate value by backpropagation in the step model or the frequency of updating the model parameters in the step model is lower compared to when the step model does not reference the model parameters of the reference step model. Specifically, in transfer learning, the learning unit 1108 sets the model parameters of the reference step model as the initial values ​​of the model parameters of the step model 1103 of the target process, and then learns the model parameters of the step model 1103 by backpropagation using the training data of the target process. Step models trained through transfer learning from a reference step model can limit their deviation from the reference step model's model parameters by setting a lower learning rate or a lower frequency of model parameter updates during backpropagation compared to other step models.

[0083] Furthermore, as in Example 3, if labels are assigned according to the material being processed, a reference step model assigned the same label as the step model of the target process can be referenced. For example, the learning unit 1108 can determine the model parameters of the step model in the target process by referencing the model parameters of a reference step model (hereinafter also referred to as "reference model parameters") that are associated with the same label as the label assigned to the step included in the processing conditions corresponding to the target process. By doing so, the accuracy of model parameter search is expected to be further improved. This is because, even in different processing processes, if the material being processed is the same, the processing mechanism will be similar, and therefore the model parameters of the step model describing the processing will also tend to be similar.

[0084] While the case of learning the model parameters of a step model in an integrated model has been described, this disclosure is not limited to this case. If the integrated model includes a step model and an additional model, it is also possible to refer to the model parameters of a reference additional model when learning the model parameters of the additional model.

[0085] (Flowchart of processes in a semiconductor device manufacturing system) Figure 21 is a flowchart showing the processes performed by the semiconductor device manufacturing system of Example 3. Compared with the flowchart of Example 2 shown in Figure 15, step S23 is a process newly added in Example 3. Steps S20 to S22 and steps S24 to S26 correspond to steps S10 to S15 in the flowchart of Example 2 shown in Figure 15, respectively.

[0086] First, in step S20, the semiconductor device manufacturing system 20 stores the processing conditions and processing results. The processing conditions and processing results are stored in the experimental database 1101.

[0087] Next, in step S21, the semiconductor device manufacturing system 20 assigns labels to the processing conditions. The labeling unit 1102 assigns a label to each step of the processing conditions. The labeled processing conditions are stored in the experimental database 1101.

[0088] Next, in step S22, the semiconductor device manufacturing system 20 performs preprocessing. The preprocessing unit 1106 generates training data by performing preprocessing on the stored data. Preprocessing includes, for example, parameter acquisition and selection processing and parameter scaling processing.

[0089] Next, in step 23, the semiconductor device manufacturing system 20 refers to a reference step model. The learning unit 1108 refers to the reference model database and searches for learned reference step models and learned reference add-on models when learning the model parameters of the integrated model. The learning unit 1108 determines, for example, whether the steps included in the target process and the steps included in the reference process are similar, and selects a reference step model or reference add-on model that has been learned using a reference process similar to the target process. The determination of whether the reference process is similar to the target process can be made, for example, based on the labels assigned to the steps. Alternatively, the determination of whether the reference process is similar to the target process may be based on the material to be processed and the parameter values ​​defined in the step. It is also possible for the user to select a reference model via the learning unit 1108.

[0090] Next, in step S24, the semiconductor device manufacturing system 20 learns the model parameters of the integrated model. Subsequently, in step S25, the semiconductor device manufacturing system 20 estimates the processing results, and in step S26, outputs the estimated results.

[0091] (GUI) Referring to Figures 22 to 27, an example of the GUI (Graphical User Interface) of the semiconductor device manufacturing system 30 in Embodiment 3 will be described. Figure 22 is a diagram showing an example of a GUI that presents the estimation results to the user in the estimation result output unit 1110. The estimation result output screen 1401 shows a processing condition input field 1402 in which the desired processing conditions can be entered, and an estimation result display field 1403 that displays the estimated processing results for the entered processing conditions. The processing condition input field 1402 has a button 1404 for deleting a row and a button 1405 for adding a row.

[0092] The user operates the processing condition input field 1402 to create the desired processing conditions. The user inputs the material to be processed, gas flow rate, processing conditions, and label for each step, creating processing conditions that include multiple steps. The estimated result display field 1403 includes the items "Top CD," "Middle CD," "Bottom CD," and "Etch Depth." Other items may also be included in the estimated results.

[0093] Figure 23 shows an example of a GUI for a user to assign labels in the labeling unit 1102. The batch label setting screen 1501 shows a case where labels are set all at once for each material to be processed. The labeling unit 1102 assigns labels to each step of the processing conditions stored in the experimental database 1101 according to the labels set on the batch label setting screen 1501.

[0094] Figure 24 shows another example of a GUI for the user to assign labels in the label assignment unit 1102. The individual label setting screen 1502 allows the user to edit labels individually for each step of the processing conditions. The individual label setting screen 1502 displays a list 1503 of processing conditions stored in the experimental database 1101 and has a scroll bar 1504. The user can assign labels individually for each step by editing the label column in the list 1503 of processing conditions.

[0095] Figure 25 shows an example of a GUI for searching and selecting a reference step model from the reference model database 1107. On the step model selection screen 1601, the user selects the step model of the target processing step that they want to learn using the reference step model. The selected step model is highlighted, for example, with a thick line. In the example shown in Figure 25, the step model with label A is selected.

[0096] Figure 26 also shows another example of a GUI for searching and selecting reference step models to be referenced from the reference model database 1107. The reference step model selection screen 1602 has a reference model search field 1603 and a search results display field 1604, and allows the user to select the reference step models that the step model selected in the step model selection screen 1601 in Figure 25 should reference. In the reference model search field 1603, the user enters a search expression related to various attributes of the reference step model, such as the material to be processed or the label. The search results display field 1604 displays a list of reference step models that match the search expression entered in the reference model search field 1603. The search results display field 1604 has radio buttons 1605 for the user to select the reference step models that they want to use for learning the step model. In the example shown in Figure 26, the SiO2 etching model of reference model ID 3 in Figure 26 is used in learning the step model of label A in Figure 25.

[0097] (Effects / Actions) Even when users have insufficient experimental data, resulting in limited training data and difficulty in ensuring the accuracy of model parameter training, it is possible to improve estimation accuracy by utilizing pre-trained reference models (step models and additional models) for similar processes. Furthermore, since a step model linked to a step of a certain processing condition can be referenced and used to estimate the processing results of another target process, the reusability of the model is improved.

[0098] (Search system) The semiconductor device manufacturing system of this disclosure can also be applied to a search system. The search system is a system for searching for setting values ​​to be set on a semiconductor device manufacturing apparatus in order to process a target object into a predetermined processing shape. For example, if the semiconductor device manufacturing apparatus is a plasma etching apparatus, the search system can search for an etching recipe for the plasma etching apparatus so as to etch the semiconductor wafer, which is the object to be processed, into a desired shape.

[0099] (Example hardware configuration) The semiconductor device manufacturing system of this disclosure can also be implemented by hardware. Figure 27 shows an example of the hardware configuration of the semiconductor device manufacturing system 20 of Embodiment 3. The semiconductor device manufacturing system 20 includes a storage device 2000, a communication device 2100, an input / output device 2200, a processor 2300, and a memory 2400.

[0100] The storage device 2000 and memory 2400 have a set of software programs and / or instructions executed by the processor 2300. The processor 2300 executes various software programs and / or instruction sets stored in the storage device 2000 and memory 2400 to perform various functions for the semiconductor device manufacturing system 20 and process data. The storage device 2000 and memory 2400 include non-volatile storage devices such as random access memory (RAM), magnetic disk storage devices, flash memory devices, or other non-volatile solid-state memory devices. While it is shown that there is one storage device 2000, processor 2300, and memory 2400 each, there may be two or more. Also, the storage device 2000 and memory 2400 may be separate devices or a single device. Memory 2400 stores, for example, a part of the operating system (OS).

[0101] The communication device 2100 is an interface for the semiconductor device manufacturing system 20 to communicate information with external devices. External devices include semiconductor device manufacturing equipment, measuring equipment for the semiconductor wafer to be processed, and a database that stores information on processing conditions and processing results.

[0102] The input / output device 2200 is used by the user to input and output information to and from the semiconductor device manufacturing system 20. The input / output device 2200 can be, for example, an output device such as a display, or an input device such as a keyboard or mouse. Alternatively, it may be a device that has both input and output functions, such as a touch panel.

[0103] The memory device 2000 stores software programs and / or instruction sets for executing the functions of the semiconductor device manufacturing system 20, and data used to execute the functions. Experimental database 2001 and reference model database 2007 correspond to experimental database 1101 and reference model database 1107 in Figure 16, respectively.

[0104] When the labeling unit 2002 is executed by the processor 2300, the processor 2300 functions as the labeling unit 1102 in Figure 16. Similarly, the step model 2003, the integrated method setting unit 2004, the integrated model 2005, the preprocessing unit 2006, the learning unit 2008, the estimation unit 2009, and the estimation result output unit 2010 are programs and / or instruction sets corresponding to the step model 1103, the integrated method setting unit 1104, the integrated model 1105, the preprocessing unit 1106, the learning unit 1108, the estimation unit 1109, and the estimation result output unit 1110 in Figure 16, respectively.

[0105] Although the hardware configuration of the semiconductor device manufacturing system 20 in Example 3 has been described, a similar hardware configuration can also be applied to the semiconductor device manufacturing systems of Example 1 and Example 2.

[0106] Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above, and various modifications are possible without departing from the spirit of the present invention.

[0107] The following describes, but is not limited to, embodiments that may constitute the present invention. (Aspect 1) A semiconductor device manufacturing system comprising a platform on which an application is implemented that takes processing conditions for a semiconductor manufacturing apparatus consisting of at least one step as input and outputs information indicating an estimated value of the processing result based on said processing conditions, A learning step in which the model parameters of the integrated model are learned using training data, comprising: at least one step model which is a model that takes the values ​​of the parameters defined in the above step as input values; an integrated model which is a model that combines the step models; and training data. An estimation step in which information about the processing result under the desired processing conditions is estimated using the trained integrated model, The application executes an output step in which the estimated processing result is output. The integrated model outputs information of the processing result using the parameter values ​​of the step corresponding to the step model as input values. A semiconductor device manufacturing system characterized in that the information of processing results obtained in advance includes information of processing results from two or more of the aforementioned steps. (Aspect 2) A semiconductor device manufacturing system according to Embodiment 1, The step model is a semiconductor device manufacturing system characterized in that it includes information about the material to be processed in the step as an input value. (Aspect 3) A semiconductor device manufacturing system according to Embodiment 1 or Embodiment 2, The application further executes a training data generation step in which training data is generated by performing preprocessing on data including the information on the processing conditions and the information on the processing results that has been acquired in advance. A semiconductor device manufacturing system characterized in that the preprocessing is a process of selecting or rejecting the parameters or a process of scaling the values ​​of the parameters. (Aspect 4) A semiconductor device manufacturing system according to any one of embodiments 1 to 3, The aforementioned step model is a neural network, A semiconductor device manufacturing system characterized in that the model parameters are learned by backpropagation. (Appendix 5) A semiconductor device manufacturing system according to any one of embodiments 1 to 4, A semiconductor device manufacturing system characterized in that an integration method setting change step is further performed by the application, in which an integration method of the step model is set or the integration method of the step model is changed. (Aspect 6) A semiconductor device manufacturing system according to any one of embodiments 1 to 5, The aforementioned integration method is a method of combining the step model corresponding to one step and the step model corresponding to the other step by using the output of the step model corresponding to one step as the input of the step model corresponding to the other step. A semiconductor device manufacturing system characterized in that one of the aforementioned steps is a step preceding the other step. (Aspect 7) A semiconductor device manufacturing system according to any one of embodiments 1 to 6, The aforementioned integration method setting change step generates an additional model which is a different model from the aforementioned step model. The aforementioned integration method is one in which the output of the step model is used as the input of the additional model. A semiconductor device manufacturing system characterized in that the model parameters of the integrated model are learned such that the output of the additional model becomes the information of the processing result. (Pattern 8) A semiconductor device manufacturing system according to any one of embodiments 1 to 7, The application further performs a labeling step in which a label is assigned to the above step. A semiconductor device manufacturing system characterized in that each of the step models corresponding to the step to which the same label is assigned is the same step model. (Aspect 9) A semiconductor device manufacturing system according to any one of embodiments 1 to 8, The aforementioned preprocessing is performed in each of the steps, A semiconductor device manufacturing system characterized in that the processing content of the pre-processing is defined for each label. (Aspect 10) A semiconductor device manufacturing system according to any one of embodiments 1 to 9, A semiconductor device manufacturing system characterized in that the label is assigned to the step based on information about the material to be processed in the step. (Aspect 11) A semiconductor device manufacturing system according to any one of embodiments 1 to 10, By referring to the model parameters of the reference step model, the model parameters of the step model in the target processing step of the step are learned. A semiconductor device manufacturing system characterized in that the reference step model is a step model that has been learned in a processing step different from the target processing step of the step. (Aspect 12) A semiconductor device manufacturing system according to any one of embodiments 1 to 11, A semiconductor device manufacturing system characterized in that, when the model parameters of the step model are learned, changes from the model parameters of the reference step model are restricted. (Aspect 13) A semiconductor device manufacturing system according to any one of embodiments 1 to 12, Model parameters of the integrated model that result in a smaller value for the penalty function defined for the integrated model are searched for. A semiconductor device manufacturing system characterized in that the penalty function includes a term indicating the degree of deviation between the model parameters of the step model and the model parameters of the reference step model. (Aspect 14) A semiconductor device manufacturing system according to any one of embodiments 1 to 13, The aforementioned step model is a neural network, The model parameters of the integrated model are learned using the backpropagation method. When the step model references the model parameters of the reference step model, the values ​​of the model parameters of the reference step model are set as the initial values ​​of the model parameters of the step model. A semiconductor device manufacturing system characterized in that the step model is configured such that the learning rate value by backpropagation in the step model or the frequency of updating the model parameters in the step model is lower compared to the case where the step model does not refer to the model parameters of the reference step model. (Aspect 15) A semiconductor device manufacturing system according to any one of embodiments 1 to 14, The model parameters of the step model are learned by referring to the model parameters of a reference step model that are assigned the same label as the label of the step. A semiconductor device manufacturing system characterized in that the reference step model is a step model that has been learned in a processing step different from the target processing step. (Aspect 16) A method for estimating processing results, which takes processing conditions for a semiconductor manufacturing apparatus consisting of at least one step as input and outputs information indicating an estimated value of the processing result based on said processing conditions, A learning step that learns the model parameters of the integrated model using training data, comprising: at least one step model which is a model that takes the values ​​of the parameters defined in the step as input values; an integrated model which is a model that integrates the step models; and training data. An estimation step of estimating information about the processing result under the desired processing conditions using the trained integrated model, The system includes an output step that outputs the estimated processing result, The integrated model outputs information of the processing result using the parameter values ​​of the step corresponding to the step model as input values. A method for estimating processing results, characterized in that the information of processing results obtained in advance includes information of processing results from two or more of the steps described above. (Aspect 17) A search method for searching for processing conditions for a desired semiconductor manufacturing apparatus, comprising at least one step, A learning step that learns the model parameters of the integrated model using training data, comprising: at least one step model which is a model that takes the values ​​of the parameters defined in the step as input values; an integrated model which is a model that integrates the step models; and training data. An estimation step of estimating information about the processing result under the desired processing conditions using the trained integrated model, The system includes an output step that outputs the processing conditions such that the estimated processing result becomes the desired processing result, The integrated model outputs information of the processing result using the parameter values ​​of the step corresponding to the step model as input values. A search method characterized in that the information of the processing results obtained in advance includes information of the processing results from two or more of the steps. (Aspect 18) A server having an application implemented that takes processing conditions for a semiconductor manufacturing apparatus consisting of at least one step as input and outputs information indicating an estimated value of the processing result based on said processing conditions, A learning step in which the model parameters of the integrated model are learned using training data, comprising: at least one step model which is a model that takes the values ​​of the parameters defined in the above step as input values; an integrated model which is a model that combines the step models; and training data. An estimation step in which information about the processing result under the desired processing conditions is estimated using the trained integrated model, The application executes an output step in which the estimated processing result is output. The integrated model outputs information of the processing result using the parameter values ​​of the step corresponding to the step model as input values. A server characterized in that the information of processing results obtained in advance includes information of processing results from two or more of the aforementioned steps. [Explanation of symbols]

[0108] 101 Experimental Database 102 Step Model 103 Integrated System Setting Unit 104 Integrated Model 105 Pre-processing section 106 Learning Department 107 Estimation part 108 Estimated Result Output Unit 201 Processing conditions 202 Processing conditions 301 Processing conditions 302 Processing conditions for Step 1 303 Step 1 Step Model 304 Processing conditions for Step 2 305 Step 2 Step Model 306 Processing conditions for Step 3 307 Step 3 Step Model 308 Processing conditions for Step 4 309 Step 4 Step Model 401 Processing conditions 402 Processing conditions for Step 1 403 Step 1 Step Model 404 Processing conditions for Step 2 405 Step 2 Step Model 406 Processing conditions for Step 3 407 Step 3 Step Model 408 Step 4 Step Model 409 Additional Models 501 Processing conditions 502 Processing conditions for Step 1 503 Step 1 Step Model 504 Processing conditions for Step 2 505 Step 2 Step Model 506 Processing conditions for Step 3 507 Step 3 Step Model 601 Experimental Database 602 Labeling Unit 603 Step Model 604 Integrated System Setting Unit 605 Integrated Model 606 Pre-processing section 607 Learning Department 608 Estimation section 609 Estimated Result Output Unit 701 Processing Condition 1 702 Processing Condition 2 801 Processing conditions for Label B 802 Processing conditions for Label A 803 Pre-processing and post-processing conditions for Label B 804 Pre-processing and post-processing conditions for Label A 901 Processing conditions 902 Processing conditions for Step 1 903 Label A Step Model 904 Processing conditions for Step 2 905 Label B Step Model 906 Processing conditions for Step 3 907 Processing conditions for Step 4 908 Additional Models 1001 Processing conditions 1002 Processing conditions for Step 1 1003 Processing conditions for Step 2 1004 Processing conditions for Step 3 1005 Processing conditions for Step 4 1006 Label A step model 1007 Label B step model 1008 Output of Step 1 1009 Output of Step 2 1010 Output of Step 3 1011 Output of Step 4 1101 Experimental Database 1102 Labeling Unit 1103 Step Model 1104 Integrated System Setting Unit 1105 Integrated Model 1106 Pre-processing section 1107 Reference Model Database 1108 Learning Department 1109 Estimation Department 1110 Estimated Result Output Unit 1201 Target Process 1202 Reference process 1301 Target process processing conditions 1302 Reference process conditions 1401 Estimated Result Output Screen 1402 Processing Condition Input Field 1403 Estimation result display field 1404 Delete row button 1405 Add row button 1501 Batch Label Setting Screen 1502 Individual Label Settings Screen 1503 Processing Condition List 1504 Scroll bar 1601 Step Model Selection Screen 1602 Reference Step Model Selection Screen 1603 Reference Model Search Field 1604 Search Results Display Area 1605 Radio buttons 2000 storage device 2001 Experimental Database 2002 Labeling Section 2003 Step Model 2004 Integrated System Setting Section 2005 Integrated Model 2006 Preprocessing 2007 Reference Model Database 2008 Learning Department 2009 Estimation Department 2010 Estimated Result Output Unit 2100 Communication equipment 2200 Input / Output Device 2300 processor 2400 memory

Claims

1. A semiconductor device manufacturing system comprising a platform on which an application is implemented that takes processing conditions for a semiconductor manufacturing apparatus consisting of at least one step as input and outputs information indicating an estimated value of the processing result based on said processing conditions, A learning step in which the model parameters of the integrated model are learned using training data, comprising: at least one step model which is a model that takes the values ​​of the parameters defined in the step as input values; an integrated model which is a model that combines the step models; and training data. An estimation step in which information about the processing result under the desired processing conditions is estimated using the trained integrated model, The application executes an output step in which the estimated processing result is output. The integrated model outputs information of the processing result using the parameter values ​​of the step corresponding to the step model as input values. A semiconductor device manufacturing system characterized in that the information of processing results obtained in advance includes information of processing results from two or more of the steps.

2. A semiconductor device manufacturing system according to claim 1, The step model is a semiconductor device manufacturing system characterized in that it includes information about the material to be processed in the step as an input value.

3. A semiconductor device manufacturing system according to claim 1, The application further executes a training data generation step in which training data is generated by performing preprocessing on data including the information on the processing conditions and the information on the processing results that has been acquired in advance. A semiconductor device manufacturing system characterized in that the preprocessing is a process of selecting or rejecting the parameters or a process of scaling the values ​​of the parameters.

4. A semiconductor device manufacturing system according to claim 1, The aforementioned step model is a neural network, A semiconductor device manufacturing system characterized in that the model parameters are learned by backpropagation.

5. A semiconductor device manufacturing system according to claim 1, A semiconductor device manufacturing system characterized in that an integration method setting change step is further performed by the application, in which an integration method of the step model is set or the integration method of the step model is changed.

6. A semiconductor device manufacturing system according to claim 5, The aforementioned integration method is a method of combining the step model corresponding to one step and the step model corresponding to the other step by using the output of the step model corresponding to one step as the input of the step model corresponding to the other step. A semiconductor device manufacturing system characterized in that one of the aforementioned steps is a step preceding the other step.

7. A semiconductor device manufacturing system according to claim 5, The aforementioned integration method setting change step generates an additional model which is a different model from the aforementioned step model. The aforementioned integration method is one in which the output of the step model is used as the input of the additional model. A semiconductor device manufacturing system characterized in that the model parameters of the integrated model are learned such that the output of the additional model becomes the information of the processing result.

8. A semiconductor device manufacturing system according to claim 1, The application further performs a labeling step in which a label is assigned to the above step. A semiconductor device manufacturing system characterized in that each of the step models corresponding to the step to which the same label is assigned is the same step model.

9. A semiconductor device manufacturing system according to claim 8, The aforementioned preprocessing is performed in each of the steps, A semiconductor device manufacturing system characterized in that the processing content of the pre-processing is defined for each label.

10. A semiconductor device manufacturing system according to claim 8, A semiconductor device manufacturing system characterized in that the label is assigned to the step based on information about the material to be processed in the step.

11. A semiconductor device manufacturing system according to claim 1, By referring to the model parameters of the reference step model, the model parameters of the step model in the target processing step of the step are learned. A semiconductor device manufacturing system characterized in that the reference step model is a step model that has been learned in a processing step different from the target processing step of the step.

12. A semiconductor device manufacturing system according to claim 11, A semiconductor device manufacturing system characterized in that, when the model parameters of the step model are learned, changes from the model parameters of the reference step model are restricted.

13. A semiconductor device manufacturing system according to claim 12, Model parameters of the integrated model are searched for that result in a smaller value for the penalty function defined for the integrated model. A semiconductor device manufacturing system characterized in that the penalty function includes a term indicating the degree of deviation between the model parameters of the step model and the model parameters of the reference step model.

14. A semiconductor device manufacturing system according to claim 12, The aforementioned step model is a neural network, The model parameters of the integrated model are learned using the backpropagation method. A semiconductor device manufacturing system characterized in that, when the step model references the model parameters of the reference step model, the values ​​of the model parameters of the reference step model are set as the initial values ​​of the model parameters of the step model, and the learning rate value by backpropagation in the step model or the frequency of updating the model parameters in the step model is set to be lower compared to when the step model does not reference the model parameters of the reference step model.

15. A semiconductor device manufacturing system according to claim 11, The model parameters of the step model are learned by referring to the model parameters of a reference step model that are assigned the same label as the label of the step. A semiconductor device manufacturing system characterized in that the reference step model is a step model that has been learned in a processing step different from the target processing step.

16. A method for estimating processing results, which takes processing conditions for a semiconductor manufacturing apparatus consisting of at least one step as input and outputs information indicating an estimated value of the processing result based on said processing conditions, A learning step that learns the model parameters of the integrated model using training data, comprising: at least one step model which is a model that takes the values ​​of the parameters defined in the step as input values; an integrated model which is a model that combines the step models; and training data. An estimation step of estimating information about the processing result under the desired processing conditions using the trained integrated model, The system includes an output step that outputs the estimated processing result, The integrated model outputs information of the processing result using the parameter values ​​of the step corresponding to the step model as input values. A method for estimating processing results, characterized in that the information of processing results obtained in advance includes information of processing results from two or more of the steps.

17. A search method for searching for processing conditions for a desired semiconductor manufacturing apparatus, comprising at least one step, A learning step that learns the model parameters of the integrated model using training data, comprising: at least one step model which is a model that takes the values ​​of the parameters defined in the step as input values; an integrated model which is a model that combines the step models; and training data. An estimation step of estimating information about the processing result under the desired processing conditions using the trained integrated model, The system includes an output step that outputs the processing conditions such that the estimated processing result becomes the desired processing result, The integrated model outputs information of the processing result using the parameter values ​​of the step corresponding to the step model as input values. A search method characterized in that the information of the processing results obtained in advance includes information of the processing results from two or more of the steps.

18. A server having an application implemented that takes processing conditions for a semiconductor manufacturing apparatus consisting of at least one step as input and outputs information indicating an estimated value of the processing result based on said processing conditions, A learning step in which the model parameters of the integrated model are learned using training data, comprising: at least one step model which is a model that takes the values ​​of the parameters defined in the step as input values; an integrated model which is a model that combines the step models; and training data. An estimation step in which information about the processing result under the desired processing conditions is estimated using the trained integrated model, The application executes an output step in which the estimated processing result is output. The integrated model outputs information of the processing result using the parameter values ​​of the step corresponding to the step model as input values. A server characterized in that the information of processing results obtained in advance includes information of processing results from two or more of the aforementioned steps.

Images