Analysis device, analysis method, and analysis program

Abstract

To provide a highly convenient analysis device, an analysis method, and an analysis program.SOLUTION: An analysis device includes a processor. The processor calculates a first contribution degree, which is a contribution degree of a first explanatory variable for a first set, on the basis of a teacher data set used for generating a predictor. The predictor is configured to receive a plurality of explanatory variables and output a predicted value corresponding to the plurality of explanatory variables. The first set is a set of values of the plurality of explanatory variables. The first explanatory variable is one of the plurality of explanatory variables. The processor calculates, on the basis of the teacher data set, a second contribution degree, which is the contribution degree of the first explanatory variable for a second set that is a set of values of the plurality of explanatory variables and is different from the first set. The processor acquires a third contribution degree by calculating a difference between the first contribution degree and the second contribution degree, and outputs the third contribution degree.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present embodiment relates to an analysis device, an analysis method, and an analysis program. [Background technology]

[0002] In recent years, machine learning has spread to various fields and is being applied to predicting complex phenomena, optimizing conditions, etc. From a practical perspective, there are increasing cases where explanations and interpretations of the predicted values ​​obtained by machine learning models are required.

[0003] SHAP (SHapley Additive exPlanations) is one of the "eXplainable AI (XAI)" machine learning technologies. With SHAP, it is possible to obtain a SHAP value, which is numerical information that indicates the contribution of each explanatory variable to the predicted value. The contribution is also called importance or contribution rate. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7153142 Summary of the Invention [Problem to be solved by the invention]

[0005] An object of one embodiment is to provide a highly convenient analysis device, analysis method, and analysis program. [Means for solving the problem]

[0006] According to one embodiment, the analysis device includes a processor. The processor calculates a first contribution, which is the contribution of a first explanatory variable, for a first set based on a teacher dataset used to generate the predictor. The predictor is configured to receive multiple explanatory variables as input and output predicted values ​​corresponding to the multiple explanatory variables. The first set is a set of values ​​for the multiple explanatory variables. The first explanatory variable is one of the multiple explanatory variables. The processor calculates a second contribution, which is the contribution of the first explanatory variable, for a second set, which is a set of values ​​for the multiple explanatory variables different from the first set, based on the teacher dataset. The processor obtains a third contribution by calculating the difference between the first contribution and the second contribution, and outputs the third contribution. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 10 is a schematic diagram for explaining an algorithm for calculating a contribution degree for each individual condition according to the embodiment. [Figure 2] FIG. 1 is a schematic diagram illustrating an example of a hardware configuration of an analysis apparatus according to an embodiment. [Figure 3] 10 is a flowchart showing an example of an operation of the analysis device according to the embodiment. [Figure 4] 4 is a diagram showing two examples of pairs of process conditions and predicted values ​​obtained by the processing of S102 to S105 shown in FIG. 3. FIG. [Figure 5] FIG. 10 is a diagram for explaining a reference value determined by a training data set according to an embodiment. [Figure 6] FIG. 10 is a diagram showing an example of a method for displaying SHAP values ​​for individual conditions according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] Machine learning can also be applied to the development of processes for manufacturing semiconductor devices (hereinafter referred to as process development). For example, if a machine learning model is generated that is configured to input process conditions as explanatory variables and output process results as predicted values, it becomes possible to optimize the process conditions using the machine learning model.

[0009] The process conditions are conditions for manufacturing a semiconductor device and are variable conditions that can be set by a developer. The process conditions may include multiple conditions. Each of the multiple conditions included in the process conditions is referred to as an individual condition. When the process conditions are input to a machine learning model, each individual condition is an explanatory variable that constitutes a group of explanatory variables.

[0010] During process development, especially in its early stages, there is a need to understand the effect of improvements from the current POR (Process of Record) conditions. For example, when a predicted value changes due to a change in the process conditions input to a predictor from the POR conditions to other conditions, developers may want to identify the degree to which each individual condition contributed to the change in the predicted value, in other words, its contribution to the change in the predicted value.

[0011] When SHAP is applied, it is possible to obtain a SHAP value for each individual condition as the contribution of each individual condition to the predicted value. However, the SHAP value is calculated based on a reference value determined by the training data set used to generate the machine learning model. In other words, the SHAP value represents the contribution of each individual condition to the change in the predicted value from the reference value. Therefore, it is difficult to interpret the SHAP value itself as the contribution of each individual condition to the change in the predicted value when the process conditions are changed from the specified conditions.

[0012] SHAP is an application of the algorithm for calculating the Shapley value from cooperative game theory to machine learning. The Shapley value has properties such as total group rationality and additivity. Total rationality is the property that the sum of each player's Shapley value is equal to the reward obtained by all players' alliance. Additivity is the property that the sum of each player's Shapley value calculated for each of two rewards is equal to the sum of each player's Shapley value calculated for the sum of the two rewards.

[0013] When players are interpreted as explanatory variables and rewards as predicted values, the SHAP value obtained by SHAP has the above-mentioned properties of global rationality and additivity, similar to the Shapley value. The analysis device 1 according to the embodiment is configured to be able to calculate the contribution of each individual condition to a change in the predicted value when the process conditions are changed, according to an algorithm that utilizes the fact that the SHAP value has global rationality and additivity.

[0014] FIG. 1 is a schematic diagram for explaining an algorithm for calculating the contribution degree for each individual condition according to the embodiment.

[0015] Two sets of values ​​of process conditions can be input to analysis device 1. If a developer wants to identify the contribution of each individual condition to a change in a predicted value when the process conditions are changed from POR conditions to candidate conditions, the developer can input POR conditions as one of the two sets and the candidate conditions as the other of the two sets to analysis device 1.

[0016] One of the two sets of process conditions input to the analysis device 1 is denoted as set Xa, and the other is denoted as set Xb. Set Xa is, for example, POR conditions, and set Xb is, for example, candidate conditions. Note that the two sets of process conditions input to the analysis device 1 are not limited to these examples.

[0017] Each set of process conditions includes two or more individual conditions. According to the example shown in Figure 1, the process conditions include individual condition X1, individual condition X2, and individual condition X3.

[0018] The analysis device 1 uses a predictor to calculate a predicted value Ya corresponding to the set Xa and a predicted value Yb corresponding to the set Xb.

[0019] Then, the analysis device 1 uses the SHAP algorithm to calculate a SHAP value for each individual condition with respect to the predicted value Yb based on a certain reference value Ybase. The SHAP value for each individual condition with respect to the predicted value Yb based on the reference value Ybase represents the contribution of each individual condition to the change in the predicted value Yb from the reference value Ybase, i.e., the difference (Yb - Ybase).

[0020] S bX1 is the SHAP value of the individual condition X1 for the difference (Yb-Ybase), and S bX2 is the SHAP value of the individual condition X2 for the difference (Yb-Ybase), and S bX3 is the SHAP value of the individual condition X3 for the difference (Yb-Ybase). Since each SHAP value has global rationality, the SHAP value S bX1 , SHAP value S bX2 , SHAP value S bX3 The sum of these is equal to the difference (Yb-Ybase).

[0021] The reference value Ybase is, for example, the average value of the predicted values ​​obtained when each process condition included in the training data used to generate the predictor is input to the predictor. The reference value Ybase is determined by the SHAP algorithm based on the training data.

[0022] The analysis device 1 also calculates a SHAP value for each individual condition with respect to the predicted value Ya based on the reference value Ybase by using the SHAP algorithm. The SHAP value for each individual condition with respect to the predicted value Yb based on the reference value Ybase represents the contribution of each individual condition to the change in the predicted value Ya from the reference value Ybase, i.e., the difference (Ya - Ybase).

[0023] S aX1 is the SHAP value of the individual condition X1 for the difference (Ya-Ybase), and S aX2 is the SHAP value of the individual condition X2 for the difference (Ya-Ybase), and S aX3 is the SHAP value of the individual condition X3 for the difference (Ya-Ybase). Since each SHAP value has global rationality, the SHAP value S aX1 , SHAP value S aX2 , SHAP value S aX3 The sum of these is equal to the difference (Ya-Ybase).

[0024] The analysis device 1 utilizes the fact that these SHAP values ​​have additivity, and calculates the SHAP value S for each individual condition with respect to the change in the predicted value (Yb-Ya). (b-a)X1 , S (b-a)X2 , S (b-a)X3 Specifically, the analysis device 1 obtains the SHAP value S for the difference (Yb-Ybase). bX1 , S bX2 , S bX3 SHAP value S for the difference (Ya-Ybase) aX1 , S aX2 , S aX3 By subtracting for each individual condition, the SHAP value S for each individual condition for the change in the predicted value (Yb - Ya) is calculated. (b-a)X1 , S (b-a)X2 , S (b-a)X3 Get.

[0025] SHAP value S (b-a)X1 is the SHAP value S bX1 From the SHAP value S aX1 The SHAP value S is obtained by subtracting the sigma and the sigma, and represents the contribution of the individual condition X1 to the change in the predicted value (Yb - Ya). (b-a)X2 is the SHAP value S bX2 From the SHAP value S aX2 The SHAP value S is obtained by subtracting the value Yb from the value Ya, and represents the contribution of the individual condition X2 to the change in the predicted value (Yb - Ya). (b-a)X3 is the SHAP value S bX3 From the SHAP value S aX3This value is obtained by subtracting the above and represents the contribution of the individual condition X3 to the change in the predicted value (Yb-Ya).

[0026] In this way, the analysis device 1 calculates the contribution of each individual condition to the change in the predicted value (Yb-Ya) when the process conditions are changed from set Xa to set Xb.

[0027] The analysis device 1, analysis method, and analysis program according to the embodiment will be described in detail below, but the present invention is not limited to the embodiment.

[0028] (Embodiment) FIG. 2 is a schematic diagram illustrating an example of a hardware configuration of the analysis device 1 according to the embodiment.

[0029] 2, the analysis device 1 includes a central processing unit (CPU) 10, a random access memory (RAM) 11, a storage device 12, an input device 13, and a display device 14. The CPU 10, the RAM 11, the storage device 12, the input device 13, and the display device 14 are electrically connected to one another via a bus 15.

[0030] The input device 13 is an HMI (Human Machine Interface) for inputting information. The input device 13 is configured by, for example, a pointing device, a keyboard, or a combination of these. Information input by an operator of the analysis device 1 operating the input device 13 is sent to the CPU 10.

[0031] The display device 14 is an HMI that can output information as an image. The display device 14 is an image display device such as a liquid crystal display, an organic EL (electro-luminescence) display, or a plasma display.

[0032] The CPU 10 is a processor that executes computer programs.

[0033] The RAM 11 is a volatile memory that operates faster than the storage device 12. The RAM 11 provides the CPU 10 with an area as a cache or buffer.

[0034] The storage device 12 is a memory that can non-volatilely store information such as data or computer programs. The storage device 12 may be configured, for example, by a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0035] The storage device 12 may be provided outside the analysis device 1, and the analysis device 1 and the storage device 12 may be connected via a network or the like. The storage device 12 may be a removable device such as a USB (Universal Serial Bus) memory, an SD (Secure Digital) card, or an external HDD.

[0036] In the embodiment, an analysis program 100, which is a computer program, is stored in storage device 12. For example, CPU 10 loads analysis program 100 from storage device 12 into RAM 11 and executes analysis program 100 loaded into RAM 11. Then, CPU 10 realizes the operation of analysis device 1 based on analysis program 100 loaded into RAM 11.

[0037] The storage device 12 stores a training data set 101. The training data set 101 includes a large number of pairs of process conditions and process results. The process conditions include, as individual conditions, parameters of a semiconductor manufacturing device (hereinafter referred to as device parameters), shapes of a pattern to be processed (hereinafter referred to as shape parameters), and the like. For example, the semiconductor manufacturing device may include a plasma processing device used for etching, etc. For a plasma processing device, gas flow rate, processing time, voltage value, and the like may be prepared as device parameters. For example, shape parameters may include design dimensions of a pattern. The process results include, for example, dimensions of a pattern obtained by processing, characteristics obtained by the process, processing rate, physical property values, and the like. The process results may include results related to these multiple parameters, or may be composed of a result related to only one parameter.

[0038] For example, a developer may collect process results for each process condition by actually processing a semiconductor device while varying the process conditions. The developer then pairs each collected process result with the process condition and accumulates it in a storage device. The accumulated group of pairs of process conditions and process results is then stored in the storage device 12 as a teacher data set 101. Note that the method for generating the teacher data set 101 is not limited to this.

[0039] Next, the operations executed by the analysis device 1 according to the embodiment will be described in detail with reference to FIGS.

[0040] 3 is a flowchart showing an example of the operation of the analysis device 1 according to the embodiment. The CPU 10 executes the series of operations shown in this diagram in accordance with the analysis program 100. In the explanation of the example shown in this diagram, the analysis program 100 will be described as the subject of the operations.

[0041] First, the analysis program 100 generates a predictor by machine learning using a teacher dataset 101 (S101). As the predictor, various machine learning models can be applied.

[0042] For example, the predictor may be configured by a neural network model having multiple layers including an input layer and an output layer. Each of the multiple layers includes one or more nodes. Each node is associated with an activation function and a weight determined by machine learning. The input layer is configured to receive process conditions, and the output layer is configured to output predicted values ​​of the process results. When the process conditions are input to the input layer, a processor (e.g., CPU 10) performs calculations using the activation functions and weights at each node, and a predicted value is output from the output layer.

[0043] In another example, the predictor may be configured by a Gaussian process regression model, a regression model (e.g., linear regression, support vector regression, Gaussian process regression, etc.), or a tree-based model (e.g., decision tree, random forest, etc.). Any machine learning model can be applied as a predictor as long as it is a model that returns a predicted value in response to an input.

[0044] An operator of the analytical device 1 can input two sets of process conditions Xa and Xb to the analytical device 1 by operating, for example, the input device 13. The method for inputting the sets Xa and Xb is not limited to this.

[0045] When the analysis program 100 acquires a set Xa of process conditions (S102), it calculates a predicted value Ya corresponding to the set Xa using a predictor (S103). The analysis program 100 inputs the set Xa to the input layer of the predictor and acquires the predicted value Ya output from the output layer of the predictor.

[0046] When the analysis program 100 acquires a set Xb of process conditions (S104), it calculates a predicted value Yb corresponding to the set Xb using a predictor (S105). The analysis program 100 inputs the set Xb to the input layer of the predictor and acquires the predicted value Yb output from the output layer of the predictor.

[0047] FIG. 4 is a diagram showing two example pairs of process conditions and predicted values ​​obtained by the processing of S102 to S105 shown in FIG. 3. In the example shown in this figure, the set of process conditions is composed of a total of seven individual conditions: equipment parameter X1, equipment parameter X2, equipment parameter X3, equipment parameter X4, equipment parameter X5, equipment parameter X6, and shape parameter X7. The predictor then outputs a predicted value Y of one parameter of the process result. Note that the number of parameters of the predicted values ​​output by the predictor is not limited to one. The predictor may output predicted values ​​of two or more parameters. In FIG. 4, the predicted value Y is, for example, a predicted value of a dimension in the process result.

[0048] 4, when set Xa is input to the predictor, "24.0" is obtained as the predicted value Y corresponding to set Xa. When set Xb is input to the predictor, "27.06" is obtained as the predicted value Y corresponding to set Xb.

[0049] Returning to the explanation of Figure 3. The analysis program 100 calculates the SHAP value for each individual condition of the set Xa based on the teacher data set 101 (S106). The analysis program 100 calculates the SHAP value using the SHAP algorithm. The SHAP value obtained by the process of S106 is referred to as the SHAP value S a According to the example shown in Figure 1, the SHAP value S aX1 , S aX2 , S aX3 Each of these is calculated as the SHAP value S a It is calculated as:

[0050] The analysis program 100 calculates the SHAP value for each individual condition of the set Xb based on the teacher data set 101 (S107). The analysis program 100 calculates the SHAP value using the SHAP algorithm. The SHAP value obtained by the process of S107 is referred to as the SHAP value S b According to the example shown in Figure 1, the SHAP value S bX1 , SbX2 , S bX3 Each of these is calculated as the SHAP value S b It is calculated as:

[0051] The SHAP value obtained by steps S106 and S107 represents the degree of contribution to the change in the predicted value from the reference value Ybase. The reference value Ybase is determined based on the training data set 101.

[0052] 5 is a diagram illustrating the reference value Ybase determined by the teacher data set 101 according to the embodiment. In this diagram, the reference value Ybase and the SHAP value for each individual condition with respect to the predicted value Yb are shown.

[0053] 5, the teacher data set 101 includes a plurality of sets of process conditions including individual conditions X1 to X7, and a plurality of process results Y corresponding to each set of process conditions. The SHAP algorithm obtains a group of predicted values ​​Y by inputting each set of process conditions included in the teacher data set 101 into a predictor. The SHAP algorithm then uses the average value of the group of predicted values ​​Y as a reference value Ybase to calculate a SHAP value for each individual condition.

[0054] In the example shown in Figure 5, "31.405" is obtained as the average value of the group of predicted values ​​Y, and therefore "31.405" is used as the reference value Ybase. As also shown in Figure 4, the predicted value Yb obtained by inputting set Xb into the predictor is "27.065". The SHAP algorithm calculates the SHAP value for each of the individual conditions X1 to X7 by subtracting "31.405" from "27.065" and allocating the difference obtained to the individual conditions X1 to X7.

[0055] Of the individual conditions X1 to X7 of set Xb, the SHAP value of the equipment parameter X1 is "+0.8", the SHAP value of the equipment parameter X2 is "-1.25", the SHAP value of the equipment parameter X3 is "-0.46", the SHAP value of the equipment parameter X4 is "-0.2", the SHAP value of the equipment parameter X5 is "+0.17", the SHAP value of the equipment parameter X6 is "-0.19", and the SHAP value of the shape parameter X7 is "-3.21".

[0056] In this way, the SHAP algorithm calculates the SHAP value for each individual condition based on the training data set 101.

[0057] Returning to FIG. 3 again. The analysis program 100 calculates the SHAP value S for each individual condition with respect to the change in the predicted value (Yb-Ya) when the process conditions are changed from the set Xa to the set Xb. (b-a) The analysis program 100 calculates the SHAP value S for each individual condition (S108). b From the SHAP value S a The SHAP value S for each individual condition is calculated by subtracting (b-a) Following the example shown in Figure 1, the SHAP value S (b-a)X1 , S (b-a)X2 , S (b-a)X3 Each of these has a SHAP value S (b-a) It is calculated as:

[0058] The analysis program 100 calculates the predicted values ​​Ya and Yb, and the SHAP value S for each individual condition. (b-a) In S109, the analysis program 100 outputs the predicted values ​​Ya, Yb and the SHAP value S for each individual condition to the display device 14, for example. (b-a) Then, the series of operations of the analysis device 1 is completed.

[0059] The order in which the processes of S101 to S107 are performed can be changed in various ways. For example, the processes of S104 and S105 may be performed before the processes of S102 and S103. Furthermore, the process of S107 may be performed before the process of S106.

[0060] FIG. 6 shows the SHAP value S for each individual condition according to the embodiment. (b-a) FIG. 10 is a diagram showing an example of a display method.

[0061] According to the display example shown in FIG. 6, the predicted value Ya is "23.965" and the predicted value Yb is "27.065".

[0062] The SHAP values ​​for each individual condition for the change from predicted value Ya to predicted value Yb are as follows: The SHAP value of the equipment parameter X1 is "+2.79", the SHAP value of the equipment parameter X2 is "+0.44", the SHAP value of the equipment parameter X3 is "+0.69", the SHAP value of the equipment parameter X4 is "+0.66", the SHAP value of the equipment parameter X5 is "-0.13", the SHAP value of the equipment parameter X6 is "+0.04", and the SHAP value of the shape parameter X7 is "+2.79".

[0063] From this example display, (1) By changing the process conditions from set Xa to set Xb, the predicted value changes from Ya to Yb. (2) The contribution of each individual condition to the change in the predicted value from Ya to Yb is greatest for X1, followed by X1, X7, X3, X2, X5, X4, and X6 in that order. (3) Each change in X1, X3, X2, X4, and X6 resulted in an increase in the predicted value, and each change in X7 and X5 resulted in a decrease in the predicted value. etc. can be read.

[0064] The predicted values ​​Ya, Yb, and the SHAP value S for each individual condition (b-a) The method of outputting this information is not limited to displaying it on the display device 14. The analysis program 100 may output this information to the storage device 12, or may output this information to another computer via a network.

[0065] The information output by the analysis program 100 includes the predicted values ​​Ya and Yb, and the SHAP value S for each individual condition. (b-a) The analysis program 100 may output any information in addition to the above. Furthermore, the analysis program 100 does not necessarily have to output the predicted values ​​Ya and Yb.

[0066] In the above example, the SHAP value was used as the numerical information representing the degree of contribution. The degree of contribution is not limited to the SHAP value. Any numerical information having overall rationality and additivity can be used as the numerical information representing the degree of contribution.

[0067] Furthermore, the analysis device 1 calculates the SHAP value for each of all the individual conditions constituting the process conditions. The analysis device 1 may be configured to calculate the SHAP value for some of the individual conditions constituting the process conditions. The analysis device 1 calculates the SHAP value for at least one individual condition out of the individual conditions constituting the process conditions.

[0068] In the process of S108 shown in FIG. 3, the SHAP value S of each individual condition is calculated with respect to the change in the predicted value (Yb-Ya) when the process condition is changed from the set Xa to the set Xb. (b-a) In S108, the analysis program 100 calculates the SHAP value S of each individual condition with respect to the change in the predicted value (Ya-Yb) when the process condition is changed from the set Xb to the set Xa. (a-b) In such a case, the analysis program 100 may be configured to calculate the SHAP value S for each individual condition in S108. a From the SHAP value S b The SHAP value S for each individual condition is calculated by subtracting (a-b) That is, the analysis program 100 obtains the SHAP value S for each individual condition. a and SHAP value S b The SHAP value for each individual condition with respect to the change in the predicted value may be obtained by calculating the difference between the predicted value and the actual value.

[0069] As described above, according to the embodiment, the CPU 10 calculates the contribution of at least one individual condition (referred to as a first individual condition) for set Xa, which is a set of values ​​of a plurality of individual conditions, based on the teacher dataset 101. The CPU 10 calculates the contribution of at least the first individual condition for set Xb, which is a set of values ​​of a plurality of individual conditions and is different from set Xa, based on the teacher dataset 101. The CPU 10 obtains the contribution of at least the first individual condition to a change in the predicted value caused by changing the process condition from set Xa to set Xb by subtracting the contribution of at least the first individual condition to set Xa from the contribution of set Xb. The CPU 10 then outputs the obtained contribution of at least the first individual condition to a change in the predicted value.

[0070] The operator of the analysis device 1 can obtain the contribution of each individual condition to the change in the predicted value when the process condition is changed from an arbitrary condition. Therefore, the analysis device 1 according to the embodiment is highly convenient for developers.

[0071] Furthermore, according to the embodiment, the CPU 10 uses a predictor to calculate a predicted value Ya corresponding to the set Xa and a predicted value Yb corresponding to the set Xb, and outputs the predicted values ​​Ya and Yb.

[0072] The developer can know the change in the predicted value when the process condition is changed from set Xa to set Xb. Therefore, the analysis device 1 according to the embodiment is highly convenient for the developer.

[0073] According to the embodiment, the CPU 10 can output the contribution of each of two or more individual conditions to a change in the predicted value. That is, the CPU 10 can operate as follows. Based on the teacher dataset 101, the CPU 10 calculates the contribution of a second individual condition, which is different from the first individual condition, for the set Xa. The CPU 10 calculates the contribution of the second individual condition for the set Cb. The CPU 10 obtains the contribution of the second individual condition to a change in the predicted value caused by changing the process condition from the set Xa to the set Xb by subtracting the contribution of the second individual condition for the set Xa from the contribution of the second individual condition for the set Xb. The CPU 10 then outputs the obtained contribution of the second individual condition to the change in the predicted value.

[0074] According to the embodiment, a predictor is used in which process conditions are input as a plurality of explanatory variables and a predicted value of a process result is output as a predicted value. That is, the embodiment describes an example in which machine learning is used in process development. The technology according to the embodiment is applicable not only to process development but also to any field.

[0075] That is, in the analysis device according to the embodiment, the processor uses a predictor configured to receive multiple explanatory variables as input and output predicted values ​​corresponding to the multiple explanatory variables. The processor calculates a first contribution, which is the contribution of a first explanatory variable among the multiple explanatory variables, for a first set, which is a set of values ​​of the multiple explanatory variables, based on a teacher dataset used to generate the predictor. The processor calculates a second contribution, which is the contribution of the first explanatory variable, for a second set, which is a set of values ​​of the multiple explanatory variables different from the first set, based on the teacher dataset. The processor obtains a third contribution by subtracting the first contribution from the second contribution, and outputs the third contribution.

[0076] The analysis device according to the embodiment is therefore highly convenient for the operator, as it allows the operator to know how the predicted value changes when the set of values ​​of multiple explanatory variables is changed.

[0077] The analysis program 100 executed by the analysis device 1 according to the embodiment may be provided by being stored in advance in a storage device 12. The storage device 12 is an example of a non-transitory, tangible computer-readable recording medium. The analysis program 100 may be provided by being recorded in the form of an installable or executable file on a non-transitory, tangible computer-readable storage medium such as a CD (Compact Disc)-ROM (Read Only Memory), a flexible disk (FD), a CD-R (Recordable), a DVD (Digital Versatile Disk), a USB (Universal Serial Bus) memory, or an SD (Secure Digital) card.

[0078] Furthermore, the analysis program 100 may be configured to be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the analysis program 100 may be configured to be provided or distributed via a network such as the Internet.

[0079] In the embodiment, the series of operations shown in Fig. 3 is realized by the CPU 10, which is a processor, executing the analysis program 100. Some or all of the series of operations shown in Fig. 3 may be realized by a logic circuit. Some or all of the series of operations shown in Fig. 3 may be realized by an analog circuit. Some or all of the series of operations shown in Fig. 3 may be realized by a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), or the like.

[0080] In the embodiment, the teacher dataset 101 has been described as being stored in advance in the storage device 12. The teacher dataset 101 may be stored in an external device. The analysis program 100 may be configured to acquire the teacher dataset 101 from the external device.

[0081] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0082] 1 Analysis device, 10 CPU, 11 RAM, 12 Storage device, 13 Input device, 14 Display device, 15 Bus, 100 Analysis program, 101 Training dataset.

Claims

1. calculating a first contribution, which is a contribution of a first explanatory variable among the plurality of explanatory variables, to a first set, which is a set of values ​​of the plurality of explanatory variables, based on a teacher dataset used in generating a predictor configured to receive a plurality of explanatory variables and output predicted values ​​corresponding to the plurality of explanatory variables; Calculating a second contribution, which is a contribution of the first explanatory variable, for a second set, which is a set of values ​​of the plurality of explanatory variables different from the first set, based on the teacher dataset; obtaining a third contribution by calculating a difference between the first contribution and the second contribution; outputting the third contribution; a processor that executes An analysis device comprising:

2. the processor uses the predictor to calculate a first predicted value corresponding to the first set and a second predicted value corresponding to the second set, and outputs the first predicted value and the second predicted value. The analysis device according to claim 1 .

3. the plurality of explanatory variables includes a second explanatory variable different from the first explanatory variable; the processor calculates, based on the teacher dataset, a fourth contribution that is a contribution of a second explanatory variable for the first set, a fifth contribution that is a contribution of the second explanatory variable for the second set, and obtains a sixth contribution by calculating a difference between the fourth contribution and the fifth contribution; The analysis device of claim 1 further comprising:

4. the plurality of explanatory variables are process conditions related to the manufacture of a semiconductor device, and the predicted value output by the predictor is a predicted value of a process result; The analysis device according to any one of claims 1 to 3.

5. calculating a first contribution, which is a contribution of a first explanatory variable among the plurality of explanatory variables, to a first set, which is a set of values ​​of the plurality of explanatory variables, based on a teacher dataset used in generating a predictor configured to receive a plurality of explanatory variables and output predicted values ​​corresponding to the plurality of explanatory variables; Calculating a second contribution, which is a contribution of the first explanatory variable, for a second set, which is a set of values ​​of the plurality of explanatory variables different from the first set, based on the teacher dataset; obtaining a third contribution by calculating a difference between the first contribution and the second contribution; outputting the third contribution; Analysis methods including.

6. calculating a first predicted value corresponding to the first set and a second predicted value corresponding to the second set using the predictor; outputting the first predicted value and the second predicted value; The analysis method according to claim 5 , further comprising:

7. the plurality of explanatory variables includes a second explanatory variable different from the first explanatory variable; Calculating a fourth contribution, which is a contribution of a second explanatory variable, for the first set based on the teacher dataset; calculating a fifth contribution of the second explanatory variable to the second set; obtaining a sixth contribution by calculating the difference between the fourth contribution and the fifth contribution; The analysis method of claim 5 further comprising:

8. the plurality of explanatory variables are process conditions related to the manufacture of a semiconductor device, and the predicted value output by the predictor is a predicted value of a process result; The analysis method according to any one of claims 5 to 7.

9. On the computer, calculating a first contribution, which is a contribution of a first explanatory variable among the plurality of explanatory variables, to a first set, which is a set of values ​​of the plurality of explanatory variables, based on a teacher dataset used in generating a predictor configured to receive a plurality of explanatory variables and output predicted values ​​corresponding to the plurality of explanatory variables; Calculating a second contribution, which is a contribution of the first explanatory variable, for a second set, which is a set of values ​​of the plurality of explanatory variables different from the first set, based on the teacher dataset; obtaining a third contribution by calculating a difference between the first contribution and the second contribution; outputting the third contribution; An analysis program that executes the above.

10. The computer, calculating a first predicted value corresponding to the first set and a second predicted value corresponding to the second set using the predictor; outputting the first predicted value and the second predicted value; The analysis program according to claim 9, further comprising:

11. the plurality of explanatory variables includes a second explanatory variable different from the first explanatory variable; The computer, Calculating a fourth contribution, which is a contribution of a second explanatory variable, for the first set based on the teacher dataset; calculating a fifth contribution of the second explanatory variable to the second set; obtaining a sixth contribution by calculating the difference between the fourth contribution and the fifth contribution; The analysis program according to claim 9 , further comprising:

12. the plurality of explanatory variables are process conditions related to the manufacture of a semiconductor device, and the predicted value output by the predictor is a predicted value of a process result; The analysis program according to any one of claims 9 to 11.

Images