Yield prediction device, model generation device, yield prediction system, and yield prediction method

Abstract

To accurately predict a yield.SOLUTION: One aspect of the present disclosure provides a yield prediction device that includes: an acquisition unit configured to acquire specific design information, which is design information for an object product; and a prediction unit that is machine-learned using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process that has the same steps as the object product's production process, and configured to predict an object product's yield based on the specific design information acquired by the acquisition unit.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a yield prediction device and the like. [Background technology]

[0002] There are known techniques for predicting product yields. For example, Patent Document 1 discloses a technique for predicting product yields using a trained model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2021-155772 Summary of the Invention [Problem to be solved by the invention]

[0004] The technology in Patent Document 1 uses training data for a single product as training data for generating a trained model, which may result in a small amount of training data and make it difficult to improve prediction accuracy.

[0005] An object of one aspect of the present disclosure is to realize a yield prediction device or the like that can predict yield with high accuracy. [Means for solving the problem]

[0006] A yield prediction device according to one aspect of the present disclosure includes an acquisition unit that acquires specific design information, which is design information for a target product, and a prediction unit that has been machine-learned using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process that has the same steps as the production process for the target product, and that predicts the yield of the target product based on the specific design information acquired by the acquisition unit.

[0007] A model generation device according to one aspect of the present disclosure includes a memory unit that stores training data including design information and actual yield values for each of multiple types of products that can be produced in a production process having the same steps as the production process of the target product, and a model generation unit that generates, through machine learning using the training data, a model that predicts the yield of the target product based on specific design information that is design information of the target product.

[0008] A yield prediction system according to one aspect of the present disclosure includes: a model generation unit that generates a model by machine learning using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process having the same process as the production process of a target product; and a prediction unit that predicts the yield of the target product based on specific design information, which is design information of the target product, in the model. Equipped with.

[0009] A yield prediction method according to one aspect of the present disclosure includes a model generation step of generating a model by machine learning using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process having the same steps as the production process of the target product, an acquisition step of acquiring specific design information that is design information for the target product, and a prediction step of predicting the yield of the target product by inputting the specific design information into the model.

[0010] The yield prediction device according to each aspect of the present disclosure may be realized by a computer. In this case, the control program for the yield prediction device that causes the computer to operate as each part (software element) of the yield prediction device to realize the yield prediction device on a computer, and the computer-readable recording medium on which the control program is recorded, also fall within the scope of the present disclosure. [Effects of the Invention]

[0011] According to one aspect of the present disclosure, yield can be predicted with high accuracy. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a block diagram showing a configuration of a yield prediction system according to a first embodiment of the present disclosure. [Figure 2] 10 is a chart showing an example of a flow of processing performed by the yield prediction system. [Figure 3] FIG. 10 is a block diagram showing the configuration of a yield prediction system according to a second embodiment of the present disclosure. [Figure 4] FIG. 10 is a diagram showing an example of a screen that a display control unit according to the second embodiment of the present disclosure causes to be displayed on a display unit. [Figure 5] FIG. 10 is a graph showing a case where there is a target product in which the predicted yield value and the actual yield value are different from each other. DETAILED DESCRIPTION OF THE INVENTION

[0013] [Embodiment 1] An embodiment of the present disclosure will be described in detail below. A yield prediction system 1 in this embodiment is a system that predicts the yield in the production of a target product. The yield prediction system 1 will be described in detail below. The yield prediction system 1 predicts the yield of one or more types of products manufactured by a production line using one or more manufacturing devices. In this embodiment, a configuration will be described in which the yield prediction system 1 predicts the yield of one or more types of products manufactured in a factory where multiple manufacturing devices are installed and multiple types of products are produced using the one or more manufacturing devices.

[0014] (Configuration of yield prediction system 1) 1 is a block diagram showing the configuration of a yield prediction system 1. As shown in FIG. 1, the yield prediction system 1 includes a design information management device 10 and a yield prediction device 20.

[0015] The design information management device 10 manages design information of various products produced in the factory. The design information management device 10 may manage information such as the size of the product, the thickness of the product, and the number of components formed on the product, as the product design information.

[0016] The yield prediction device 20 predicts the yield of manufactured products. The yield prediction device 20 includes a control unit 30, a memory unit 40, an input unit 50 that receives input to the yield prediction device 20, and a display unit 60 that displays various information. The input unit 50 may be, for example, a keyboard, a mouse, or a microphone. The display unit 60 may be, for example, a display device such as a monitor.

[0017] The storage unit 40 stores various data used by the control unit 30. The storage unit 40 stores a control program 41, which is a program for performing various controls of the yield prediction device 20. The storage unit 40 stores training data 42, which is used by a model generation unit 32A (described later) to generate a model M. The training data 42 includes design information 42A for various products and actual yield values 42B for each product.

[0018] The control unit 30 comprehensively controls each unit of the yield prediction device 20. The control unit 30 includes an acquisition unit 31, a prediction unit 32, and a display control unit 33 that controls the display of the display unit 60.

[0019] The acquisition unit 31 acquires design information of a target product for which a yield is to be predicted from the design information management device 10. In the following description, the design information of the target product will be referred to as specific design information. The acquisition unit 31 may output the acquired specific design information to the prediction unit 32. The acquisition unit 31 may store the acquired specific design information in the storage unit 40.

[0020] The prediction unit 32 predicts the yield of the target product and includes a model generation unit 32A and a prediction execution unit 32B.

[0021] The model generation unit 32A generates a model M, which is a trained model that outputs a predicted value of the yield of the target product. In detail, the model generation unit 32A first reads training data 42 for generating the model M from the training data 42 stored in the storage unit 40. The training data 42 for generating the model M includes design information 42A of multiple types of products that can be produced in a production process having the same steps as the production process for producing the target product, and actual yield values 42B of the multiple types of products. The multiple types of products can also be considered to be multiple products that can be produced in the same production line as the production line for producing the target product. The multiple types of products can also be considered to be products with mutually different designs. In the following explanation, a product that can be produced in a production process having the same steps as the production process for producing the target product will be referred to as a reference product.

[0022] The target product may be a product whose design is different from any of the above-mentioned multiple types of products. That is, the model generation unit 32A can generate a model M that can predict the yield of a target product that has not been produced so far. In this case, the acquisition unit 31 acquires specific design information of the target product whose design is different from any of the above-mentioned multiple types of products.

[0023] Next, the model generation unit 32A generates a model M that outputs a predicted value of the yield of the target product by inputting specific design information of the target product through machine learning using training data 42 that uses the read design information 42A of the reference product as an explanatory variable and the read actual yield value 42B of the reference product as a target variable. The model generation unit 32A may store the generated model M in the storage unit 40. The machine learning method for generating the model M is not particularly limited. The model M may be, for example, a linear regression model.

[0024] The yield prediction system 1 according to one embodiment of the present disclosure may include a model generation device having the functions of the model generation unit 32A and a storage unit storing training data 42, and the model M generated by the model generation device may be stored in the storage unit 40 of the yield prediction device 20. In this case, the prediction execution unit 32B of the yield prediction device 20 may input specific design information of the target product to the model M generated by the model generation device, thereby predicting the yield of the target product.

[0025] The prediction execution unit 32B outputs a predicted value of the yield of the target product by inputting the specific design information of the target product acquired by the acquisition unit 31 into the model M generated by the model generation unit 32A. The display control unit 33 may cause the display unit 60 to display the predicted value of the yield of the target product.

[0026] (Processing performed by the yield prediction device 20) Next, the flow of processing performed by the yield prediction system 1 will be described with reference to Fig. 2. Fig. 2 is a chart showing an example of the flow of processing performed by the yield prediction system 1.

[0027] 2, the model generation unit 32A generates a model M (step S1, model generation step). Specifically, the model generation unit 32A reads out design information 42A of multiple types of products that can be produced in a production process having the same steps as the production process for producing the target product, and actual yield values 42B of the multiple types of reference products from the storage unit 40. That is, the model generation unit 32A reads out design information 42A of multiple types of reference products and actual yield values 42B of the multiple types of reference products from the storage unit 40. The model generation unit 32A generates a model M by machine learning using training data including the read design information 42A and actual yield values 42B.

[0028] Next, the acquisition unit 31 acquires the specific design information of the target product from the design information management device 10 (step S2, acquisition step).

[0029] Next, the prediction execution unit 32B inputs specific design information into the model M to predict the yield of the target product (step S3, prediction execution step).

[0030] Here, when a model is generated using only training data for a single product, as in Patent Document 1, the amount of data available for use as training data is small, making it impossible to improve prediction accuracy. In contrast, in the yield prediction system 1 of this embodiment, as described above, data for multiple types of reference products can be used as training data 42 for generating a model. Therefore, the model generation unit 32A can generate the model M using more training data 42 than when it uses training data for only one product. As a result, the prediction accuracy of the model M can be improved.

[0031] [Embodiment 2] Other embodiments of the present disclosure will be described below. For convenience of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0032] Manufacturing equipment may deteriorate over time. As a result, the yield of manufactured products may change over time, and the predicted yield value based on the previously used model may diverge from the actual yield. Furthermore, when manufacturing a new product with a design that has not been manufactured before, the prediction based on the previously used model M may not be able to accurately predict the yield. In this embodiment, a yield prediction system that can maintain high yield prediction accuracy will be described.

[0033] Fig. 3 is a block diagram showing the configuration of a yield prediction system 1A in this embodiment. As shown in Fig. 3, the yield prediction system 1A includes, in addition to the configuration of the yield prediction system 1 in embodiment 1, an actual yield value management device 11 that manages the actual yield value for each of the manufactured products. The yield prediction system 1A includes a yield prediction device 20A instead of the yield prediction device 20 in embodiment 1. The yield prediction device 20A includes a control device 30A instead of the control device 30 in embodiment 1. The control device 30A includes a display control device 33A instead of the display control device 33 in embodiment 1.

[0034] The display control unit 33A causes the display unit 60 to display information indicating the predicted value of the yield of the target product predicted by the prediction unit 32 and the actual value of the yield of the target product. The display control unit 33A acquires the actual value of the yield of the target product from the actual yield value management device 11.

[0035] FIG. 4 is a diagram illustrating an example of a screen displayed by the display control unit 33A on the display unit 60. As shown in FIG. 4, the display control unit 33A may cause the display unit 60 to display a graph G plotting the target products, with the actual yield values of the target products on the horizontal axis and the predicted yield values of the target products predicted by the prediction unit 32 on the vertical axis. Each point shown in FIG. 4 represents a plot of the target products for each lot. FIG. 4 illustrates a graph G plotted for multiple types of target products. As shown in FIG. 4, the display control unit 33A may cause the display unit 60 to display a graph G including a line L connecting points where the actual yield values and the predicted yield values are the same. This makes it easier for the user to visually recognize the difference between the actual yield values and the predicted yield values. In the graph shown in FIG. 4, plots for all types of target products are plotted near the line L, indicating the high prediction accuracy of model M.

[0036] 5 is a graph G showing a case where there is a discrepancy between the predicted yield and the actual yield for a target product. For example, when a manufacturing device has deteriorated or when predicting the yield of a new product with a design that has not been manufactured before, the predicted yield and the actual yield for the target product may differ significantly, as in the case of the target product included in region C in FIG. 5.

[0037] In the present embodiment, the yield prediction system 1A generates a new model M when the frequency with which the difference between the predicted yield and the actual yield of a target product produced during a predetermined period exceeds a predetermined value, in other words, when the percentage of times when the difference exceeds a predetermined value, exceeds a criterion. For example, if a user determines from the graph displayed on the display unit 60 that the percentage of times when the difference between the predicted yield and the actual yield is 10% or more exceeds the criterion of 20% overall, the user may instruct the yield prediction device 20A to generate a new model M. The predetermined value is not particularly limited, and may be 10% as described above, or another value. The criterion is not particularly limited, and may be more than 20% as described above, or another value. In the following description, a product in which the difference between the predicted yield and the actual yield of the target product exceeds a predetermined value will be referred to as a deviation product.

[0038] In the yield prediction device 20A of this embodiment, when a user instructs to generate a new model M, the model generation unit 32A generates a new model M. In other words, the model generation unit 32A updates the model M (update step). In this case, the model generation unit 32A may generate the new model M using training data including specific design information and actual yield values for the deviation product. As a result, the newly generated model M becomes a model that can accurately predict the yield of the deviation product.

[0039] When updating model M, model generation unit 32A may use training data 42 acquired at a time closer to the time of updating model M than the training data 42 used to generate model M before the update. This allows model M to be a model that better reflects the deterioration state of the manufacturing equipment at the time of prediction. As a result, model M can output a predicted value of yield that reflects the influence of deterioration of the manufacturing equipment, thereby improving the accuracy of yield prediction.

[0040] In the yield prediction system 1 of the present embodiment, the yield prediction device 20 includes the model generation unit 32A and the prediction execution unit 32B, but the configuration is not limited to this. In the yield prediction system 1 of one aspect of the present disclosure, the functions of the model generation unit 32A and the functions of the prediction execution unit 32B may be installed in separate devices.

[0041] [Software implementation example] The functions of the yield prediction device 20 / 20A (hereinafter referred to as the "device") can be realized by a program that causes a computer to function as the device, and a program that causes a computer to function as each control block of the device (particularly each part included in the control unit 30 / 30A).

[0042] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The control device and storage device execute the program, thereby realizing the functions described in each of the above embodiments.

[0043] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0044] In addition, some or all of the functions of each of the control blocks can be realized by logic circuits. For example, integrated circuits in which logic circuits that function as each of the control blocks are formed are also included in the scope of the present disclosure. In addition, the functions of each of the control blocks can also be realized by, for example, a quantum computer.

[0045] The invention according to the present disclosure has been described above based on the drawings and examples. However, the invention according to the present disclosure is not limited to the above-described embodiments. In other words, the invention according to the present disclosure can be modified in various ways within the scope of the present disclosure, and embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the invention according to the present disclosure. In other words, it should be noted that a person skilled in the art can easily make various modifications or corrections based on the present disclosure. It should also be noted that these modifications or corrections are included in the scope of the present disclosure.

[0046] 〔summary〕 A yield prediction device according to aspect 1 of the present disclosure includes an acquisition unit that acquires specific design information, which is design information for a target product, and a prediction unit that has been machine-learned using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process that has the same steps as the production process for the target product, and that predicts the yield of the target product based on the specific design information acquired by the acquisition unit.

[0047] A yield prediction device according to a second aspect of the present disclosure may be configured in the first aspect above such that the acquisition unit acquires the specific design information of the target product, the design of which is different from any of the plurality of types of products.

[0048] A yield prediction device according to aspect 3 of the present disclosure may be configured in the above-mentioned aspect 1 or 2 to include a display control unit that causes a display unit to display a graph plotting the target products with the actual yield value on the horizontal axis and the predicted yield value predicted by the prediction unit on the vertical axis.

[0049] A model generation device according to a fourth aspect of the present disclosure includes a memory unit that stores training data including design information and actual yield values for each of multiple types of products that can be produced in a production process having the same steps as the production process of the target product, and a model generation unit that generates, through machine learning using the training data, a model that predicts the yield of the target product based on specific design information that is the design information of the target product.

[0050] A model generation device according to aspect 5 of the present disclosure may be configured in the above-mentioned aspect 4 such that the model generation unit generates a new model using training data including the specific design information and the actual yield value for a target product for which the difference between the predicted yield value predicted based on the specific design information and the actual yield value for the target product is equal to or greater than a predetermined value.

[0051] A yield prediction system according to aspect 6 of the present disclosure includes a model generation unit that generates a model through machine learning using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process having the same steps as the production process of a target product, and a prediction execution unit that predicts the yield of the target product by inputting specific design information, which is the design information of the target product, into the model.

[0052] A yield prediction method according to aspect 7 of the present disclosure includes a model generation step of generating a model by machine learning using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process having the same steps as the production process of the target product, an acquisition step of acquiring specific design information that is design information for the target product, and a prediction execution step of predicting the yield of the target product by inputting the specific design information into the model.

[0053] The yield prediction method according to aspect 8 of the present disclosure may be configured in the above-mentioned aspect 7, further including an update step of generating a new model when the frequency with which the difference between the predicted yield value and the actual yield value of the target product is equal to or greater than a predetermined value exceeds a standard. [Explanation of symbols]

[0054] 1. 1A Yield Prediction System 20, 20A Yield Prediction Device 31 Acquisition Department 32 Prediction Department 32A Model Generation Unit 32B Prediction Execution Unit 33A Display control unit 40 Storage section 42 Training Data 42A design information 42B Actual value 60 Display

Claims

1. an acquisition unit that acquires specific design information that is design information of the target product; a prediction unit that is machine-learned using training data including design information and actual yield values for each of a plurality of types of products that can be produced in a production process having the same process as the production process of the target product, and that predicts the yield of the target product based on the specific design information acquired by the acquisition unit; A yield prediction device comprising:

2. 2 . The yield prediction device according to claim 1 , wherein the acquisition unit acquires the specific design information of the target product, the design of which is different from any of the plurality of types of products.

3. 2. The yield prediction device according to claim 1, further comprising a display control unit that causes a display unit to display a graph in which the target products are plotted, with the actual yield value on the horizontal axis and the predicted yield value predicted by the prediction unit on the vertical axis.

4. a storage unit that stores training data including design information and actual yield values for each of a plurality of types of products that can be produced in a production process that has the same steps as the production process of the target product; a model generation unit that generates a model that predicts the yield of the target product based on specific design information that is design information of the target product by machine learning using the training data.

5. 5. The model generating device according to claim 4, wherein the model generating unit generates the new model using training data including the specific design information and the actual yield value for a target product for which a difference between a predicted yield value predicted based on the specific design information and the actual yield value for the target product is equal to or greater than a predetermined value.

6. a model generation unit that generates a model by machine learning using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process that has the same process as the production process of the target product; a prediction execution unit that predicts the yield of the target product by inputting specific design information, which is design information of the target product, into the model; A yield prediction system comprising:

7. a model generation step of generating a model by machine learning using training data including design information and actual yield values for each of multiple types of products that can be produced in a production process having the same steps as the production process of the target product; an acquisition step of acquiring specific design information which is design information of the target product; a prediction execution step of predicting the yield of the target product by inputting the specific design information into the model; A method for predicting yield, comprising:

8. 8. The yield prediction method according to claim 7, further comprising an update step of generating a new model when the frequency at which the difference between the predicted yield value and the actual yield value of the target product is equal to or greater than a predetermined value exceeds a standard.

Images