TWI830812B - Substrate processing condition setting support method, substrate processing system, storage medium, and learning model - Google Patents

Abstract

An object of the invention is to provide a condition setting support method which is effective for simplifying the task of setting processing conditions for substrate processing. The substrate processing condition setting support method includes a step of inputting, into a machine learning device, a data set that contains processing conditions for substrate processing performed by a substrate processing device including the supply of a process liquid to the substrate and also contains results data related to the quality of the substrate processing, and a step of deriving recommended processing conditions for substrate processing based on a learning model which is generated by the machine learning device using machine learning based on multiple data sets and which outputs prediction data related to the quality of substrate processing according to the input processing conditions.

Description

Substrate processing condition setting support method, substrate processing system, recording medium and learning model

The content disclosed in the present invention relates to a substrate processing condition setting support method, a substrate processing system, a recording medium and a learning model.

Patent Document 1 discloses a device that forms a photosensitive coating film on the surface of a substrate, and performs a development process on the photosensitive coating film after the exposure process of the photosensitive coating film.

[Known technical documents]

[Patent Document]

Patent Document 1: Japanese Patent Application Publication No. 2017-73522

The disclosure of the present invention provides a condition setting support method, which is effective in simplifying the setting operation of processing conditions for substrate processing.

A condition setting support method for substrate processing according to one aspect of the present invention includes the following steps: setting processing conditions including substrate processing (including supply of processing liquid to the substrate) performed by a substrate processing device, and information about the substrate. The data set of the actual performance data of the processing quality is input to the machine learning device; and based on the model generated by the "machine learning device through machine learning based on the plurality of data sets", that is, in response to the input of the processing conditions, the information about the substrate is "Learning model for predicting data output of processing quality" to derive recommended processing conditions for substrate processing.

According to the disclosure of the present invention, a condition setting support method can be provided, which is effective in simplifying the setting operation of processing conditions for substrate processing.

1:Substrate processing system

2: Coating and developing device (substrate processing device)

3: Exposure device

4: Transport block

5: Processing blocks

6:Interface block

7:Condition setting system

11,12,13,14: Processing module (processing department)

20,50: Rotation holding part

21,51,83:Maintenance Department

22,52: Rotary drive unit

30:Developer supply part

31,41,61:Nozzle

32,42: Nozzle transfer part

33,43,62:Liquid source

40: Flushing fluid supply part

60: Film-forming fluid supply department

70:Quality inspection device

80: Post-processing inspection department

81,95,96,97:Photography Department

82: Projection reflection part

84: Linear drive department

86:Half mirror

87:Light source

90: Processing Inspection Department

91: Liquid splash detection department

92: Liquid formation state detection part

93: Liquid drop detection department

94: Irradiation Department

100:Control device

111: Processing condition storage department

112: Processing Control Department

113:Data Acquisition Department

114:Data Entry Department

115: Recommendation conditions derivation department

116: Condition Evaluation Department

117: Duplicate Management Department

118:Performance Data Correction Department

121: Evaluation condition input part

122:Search result acquisition department

190,290:Circuit

191,291:processor

192,292: memory

193,293: storage

194:Display device

195:Input device

196:Input/output port

197,294: Communication port

200: Mechanical learning device

211: Search Operation Department

212:Data Acquisition Department

213:Data Storage Department

214: Model Generation Department

215: Model Preservation Department

216:Condition search department

A1,A8: transfer arm

A3: Carrying arm

A7:Lifting arm

C:Vehicle

S01~S09, S11~S14, S21~S28, S31~S39, S41, S51~S58: Steps

U1: coating unit

U2, U4: heat treatment unit

U3:Developing unit

U10,U11: Storage Department

W:wafer

Wa: surface

FIG. 1 is a schematic diagram showing the structure of a substrate processing system according to an exemplary embodiment.

FIG. 2 is a schematic diagram illustrating the schematic configuration of a coating unit.

FIG. 3 is a schematic diagram illustrating the schematic structure of the developing unit.

FIG. 4 is a schematic diagram illustrating the schematic configuration of the post-process inspection device.

FIG. 5 is a schematic diagram illustrating the schematic configuration of the inspection device during processing.

FIG. 6 is a block diagram illustrating the functional configuration of the control device and the machine learning device.

FIG. 7 is a block diagram illustrating the hardware configuration of the control device and the machine learning device.

FIG. 8 is a flowchart illustrating a condition setting support procedure executed by the control device.

FIG. 9 is a flowchart illustrating a condition setting support procedure further executed by the control device.

FIG. 10 is a flowchart illustrating the condition setting support procedure executed by the machine learning device.

FIG. 11 is a flowchart illustrating a condition setting support procedure further executed by the machine learning device.

Various exemplary embodiments will be described below. In the description, the same elements or elements with the same function are given the same symbols, and repeated explanations are omitted.

[Substrate processing system]

The substrate processing system 1 is a system that forms a photosensitive coating film on the surface of a substrate and develops the photosensitive coating film after exposure processing. The substrate to be processed is, for example, a semiconductor wafer W. The photosensitive coating film is, for example, a photoresist film.

As illustrated in FIG. 1 , the substrate processing system 1 includes a coating and developing device 2 and a control device 100 . The coating and developing device 2 includes a carrying block 4, a processing block 5, and an interface block 6.

The carrying block 4 carries out the introduction of the wafer W (substrate) into the coating and developing device 2 and the removal of the wafer W from the coating and developing device 2 . For example, the carrying block 4 can support a plurality of carriers C for the wafer W and has a built-in transfer arm A1. The carrier C stores, for example, a plurality of circular wafers W. The transfer arm A1 takes out the unprocessed wafer W from the carrier C and returns the processed wafer W to the carrier C.

The processing block 5 is provided with a plurality of processing modules 11, 12, 13, and 14. The processing modules 11, 12, and 13 (processing units) perform film formation processing: coating a film-forming liquid (film-forming processing liquid) on the surface Wa of the wafer W to form a coating film. For example, the processing modules 11, 12, and 13 are built with a coating unit U1, a heat treatment unit U2, and a transfer arm A3 that transfers the wafer W to these units.

The processing module 11 forms an underlying film on the surface of the wafer W through the coating unit U1 and the heat treatment unit U2. The coating unit U1 of the processing module 11 applies the processing liquid for forming the lower layer film on the wafer W. The heat treatment unit U2 of the treatment module 11 performs various heat treatments accompanying the formation of the underlying film.

The processing module 12 forms a photoresist film on the lower film through the coating unit U1 and the heat treatment unit U2. The coating unit U1 of the processing module 12 applies the processing liquid for forming the photoresist film on the underlying film. The heat treatment unit U2 of the processing module 12 performs various heat treatments accompanying the formation of the photoresist film.

The processing module 13 forms an upper layer film on the photoresist film through the coating unit U1 and the heat treatment unit U2. The coating unit U1 of the processing module 13 applies the liquid for forming the upper layer film on the photoresist film. The heat treatment unit U2 of the treatment module 13 performs various heat treatments accompanying the formation of the upper layer film.

As illustrated in FIG. 2 , the coating unit U1 includes a rotation holding unit 50 and a film-forming liquid supply unit 60 . The rotation holding unit 50 holds the wafer W and rotates it. For example, the rotation holding part 50 includes a holding part 51 and a rotation driving part 52 . The holding portion 51 supports the horizontally arranged wafer W, and holds the wafer W by, for example, vacuum suction. The rotation drive unit 52 uses, for example, an electric motor or the like as a power source to rotate the holding unit 51 around a vertical axis. Thereby, the wafer W held by the holding part 51 also rotates.

The film-forming liquid supply unit 60 supplies the film-forming liquid to the surface Wa of the wafer W held by the holding unit 51 . For example, the film-forming liquid supply unit 60 includes a nozzle 61 and a liquid source 62 . The nozzle 61 is arranged above the wafer W held by the holding part 51 and sprays the processing liquid downward. The liquid source 62 sends the processing hydraulic pressure to the nozzle 61 .

Returning to FIG. 1 , the processing module 14 (processing unit) performs a development process: a development treatment liquid is supplied to the overexposed photoresist film (photosensitive coating film) on the surface Wa of the wafer W. For example, the processing module 14 includes a developing unit U3, a heat treatment unit U4, and a transport arm A3 for transporting the wafer W to these units. The processing module 14 performs development processing on the exposed photoresist film through the developing unit U3 and the heat treatment unit U4. The developing unit U3 applies a developer solution (a treatment liquid for development) to the surface of the exposed wafer W, and then washes it away with a rinse liquid (a treatment liquid for rinse), thereby performing development treatment of the photoresist film. The heat treatment unit U4 performs various heat treatments accompanying the development process. Specific examples of the heat treatment include heat treatment before development (PEB: Post Exposure Bake), heat treatment after development (PB: Post Bake), and the like.

As illustrated in FIG. 3 , the developing unit U3 includes a rotation holding unit 20 , a developer supply unit 30 , and a rinse liquid supply unit 40 . The rotation holding unit 20 holds and rotates the wafer W. For example, the rotation holding part 20 includes a holding part 21 and a rotation driving part 22. The holding portion 21 supports the horizontally arranged wafer W, and holds the wafer W by, for example, vacuum suction. The rotation drive unit 22 uses, for example, an electric motor or the like as a power source to rotate the holding unit 21 around a vertical axis. Thereby, the wafer W held by the holding part 21 also rotates.

The developer supply unit 30 supplies the developer to the surface Wa of the wafer W held by the holding unit 21 . For example, the developer supply unit 30 includes a nozzle 31 , a nozzle transfer unit 32 , and a liquid source 33 . The nozzle 31 is arranged above the wafer W held by the holding part 21 and sprays the developer downward. Nozzle transfer part 32, An electric motor or the like is used as a power source to move the nozzle 31 in the horizontal direction. The liquid source 33 sends the developing hydraulic pressure to the nozzle 31 .

The rinse liquid supply unit 40 supplies the rinse liquid to the surface Wa of the wafer W held by the holding unit 21 . For example, the flushing liquid supply unit 40 includes a nozzle 41 , a nozzle transfer unit 42 , and a liquid source 43 . The nozzle 41 is arranged above the wafer W held by the holder 21 and sprays the rinse liquid downward. The nozzle transfer unit 42 uses an electric motor or the like as a power source to move the nozzle 41 in the horizontal direction. The liquid source 43 sends flushing hydraulic pressure to the nozzle 41 .

Returning to FIG. 1 , the interface block 6 transfers the wafer W between the wafer W and the exposure device (not shown) that performs the exposure process of the photoresist film formed on the wafer W. For example, the interface block 6 has a built-in transfer arm A8 connected to the exposure device. The transfer arm A8 transfers the wafer W before exposure processing to the exposure device, and receives the wafer W after exposure processing from the exposure device.

Between the processing block 5 and the carrying block 4, a storage unit U10 is provided. The storage part U10 is divided into a plurality of small units arranged side by side in the up and down direction, and the wafer W can be stored in each small unit. The storage unit U10 is used for transferring the wafer W between the carrying block 4 and the processing block 5 . A lifting arm A7 is provided near the storage part U10. The lifting arm A7 lifts and lowers the wafer W between the small units of the storage unit U10. Between the processing block 5 and the interface block 6, a storage portion U11 is provided. The storage part U11 is also divided into a plurality of small units arranged side by side in the up and down direction, and the wafer W can be stored in each small unit. The storage unit U11 is used for transferring the wafer W between the processing block 5 and the interface block 6 .

The control device 100 controls, for example, the coating and developing device 2 so that the coating and developing process is performed in the following sequence. First, the control device 100 controls the transfer arm A1 to transport the wafer W in the carrier C to the storage unit U10, and controls the lifting arm A7 to place the wafer W in the small unit for the processing module 11.

Next, the control device 100 controls the transport arm A3 to transport the wafer W in the storage unit U10 to the coating unit U1 and the heat treatment unit U2 in the processing module 11 . In addition, the control device 100 controls the coating unit U1 and the heat treatment unit U2 to form an underlying film on the surface of the wafer W. Then, the control device 100 controls the transfer arm A3 to return the wafer W with the underlying film to the storage unit U10 , and controls the lifting arm A7 to place the wafer W in the small unit for the processing module 12 .

Next, the control device 100 controls the transport arm A3 to transport the wafer W in the storage unit U10 to the coating unit U1 and the heat treatment unit U2 in the processing module 12 . In addition, the control device 100 controls the coating unit U1 and the heat treatment unit U2 to form a photoresist film on the lower film of the wafer W. Then, the control device 100 controls the transfer arm A3 to return the wafer W to the storage unit U10 and controls the lifting arm A7 to place the wafer W in the small unit for the processing module 13 .

Next, the control device 100 controls the transport arm A3 to transport the wafer W in the storage unit U10 to each unit in the processing module 13 . In addition, the control device 100 controls the coating unit U1 and the heat treatment unit U2 to form an upper layer film on the photoresist film of the wafer W. Then, the control device 100 controls the transport arm A3 to transport the wafer W to the storage unit U11.

Next, the control device 100 controls the transfer arm A8 to send the wafer W in the storage part U11 to the exposure device 3 . Then, the control device 100 controls the transfer arm A8 to receive the overexposed wafer W from the exposure device 3 and arrange it in the small unit for the processing module 14 of the storage unit U11.

Next, the control device 100 controls the transport arm A3 to transport the wafer W in the storage unit U11 to each unit in the processing module 14, and controls the developing unit U3 and the heat treatment unit U4 to develop the photoresist film of the wafer W. handle. Then, the control device 100 controls the transfer arm A3 to return the wafer W to the storage unit U10, and controls the lifting arm A7 and the transfer arm A1 to return the wafer W to the carrier C. In the above manner, the coating and development process is completed.

In addition, the specific structure of the substrate processing system is not limited to the types illustrated above. The substrate processing system may have any configuration as long as it includes a processing unit that performs substrate processing including "supply of processing liquid to the substrate" and a control device 100 that can control the processing unit.

[Condition setting support system]

The substrate processing system 1 further includes a condition setting system 7 . The condition setting system 7 includes a quality inspection device 70 . In addition, at least part of the condition setting system 7 is composed of the above-mentioned control device 100 . That is, the condition setting system 7 includes the quality inspection device 70 and the control device 100 . The quality inspection device 70 detects information on the quality of the substrate processing performed by the coating and developing device 2 .

The control device 100 is configured to perform the following steps: following preset processing conditions, causing the coating and developing device 2 (substrate processing device) to perform substrate processing including supplying the processing liquid to the wafer W; and obtaining from the quality inspection device 70 Actual performance data on the quality of substrate processing according to the processing conditions; inputting a data set including the processing conditions of the substrate processing and actual performance data of the substrate processing to the machine learning device 200; and in response to the input of the processing conditions, information about the substrate The prediction data of the processing quality is output in a manner that the recommended processing conditions for substrate processing are derived based on the learning model generated by the machine learning device 200 through machine learning based on the complex data set. Forecast data, such as data that predicts the performance data mentioned above material. Performance data can be any data regarding the quality of substrate processing. The information on the quality of the substrate after substrate processing is related to the quality of the substrate processing. In addition, the supply state of the processing liquid during substrate processing is also related to the quality of the substrate processing.

The condition setting system 7 may further include a machine learning device 200 . The machine learning device 200 is configured to perform the following steps: obtain the above-mentioned data set; and generate the above-mentioned learning model through machine learning based on a plurality of data sets. The machine learning device 200 can be stored in the same casing as the control device 100 , or can be installed at a location far away from the control device 100 . When installed at a location far away from the control device 100, the machine learning device 200 is connected to the control device 100 via, for example, a local area network. The machine learning device 200 can also be connected to the control device 100 via a wide area network such as the Internet. Below, the structure of each part is demonstrated in detail.

(Quality data detection device)

The quality inspection device 70 includes, for example, a post-process inspection unit 80 shown in FIG. 4 . The post-processing inspection unit 80 detects information on the quality of the substrate after the substrate processing. As an example, the post-process inspection unit 80 detects information about the line width of the photoresist pattern formed on the surface of the wafer W after the development process. For example, the post-processing inspection unit 80 detects image information that can identify the difference in line width of the photoresist pattern as a difference in at least any one of hue, brightness, and chroma.

Specifically, the post-processing inspection unit 80 includes a holding unit 83 , a linear drive unit 84 , an imaging unit 81 , and a projection reflection unit 82 . The holding portion 83 holds the wafer W horizontally. The linear drive unit 84 uses, for example, an electric motor or the like as a power source to move the holding unit 83 along a horizontal linear path. The imaging unit 81 obtains image data of the surface of the wafer W. The imaging unit 81 is provided on one end side of the post-process inspection unit 80 in the moving direction of the holding unit 83 and faces the other end side in the moving direction.

The projection reflection part 82 projects light to the imaging range and guides the reflected light from the imaging range to the imaging part 81 side. For example, the projection reflection unit 82 includes a half mirror 86 and a light source 87 . The half mirror 86 is provided at a higher position than the holding portion 83 in the middle of the moving range of the holding portion 83, and reflects light from below toward the imaging portion 81 side. The light source 87 is provided above the half-reflecting mirror 86 and irradiates illumination light downward through the half-reflecting mirror 86 .

The post-process inspection unit 80 operates as follows to obtain image data of the surface of the wafer W. First, the linear drive part 84 moves the holding part 83. Thereby, the wafer W passes under the half mirror 86 . During this passing process, the reflected light from each part of the surface of the wafer W is sequentially sent to the imaging unit 81 . The imaging unit 81 images the reflected light from each part of the surface of the wafer W to obtain image data of the surface of the wafer W. Through this, the image information of the photoresist pattern is detected.

The post-process inspection unit 80 can also detect information on the film thickness of the coating film formed on the surface of the wafer W after the film formation process. For example, the post-processing inspection unit 80 detects image information that can identify a difference in film thickness of the coating film as a difference in at least any one of hue, brightness, and chroma. This image information can also be detected through the structure illustrated in Figure 4.

The quality inspection device 70 may further include an in-process inspection unit 90 shown in FIG. 5 . The in-process inspection unit 90 detects information on the supply status of the processing liquid during substrate processing. As an example, the in-process inspection unit 90 detects information on the supply status of the developer during the development process. For example, the in-process inspection unit 90 includes a liquid splash detection unit 91 , a liquid formation state detection unit 92 , and a liquid drop detection unit 93 .

The liquid splash detection unit 91 detects information on the occurrence state of liquid splash during the supply of developer. For example, the liquid splash detection unit 91 includes an irradiation unit 94 and an imaging unit 95 . The irradiation part 94 is fixed to, for example, the nozzle 31 or the like, and irradiates the laser light above the wafer W in the horizontal direction. The installation height of the irradiation part 94 is set to a height that liquid droplets splashed from the surface Wa can reach. The imaging unit 95 acquires image data of the irradiation range of the laser light from the irradiation unit 94 . If liquid splash occurs, the scattered droplets of the liquid cause scattering of laser light, etc., and the image data acquired by the imaging unit 95 changes. Therefore, the image data acquired by the imaging unit 95 includes information on the generation state of droplets.

The liquid formation state detection unit 92 detects information on the state of formation of a liquid film of the developer on the surface Wa. For example, the liquid formation state detection unit 92 includes an imaging unit 96 . The imaging unit 96 acquires image data of the surface Wa of the wafer W held in the holding unit 21 . The image data acquired by the imaging unit 96 includes information on the formation state of the liquid film.

The liquid dripping detection unit 93 detects information on the occurrence state of liquid dripping of the developer from the nozzle 31 . Liquid dripping refers to a phenomenon in which the developer drops from the nozzle 31 outside the preset supply period of the developer. For example, the liquid drop detection unit 93 includes an imaging unit 97 . The imaging unit 97 acquires image data of the nozzle 31 and its lower part. The image data acquired by the imaging unit 97 includes information on the occurrence state of liquid dripping.

The in-process inspection unit 90 can also detect information on the supply status of the film-forming liquid during the film-forming process. In this case, information on the supply status of the film-forming liquid to the coating unit U1 can also be detected by using the same configuration as the liquid splash detection unit 91 , the liquid formation state detection unit 92 , and the liquid drop detection unit 93 .

(Control device and machine learning device)

As shown in FIG. 6 , the control device 100 includes a processing condition storage unit 111 , a processing control unit 112 , a data acquisition unit 113 , a data input unit 114 , and a recommendation unit as a functional configuration (hereinafter referred to as "functional module"). Condition derivation unit 115.

The processing condition storage unit 111 records preset processing conditions. For example, the processing condition storage unit 111 records the development processing conditions of the processing module 14 . The development treatment conditions include the heat treatment conditions of the heat treatment unit U4 and the liquid treatment conditions of the development unit U3. The liquid processing conditions of the developing unit U3 include procedures such as supply of developer, supply of rinse liquid, and drying (drying by rotation). In addition, the liquid processing conditions of the above-mentioned developing unit U3 include: the rotation speed of the wafer W in each program, the supply amount of the developer, the supply time of the developer, the supply amount of the rinse liquid, the ejection time of the rinse liquid, and the rejection time. Dry time etc. When the nozzle transfer unit 32 moves the nozzle 31 and supplies the developer, the liquid processing conditions of the developing unit U3 may further include: the movement start position, movement speed, movement end position, etc. of the nozzle 31 during supply of the developer. .

The processing condition storage unit 111 may also record the film formation processing conditions of the processing modules 11, 12, and 13. The film forming treatment conditions include liquid treatment conditions of the coating unit U1 and heat treatment conditions of the heat treatment unit U2. The liquid treatment conditions of the coating unit U1 include procedures such as supply of film-forming liquid. In addition, the liquid processing conditions of the coating unit U1 include the rotation speed of the wafer W in each program, the supply amount of the film-forming liquid, the supply time of the film-forming liquid, and the like.

The processing control unit 112 causes the processing unit to perform substrate processing in accordance with the processing conditions recorded in the processing condition storage unit 111 . For example, the process control unit 112 causes the processing module 14 to execute the development process in accordance with the development process conditions recorded in the process condition storage unit 111 . As an example, the processing control unit 112 follows the preset Under certain heat treatment conditions, the heat treatment unit U4 is controlled to perform heat treatment on the exposed wafer W (such as the above-mentioned PEB). Then, the processing control unit 112 controls the developing unit U3 to perform developing processing on the wafer W according to the preset liquid processing conditions. Then, the process control unit 112 follows the preset heat treatment conditions and controls the heat treatment unit U4 to perform heat treatment on the wafer W (for example, the above-mentioned PB).

The process control unit 112 may also cause the process modules 11, 12, and 13 to perform film formation processing in accordance with the film formation process conditions recorded in the process condition storage unit 111. As an example, the process control unit 112 follows preset liquid processing conditions and controls the coating unit U1 to apply the film-forming liquid to the surface Wa of the wafer W. Then, the process control unit 112 controls the heat treatment unit U2 to perform heat treatment on the wafer W according to the preset heat treatment conditions.

The data acquisition unit 113 acquires actual performance data on the quality of substrate processing according to the processing conditions. The data acquisition unit 113 can acquire performance data including performance values of a plurality of items. The actual performance values of the plurality of items may include a post-processing item indicating the quality of the wafer W after substrate processing, and an actual performance value of an in-process item indicating the supply status of the processing liquid during substrate processing. For performance values of multiple projects, performance data containing multiple performance values of the same type can also be obtained. A plurality of performance values of the same type means a plurality of performance values that should ideally be the same value. Specific examples of a plurality of performance values of the same type include a plurality of performance values obtained at a plurality of places.

For example, the data acquisition unit 113 acquires, as an example of a post-processing item, an actual performance value (hereinafter referred to as "line width performance value") indicating the line width of the photoresist pattern formed on the surface Wa of the wafer W by the development process. Performance value. Specifically, the data acquisition unit 113 acquires the actual line width performance value based on the information detected by the post-processing inspection unit 80 . The data acquisition unit 113 may also acquire actual line width performance values at multiple locations on the surface Wa based on the information detected by the post-processing inspection unit 80 .

The data acquisition unit 113 acquires an actual performance value indicating the supply status of the developer during the development process as an example of the item being processed. Specifically, based on the information detected by the in-process inspection unit 90 , the data acquisition unit 113 acquires the actual performance value of the presence or absence of liquid splashing of the developer, defective liquid film formation, and liquid dripping.

The data acquisition unit 113 may also acquire, as an example of a post-processing item, an actual performance value indicating the film thickness of the coating film formed on the surface Wa of the wafer W by the film formation process (hereinafter referred to as the "film thickness performance value"). actual performance value. Specifically, the data acquisition unit 113 can acquire the actual film thickness performance value based on the information detected by the post-process inspection unit 80 . The data acquisition unit 113 may also acquire actual film thickness performance values at multiple locations on the surface Wa based on the information detected by the post-process inspection unit 80 .

The data acquisition unit 113 may also acquire an actual performance value indicating the supply status of the film-forming liquid in the film-forming process as an example of an item being processed. Specifically, the data acquisition unit 113 can acquire the actual performance value of the presence or absence of liquid splashing of the film-forming liquid, formation failure of the liquid film, and liquid dripping based on the information detected by the in-process inspection unit 90 .

The data input unit 114 inputs a data set to the model generation unit 214 (described later) of the machine learning device 200; the data set includes processing conditions and performance data corresponding to the processing conditions. The data input unit 114 can select a data set to be input to the model generation unit 214 based on the actual performance values of the items being processed. For example, the data input unit 114 may exclude a data set in which the supply status of the processing liquid is defective from being input to the model generation unit 214 . Specific examples of a defective supply state of the processing liquid include the occurrence of at least one of the above-described liquid splashing, defective liquid film formation, and liquid dripping.

The recommended condition deriving unit 115 derives recommended processing conditions for substrate processing based on the learning model generated by the model generating unit 214 through machine learning based on the complex data set. As will be described later, the learning model is generated by outputting prediction data on the quality of substrate processing in response to input of processing conditions. Recommended processing conditions are processing conditions that are judged to be recommended based on the learning model and the established evaluation conditions of the prediction data.

For example, the recommendation condition derivation unit 115, as a more subdivided functional module, includes an evaluation condition input unit 121 and a search result acquisition unit 122. The evaluation condition input unit 121 inputs the evaluation conditions of the prediction data to the condition search unit 216 (described later) of the machine learning device 200 . The evaluation conditions are conditions for determining whether the prediction data is of an acceptable level.

The evaluation condition input unit 121 may also input evaluation conditions for evaluating the predicted values of multiple items into the condition search unit 216 . The evaluation condition input unit 121 may also input an evaluation condition including a condition regarding "variation in the predicted values of at least a part of a plurality of items" into the condition search unit 216 . For example, the evaluation conditions include the method for deriving the evaluation score of the prediction data and the allowable level of the evaluation score.

As an example, the evaluation condition input unit 121 inputs an evaluation condition for evaluating the predicted value of the line width at plural locations on the surface Wa (hereinafter referred to as "line width predicted value") to the condition search unit 216 . The evaluation condition, as an example of the derivation method of the above-mentioned evaluation score, includes a different calculation formula (for example, a standard deviation calculation formula) for the line width prediction value of at least a part (for example, all locations) of a plurality of locations. This evaluation condition, as the allowable level of the above-mentioned evaluation score, includes a varying allowable upper limit value calculated by the above-mentioned equation.

The evaluation condition input unit 121 may input evaluation conditions for evaluating the predicted values of the film thickness at plural locations on the surface Wa to the condition search unit 216. This evaluation condition is used as the derivation method of the above evaluation score. An example of includes a different calculation formula (such as a standard deviation calculation formula) for the predicted value of film thickness at at least a part (for example, the whole place) of a plurality of locations. This evaluation condition, as the allowable level of the above-mentioned evaluation score, includes a varying allowable upper limit value calculated by the above-mentioned equation.

The search result acquisition unit 122 acquires the recommended processing conditions derived by the condition search unit 216 and saves them in the processing condition storage unit 111 . As will be described later, the recommended processing conditions are derived based on the plurality of data sets, the learning model, and the evaluation conditions input by the evaluation condition input unit 121 .

Here, the process control unit 112 can cause the processing unit to further perform substrate processing in compliance with the recommended processing conditions. The data acquisition unit 113 can further acquire additional performance data on the quality of substrate processing that follows recommended processing conditions. The data input unit 114 can further input the additional data set including the recommendation processing conditions and additional performance data to the model generation unit 214. The recommendation condition derivation unit 115 can update the recommendation processing conditions based on the learning model updated by the model generation unit 214 based on the additional data set. The update of the learning model refers to generating a new learning model based on the plurality of data sets added to the additional data set. The update of the recommendation processing conditions means deriving new recommendation processing conditions based on the learning model updated by the model generation unit 214 .

In this case, the control device 100 may further include a condition evaluation unit 116 and a duplication management unit 117. The condition evaluation unit 116 evaluates whether the recommended processing conditions can be adopted. The repetition management unit 117 repeats at least the following steps until the evaluation result of the condition evaluation unit 116 becomes applicable.

i) The processing control unit 112 causes the processing unit to further perform substrate processing in compliance with recommended processing conditions.

ii) The data acquisition unit 113 further acquires additional performance data.

iii) The data input unit 114 further inputs the additional data set to the model generation unit 214.

iv) The recommendation condition derivation unit 115 updates the recommendation processing conditions based on the learning model updated by the model generation unit 214 based on the additional data set.

The evaluation method of the recommended processing conditions by the condition evaluation unit 116 is not particularly limited. For example, the condition evaluation unit 116 evaluates whether the recommended processing condition can be adopted based on the evaluation result of the above-mentioned additional performance data based on the predetermined evaluation conditions. This evaluation condition may be the same as the evaluation condition of the above-mentioned prediction data. For example, the evaluation conditions include a method of deriving evaluation scores for additional performance data and an allowable level of evaluation scores.

As an example, the condition evaluation unit 116 evaluates the actual line width performance values at a plurality of locations on the surface Wa based on predetermined evaluation conditions. The evaluation conditions, as an example of the derivation method of the above-mentioned evaluation score, include a calculation formula (for example, a standard deviation calculation formula) of different line width performance values at at least a part of a plurality of locations (for example, all locations). This evaluation condition, as the allowable level of the above-mentioned evaluation score, includes a varying allowable upper limit value calculated by the above-mentioned equation.

The evaluation condition input unit 121 may also evaluate the film thickness performance values at a plurality of locations on the surface Wa based on predetermined evaluation conditions. The evaluation conditions include, as an example of the derivation method of the above-mentioned evaluation score, a formula (for example, a formula for a standard deviation) of varying film thickness performance values at at least a part of a plurality of locations (for example, all locations). This evaluation condition, as the allowable level of the above-mentioned evaluation score, includes a varying allowable upper limit value calculated by the above-mentioned equation.

The condition evaluation unit 116 may evaluate whether the latest recommendation processing condition can be adopted based on whether the difference between the latest recommendation processing condition and the past recommendation processing condition (for example, the previous recommendation processing condition) is an acceptable level. It is assumed that the recommended processing conditions gradually converge to one condition by the repetitive processing performed by the repetitive management unit 117 . borrow By narrowing the difference between the latest recommended processing conditions and the past recommended processing conditions to an allowable level, recommended processing conditions that are close to the convergence result can be adopted.

The condition evaluation unit 116 may also evaluate whether the latest recommendation processing condition can be adopted based on whether the difference between the latest additional performance data and the past additional performance data is an allowable level. The condition evaluation unit 116 may also evaluate whether the latest recommendation processing conditions can be adopted based on whether the difference between the evaluation score of the latest additional performance data and the evaluation score of the past additional performance data is an allowable level.

The control device 100 may further include a performance data correction unit 118 . The performance data correction unit 118 , before the data input unit 114 inputs the data set to the model generation unit 214 , eliminates components from the performance data of the data set that are derived from factors different from the substrate processing of the processing unit of the coating and developing device 2 . For example, the performance data correction unit 118 eliminates uneven components derived from the exposure process from the line width performance values at the plurality of locations. Specifically, the performance data correction unit 118 excludes the uneven pattern unique to the exposure process that has been investigated in advance from the line width performance values at multiple locations.

The machine learning device 200 includes, as functional modules, a search operation unit 211, a data acquisition unit 212, a data storage unit 213, a model generation unit 214, a model storage unit 215, and a condition search unit 216. The search calculation unit 211 is a machine learning engine of the machine learning device 200 . For example, the search operation unit 211 searches for a solution using a genetic algorithm based on preset learning conditions. The learning conditions include the first-generation individuals, the derivation method of the individual's evaluation score, and the allowable level of the evaluation score.

The search calculation unit 211 obtains a plurality of individuals of the first generation and calculates the evaluation score of each individual. Then, the search operation unit 211 eliminates individuals whose evaluation scores are far from the allowable level, and uses operations such as crossover, inversion, and mutation to evolve the plural individuals into the plural individuals of the next generation. That Then, the search calculation unit 211 derives individuals whose evaluation scores are at the allowable level by repeating the derivation of the individual's evaluation score, the elimination of individuals, and the evolution of individuals.

The data acquisition unit 212 acquires the above-mentioned data set and the additional data set from the data input unit 114. The data storage unit 213 stores the data set acquired by the data acquisition unit 212 as a database for learning.

The model generation unit 214 generates the above-mentioned learning model through machine learning based on the plurality of data sets stored in the data storage unit 213 . The model generation unit 214 may also generate a learning model through machine learning including an operation process of searching for the above-mentioned learning model using a genetic program. For example, the model generation unit 214 generates a learning model including a complex model expression that outputs predicted values of a plurality of items respectively in response to input of processing conditions. In generating each model expression, the model generation unit 214 sets the above-mentioned learning conditions for deriving the model expression, and requests the search calculation unit 211 to derive the model expression that follows the learning conditions.

For example, the model generation unit 214 generates a plurality of temporary model expressions that generate predicted values in response to the input of processing conditions, so that they are the plurality of individuals of the first generation. The temporary model expression uses various operators and random numerical values as elements, and expresses the calculation expression in a tree structure. The model generation unit 214 uses the mean difference score, which represents the mean difference between the predicted value and the actual performance value based on the temporary model expression, as the evaluation score of the above-mentioned learning conditions, and defines its derivation method. For example, the model generation unit 214 defines a derivation method including at least the following sequence.

a1) Input the processing conditions of the complex group data set into the temporary model formula and derive the complex predicted value.

a2) Derive the mean difference score that represents the mean difference between the complex predicted value and the actual performance value of the complex group data set.

The mean difference score can be any value if it can represent the mean difference between the complex predicted value and the actual performance value of the complex data set. A specific example of the deviation score from the mean is the square of the difference between the predicted value and the actual performance value. and, or the square root of the sum of squares, etc. The model generation unit 214 sets the upper limit value set in advance for the mean difference score to the allowable level of the evaluation score in the above-mentioned learning conditions.

The search operation unit 211 derives a model expression in which the mean difference score is equal to or less than the upper limit by repeating the derivation of the mean difference score of the temporary model expression, the elimination of the temporary model expression, and the evolution of the temporary model expression. The model generation unit 214 obtains the model expression derived by the search calculation unit 211 and saves it in the model storage unit 215 . Through the above procedure, the model generation unit 214 stores each model expression in the model storage unit 215, thereby causing the model storage unit 215 to generate a learning model including the plural model expressions.

The condition search unit 216 derives recommendation processing conditions based on the plurality of data sets recorded in the data storage unit 213 , the learning model recorded in the model storage unit 215 , and the evaluation conditions input by the evaluation condition input unit 121 . The condition search unit 216 may also derive the recommended processing conditions through a search process including an operation process of searching for recommended processing conditions using a genetic algorithm. For example, the condition search unit 216 sets the above-mentioned learning conditions for deriving the recommended processing conditions, and requests the search calculation unit 211 to derive the recommended processing conditions that comply with the learning conditions.

For example, the condition search unit 216 sets the processing condition of the plurality of data sets recorded by the data storage unit 213 to be the plurality of individuals of the first generation. Each processing condition represents conditions for multiple items in a tree structure.

The condition search unit 216 defines the method for deriving the evaluation scores of the learning items so as to include at least the following order.

b1) Input the processing conditions of the complex data set into the learning model recorded in the model storage unit 215, and derive the prediction data.

b2) Derive the evaluation score of the prediction data according to the derivation method of the evaluation conditions input by the evaluation condition input unit 121 .

The condition search unit 216 makes the allowable level of the evaluation condition input by the evaluation condition input unit 121 be the allowable level of the evaluation score of the above-mentioned learning condition.

The search operation unit 211 derives recommended processing conditions with evaluation scores as allowable levels by repeatedly deriving evaluation scores of processing conditions, eliminating processing conditions, and evolving processing conditions. The condition search unit 216 acquires the recommended processing conditions derived by the search operation unit 211 and outputs them to the search result acquisition unit 122 .

FIG. 7 is a block diagram illustrating the hardware configuration of the control device 100 and the machine learning device 200. The control device 100 includes a circuit 190 . The circuit 190 includes at least one processor 191, memory 192, storage 193, display device 194, input device 195, input/output port 196, and communication port 197. The storage 193 is a non-volatile recording medium (such as flash memory) that can be read by a computer. For example, the memory 193 records a program for causing the control device 100 to perform the following steps: causing the coating and developing device 2 to perform substrate processing according to preset processing conditions; and obtaining data on the substrate processing according to the processing conditions from the quality inspection device 70 . Quality performance data; input the data set including the processing conditions of the substrate processing and the performance data of the substrate processing into the machine learning device 200; derive the substrate based on the above learning model generated by the machine learning device 200 based on the plurality of sets of data sets Recommended handling conditions for handling. For example, the memory 193 includes a recording area for recording programs constituting the above-mentioned functional modules and a recording area allocated to the processing condition storage unit 111 .

The display device 194 is used for displaying recommended processing conditions and the like. The display device 194 and the input device 195 function as a user interface of the control device 100 . The display device 194 includes, for example, a liquid crystal display Device, etc., used to display information to users. The input device 195, such as a keyboard, obtains the user's input information. The display device 194 and the input device 195 may be integrated into a so-called touch panel. The input device 195 is used for inputting processing conditions and evaluation conditions, etc.

The memory 192 temporarily records the program loaded from the storage 193, the operation results of the processor 191, etc. The processor 191 cooperates with the memory 192 to execute the above program, thereby controlling the coating and developing device 2 . The input/output port 196 performs input/output of electrical signals with the display device 194 and the input device 195 in response to instructions from the processor 191 . The communication port 197 implements network communication with the machine learning device 200 in response to instructions from the processor 191 .

The machine learning device 200 includes a circuit 290. The circuit 290 includes a processor 291, a memory 292, a storage 293, and a communication port 294. The storage 293 is a non-volatile recording medium (such as flash memory) that can be read by a computer. For example, the memory 293 records a program for causing the machine learning device 200 to execute the following steps: obtain the above-mentioned data set; and generate the above-mentioned learning model through machine learning based on a plurality of sets of data sets. For example, the memory 293 includes a recording area for recording programs constituting the above-mentioned functional modules, and a recording area allocated to the data storage unit 213 and the model storage unit 215 .

The memory 292 temporarily records the program loaded from the storage 293, the operation results of the processor 291, etc. The processor 291 cooperates with the memory 292 to execute the above program, thereby generating the above learning model. The communication port 294 implements network communication with the control device 100 in response to instructions from the processor 291 .

[Condition setting support sequence]

Next, as an example of the condition setting support method, the condition setting support procedure executed by the control device 100 and the machine learning device 200 will be described. The condition setting support program executed by the control device 100 includes a deriving sequence of recommended processing conditions and a refreshing sequence of recommended processing conditions. The condition setting support sequence executed by the machine learning device 200 includes a learning model generation sequence and a search sequence for recommended processing conditions. Below, specific examples of each sequence are given.

(Recommended export order of processing conditions)

The derivation sequence of the recommended processing conditions performed by the control device 100 includes the following steps: following the preset processing conditions, causing the coating and developing device 2 to perform substrate processing, which substrate processing includes supply of processing liquid to the wafer W; obtaining information on compliance The actual performance data of the quality of the substrate processing according to the processing conditions; a data set including the processing conditions of the substrate processing and the actual performance data of the substrate processing is input to the machine learning device 200; the machine learning device 200 generates based on a plurality of sets of data sets The above learning model derives recommended processing conditions. The derivation of recommended processing conditions may include: inputting the evaluation conditions of the prediction data into the machine learning device 200; and obtaining the recommended processing conditions derived by the machine learning device 200 based on the plurality of data sets, learning models, and evaluation conditions.

As illustrated in FIG. 8 , the control device 100 first executes steps S01, S02, and S03. In step S01 , the process control unit 112 causes the coating and developing device 2 to start substrate processing in accordance with the processing conditions recorded in the processing condition storage unit 111 . In step S02, the data acquisition unit 113 acquires the actual performance value of the item being processed. The data acquisition unit 113 can also acquire actual performance values of a plurality of items being processed. For example, the data acquisition unit 113 acquires actual performance values of the presence or absence of liquid splashing of the developer, defective liquid film formation, and liquid dripping based on the information detected by the in-process inspection unit 90 . The data acquisition unit 113 may also acquire the presence or absence of liquid splashing of the film-forming liquid, formation failure of the liquid film, and liquid droplets based on the information detected by the in-process inspection unit 90 The actual performance value that fell. In step S03, the process control unit 112 confirms whether the substrate processing complying with the processing conditions has been completed.

In step S03, it is determined that the substrate processing has not been completed, and the control device 100 returns the process to step S02. Thereafter, the acquisition of actual performance values of the items being processed is continued until the substrate processing is completed. In step S03, it is determined that the substrate processing is completed, and the control device 100 executes step S04. In step S04, the data input unit 114 confirms whether a defect has occurred in the supply state of the processing liquid based on the actual performance value of the item being processed.

In step S04, it is determined that the supply state of the processing liquid is not defective, and the control device 100 executes steps S05, S06, and S07. In step S05, the data acquisition unit 113 acquires the actual performance value of the above-mentioned processed items. The data acquisition unit 113 can also acquire the actual performance values of multiple processed items. For example, the data acquisition unit 113 acquires the above-mentioned actual line width performance values at a plurality of locations on the surface Wa based on the information detected by the post-processing inspection unit 80 . The data acquisition unit 113 can acquire the above-mentioned actual film thickness performance values at a plurality of locations on the surface Wa based on the information detected by the post-process inspection unit 80 . In step S06, the performance data correction unit 118 excludes components originating from factors different from the substrate processing from the performance values of the plurality of post-process items. In step S07, the data input unit 114 inputs a data set including processing conditions and actual performance data (actual performance values of plural-processed items) corresponding to the processing conditions to the machine learning device 200.

Next, the control device 100 executes step S08. In step S04, it is determined that the supply state of the processing liquid is defective, and the control device 100 executes step S08 without executing steps S05, S06, and S07. In step S08, the data input unit 114 confirms whether the input of the number of data sets required for machine learning in the machine learning device 200 has been completed.

In step S08, it is determined that the input of the number of data sets required for machine learning has not been completed, and the control device 100 executes step S09. In step S09, the processing control unit 112 changes the processing conditions. For example, the processing control unit 112 changes the processing conditions based on the user's input to the input device 195 or the like. Then, the control device 100 returns the process to step S01. Thereafter, changing the processing conditions, executing the substrate processing, and inputting the data set are repeated until the input of the number of data sets required for machine learning is completed.

In step S08, it is determined that the input of the number of data sets required for machine learning has been completed, and the control device 100 executes steps S11, S12, S13, and S14. In step S11, the evaluation condition input unit 121 waits for a learning completion notification from the machine learning device 200. In step S12, the evaluation condition input unit 121 sets evaluation conditions for the prediction data. For example, the evaluation condition input unit 121 sets the evaluation conditions for the prediction data based on the user's input to the input device 195 or the like. In step S13, the evaluation condition input unit 121 inputs the evaluation condition set in step S12 to the machine learning device 200. In step S14 , the search result acquisition unit 122 acquires the recommended processing conditions derived by the machine learning device 200 based on the plurality of data sets, learning models, and evaluation conditions input by the evaluation condition input unit 121 , and saves them to the processing module 11 . Through the above steps, the export sequence of recommended processing conditions is completed.

(Recommended refresh order for processing conditions)

The refresh sequence of the recommended processing conditions performed by the control device 100 includes the following steps: causing the coating and developing device 2 to further perform substrate processing according to the recommended processing conditions; further obtaining additional performance data on the quality of substrate processing following the recommended processing conditions; The additional data set including the recommended processing conditions and the additional performance data is further input to the machine learning device 200; and the recommended processing conditions are updated based on the learning model updated by the machine learning device 200 based on the additional data set. This refresh sequence may further include the steps of evaluating the recommended processing conditions; this refresh sequence may further include the evaluation results of the recommended processing conditions. Until the predetermined level is reached, the following steps are repeated: making the coating and developing device 2 further perform substrate processing according to the recommended processing conditions; further acquiring additional performance data; further inputting the additional data set to the machine learning device 200; and based on the machine learning device 200. Add the updated learning model to the data set and update the recommended processing conditions.

As illustrated in FIG. 9 , the control device 100 first executes steps S21, S22, S23, S24, and S25. In step S21 , the process control unit 112 causes the coating and developing device 2 to perform substrate processing in compliance with the recommended processing conditions recorded in the processing condition storage unit 111 . In step S22, the data acquisition unit 113 acquires the additional performance value of the above-mentioned processed item. The data acquisition unit 113 can also acquire the additional performance values of the plurality of processed items. In step S23, the performance data correction unit 118 excludes components originating from factors different from the substrate processing from the additional performance values of the plurality of post-process items. In step S24, the condition evaluation unit 116 evaluates the recommendation processing conditions. In step S25, the repetition management unit 117 confirms whether the recommended processing condition can be adopted based on the evaluation result in step S24.

In step S25, if it is determined that the recommended processing conditions cannot be adopted, the control device 100 executes steps S26, S27, and S28. In step S26, the data input unit 114 inputs an additional data set including processing conditions and additional performance data (additional performance values of plural processed items) corresponding to the processing conditions to the machine learning device 200. In step S27 , the search result acquisition unit 122 waits for an update completion notification of the learning model from the machine learning device 200 . In step S28 , the search result acquisition unit 122 acquires the recommended processing conditions updated by the machine learning device 200 based on the additional data set and saves them to the processing module 11 . Then, the control device 100 returns the process to step S21. Thereafter, the acquisition of additional performance data and the updating of the recommended processing conditions are repeated until the recommended processing conditions become applicable.

In step S25, it is determined that the recommended processing conditions can be adopted, and the control device 100 completes the processing. Through the above steps, the refresh sequence of recommended processing conditions is completed.

(Production order of learning models)

The sequence of generating a learning model performed by the machine learning device 200 includes the following steps: obtaining the above-mentioned data set; and generating a learning model through machine learning based on a plurality of sets of data sets. The step of generating a learning model through machine learning may include a computational process of searching for a learning model through genetic programming. It is possible to generate a learning model including a complex model formula that outputs predicted values of complex items respectively in response to the input of processing conditions.

As illustrated in FIG. 10 , the machine learning device 200 first executes steps S31, S32, and S33. In step S31, the data acquisition unit 212 waits for the input of the data set from the data input unit 114. In step S32, the data acquisition unit 212 stores the input data set into the data storage unit 213. In step S33, the data storage unit 213 confirms whether the number of data sets stored in the data storage unit 213 has reached the number required for machine learning.

In step S33, it is determined that the number of stored data sets has not reached the number required for machine learning, and the control device 100 returns the process to step S31. Thereafter, the acquisition of the data set is repeated until the number of data sets required for machine learning is stored.

In step S33, it is determined that the number of stored data sets reaches the number required for machine learning, and the control device 100 executes steps S34, S35, and S36. In step S34, the model generation unit 214 sets the above-described learning conditions for deriving a model expression corresponding to any predicted value, and requests the search operation unit 211 to derive a model expression that conforms to the learning conditions. For example, the model generation unit 214 generates a model based on input of processing conditions. Generate a plurality of temporary model expressions of predicted values, so that they are equal to the plurality of individuals of the first generation. In addition, the model generation unit 214 uses the above-mentioned mean difference score as an evaluation score, and defines its derivation method so that the upper limit of the above-mentioned mean difference score becomes the allowable level of the evaluation score. In step S35, the search calculation unit 211 calculates the mean difference score of each temporary model expression according to the above learning conditions. In step S36, the search calculation unit 211 follows the above-mentioned learning conditions and confirms whether there is a temporary model expression in which the deviation score from the mean is less than the above-mentioned upper limit value.

In step S36, when it is determined that there is no temporary model expression in which the deviation score from the mean is equal to or less than the upper limit value, the machine learning device 200 executes step S37. In step S37, the search operation unit 211 eliminates temporary model expressions whose deviation from the mean exceeds the upper limit, and evolves a plurality of temporary model expressions into the next generation through operations such as crossover, inversion, and mutation. A plurality of temporary model equations. Then, the machine learning device 200 returns the process to step S35. Thereafter, the derivation of the deviation score from the mean of the temporary model expression, the elimination of the temporary model expression, and the evolution of the temporary model expression are repeated until the derivation of the deviation score from the mean becomes a temporary model expression below the upper limit.

In step S36, it is determined that there is a temporary model expression in which the deviation score from the mean is equal to or less than the upper limit value, and the machine learning device 200 executes steps S38 and S39. In step S38, the search operation unit 211 selects the temporary model equation with the best (minimum) mean difference score, and saves it in the model storage unit 215 as a model equation of the learning model. In step S39, the model generation unit 214 confirms whether the derivation of all model expressions used to constitute the learning model (that is, all model expressions required to derive the predicted values of plural items) has been completed.

In step S39, it is determined that the derivation of all model expressions has not been completed, and the machine learning device 200 executes step S41. In step S41, the model generation unit 214 changes the model expression of the export object. In other words, the model generation unit 214 changes the items to be predicted in the model expression. Then, the machine learning device 200 makes The process returns to step S34. Thereafter, the setting of learning conditions and the derivation of model expressions based on the settings are repeated until the derivation of all model expressions is completed.

In step S39, it is determined that the derivation of all model expressions is completed, and the machine learning device 200 completes generation of the learning model. Through the above steps, the sequence of generating the learning model is completed.

(Search order for recommended processing conditions)

The search sequence for recommended processing conditions performed by the machine learning device 200 includes the following steps: deriving recommended processing conditions for substrate processing based on evaluation conditions of a plurality of data sets, learning models, and prediction data. The derivation of recommended processing conditions may include an operation process of searching for recommended processing conditions through a genetic algorithm. Recommended processing conditions can also be derived based on complex number group data sets, complex number model expressions, and evaluation conditions for evaluating predicted values of complex number items. For example, the recommendation processing conditions can be derived based on the evaluation conditions including the condition "variation in the predicted values of multiple items".

As illustrated in FIG. 11 , the machine learning device 200 first executes steps S51 and S52. In step S51, the condition search unit 216 waits for input of evaluation conditions from the evaluation condition input unit 121. In step S52, the condition search unit 216 sets the above-mentioned learning conditions for deriving the recommended processing conditions, and requests the search calculation unit 211 to derive the recommended processing conditions that comply with the learning conditions. For example, the condition search unit 216 makes the processing conditions of the plurality of data sets recorded by the data storage unit 213 be plural individuals of the first generation. In addition, the condition search unit 216 defines the evaluation score derivation method and the allowable level of the evaluation score based on the evaluation conditions input by the evaluation condition input unit 121 .

Next, the machine learning device 200 executes steps S53, S54, and S55. In step S53, the search operation unit 211 inputs each processing condition into the learning model recorded in the model storage unit 215, and derives prediction data. In step S54, the search calculation unit 211 derives the evaluation score of the prediction data. In step S55, the search calculation unit 211 confirms whether there is a processing condition in which the evaluation score is an allowable level.

In step S55, when it is determined that the processing condition for the evaluation score to be the allowable level does not exist, the machine learning device 200 executes step S56. In step S56, the search operation unit 211 eliminates the processing conditions whose evaluation scores are far from the allowable level, and evolves the complex processing conditions into the next generation's complex processing conditions through operations such as crossover, inversion, and mutation. Then, the machine learning device 200 returns the process to step S53. Thereafter, the derivation of the evaluation score of the processing condition, the elimination of the processing condition, and the evolution of the processing condition are repeated until the derived evaluation score becomes the processing condition of the allowable level.

In step S55, when it is determined that the processing condition for the evaluation score to be an allowable level exists, the machine learning device 200 executes steps S57 and S58. In step S57, the search calculation unit 211 sets the processing condition in which the evaluation score is the optimal value as the recommended processing condition. In step S58 , the condition search unit 216 obtains the recommended processing conditions derived by the search operation unit 211 and outputs them to the search result acquisition unit 122 . Through the above steps, the search sequence of recommended processing conditions is completed.

The derivation of recommended processing conditions is not limited to the operation process of searching for recommended processing conditions through the above genetic algorithm. For example, in step S55, the recommended processing conditions may also be derived by repeating the operation process of changing the processing conditions and deriving the evaluation score until the evaluation score reaches the allowable level.

[Specific example]

As an example, the setting support procedure of the processing conditions of the development process in the development unit U3 is specifically illustrated. The processing conditions of the development process in the development unit U3 include, for example: the rotation speed of the wafer W, the supply amount of the developer, the supply time of the developer, the supply amount of the rinse liquid, the ejection time of the rinse liquid, and the ejection time of the rinse liquid. The drying time, the movement start position of the nozzle 31, the movement speed of the nozzle 31, the movement end position of the nozzle 31, etc. Among them, items necessary for the recommended processing conditions include, for example, the rotation speed of the wafer W during supply of the developer solution and the moving speed of the nozzle 31 . In this case, in the above-mentioned steps S01 to S09, the following steps are repeated: changing the rotation speed of the wafer W and the moving speed of the nozzle 31, and inputting the data set to the machine learning device 200.

For example, in steps S01 to S09, while the rotation speed of the wafer W is 200 rpm, the moving speed of the nozzle 31 is set to 15 mm/s, 20 mm/s, and 25 mm/s; then, the rotation speed of the wafer W is set to 200 rpm. In the state of 250 rpm, the moving speed of the nozzle 31 was set to 15 mm/s, 20 mm/s, and 25 mm/s; then, in the state of the rotation speed of the wafer W being 300 rpm, the moving speed of the nozzle 31 was set to 15 mm/s. s, 20mm/s, 25mm/s. In step S04 executed under any one of these processing conditions, it is determined that the supply state of the processing liquid is defective, and the data set corresponding to the processing condition is excluded from being input to the machine learning device 200 . In this case, in order to obtain the number of data sets required for machine learning, the processing conditions are further changed in step S09. For example, under the processing conditions of a rotation speed of 300 rpm and a movement speed of 25 mm/s, it is determined that splashing of the developer solution occurs, the rotation speed is changed to 290 rpm, and the rotation speed is again 290 rpm and the movement speed is 25 mm/s. performance data under the conditions.

In step S05, for example, the average value of the line widths of the divided regions of the wafer W divided into n places is obtained as n line width actual performance values. The data set for this situation is illustrated below.

Processing conditions: rotation speed of wafer W = 200 rpm, moving speed of nozzle = 15 mm/s

Actual performance data: W1=23nm, W2=28nm, W3=31nm,. ． ． Wn=24nm (Wi: average line width of divided area i)

The learning model generated in the machine learning device 200 based on this data set, for example, outputs a predicted value of the average line width in the divided area at n in response to inputs of the rotation speed of the wafer W and the moving speed of the nozzle. In step S12, as the calculation formula of the evaluation score, for example, the calculation formula of the standard deviation of n line width prediction values is set; as the above-mentioned tolerance level, the tolerance value of the standard deviation is set. Based on the evaluation conditions set in this way, in the machine learning device 200, the recommended values of the rotation speed of the wafer W and the moving speed of the nozzle 31 (for example, the rotation speed of the wafer W = 234 rpm, the moving speed of the nozzle 31 = 22 rpm ), derived as the above recommended processing conditions.

[Effects of this embodiment]

As described above, the substrate processing condition setting support method of this embodiment includes the following steps: preparing a data set including the processing conditions of the substrate processing performed by the coating and developing device 2 and the actual performance data on the quality of the substrate processing, Input to the machine learning device 200, the substrate processing includes supply of processing liquid to the wafer W; and according to the model generated by the machine learning device 200 through machine learning based on a plurality of sets of data sets, that is, in response to the input of the processing conditions, A learning model outputs prediction data regarding the quality of substrate processing, and derives recommended processing conditions for substrate processing.

By setting the support method based on this condition, recommended processing conditions are derived based on the learning model generated by machine learning, so appropriate processing conditions can be searched efficiently. Therefore, it is effective in simplifying the work of setting processing conditions for substrate processing.

The substrate processing condition setting support method may include the following steps: making the coating and developing device 2 follow the recommended processing conditions to further perform the substrate processing; further obtaining additional performance data on the quality of the substrate processing following the recommended processing conditions; including the recommended processing conditions. The additional data set and the additional performance data are further input to the machine learning device 200; and the machine learning device 200 based on the additional data Collect the updated learning model and update the recommended processing conditions. In this case, the recommended processing conditions are updated based on feedback from the recommended processing conditions and additional performance data. Therefore, more appropriate processing conditions can be searched for efficiently.

The condition setting support method for substrate processing may further include a step of evaluating recommended processing conditions; in addition, the condition setting supporting method repeats the following steps until the evaluation result of the recommended processing conditions reaches a predetermined level: making the coating and developing device 2 follow the recommended processing further perform substrate processing according to the conditions; further obtain additional performance data; further input the additional data set to the machine learning device 200; and update the recommended processing conditions based on the learning model updated by the machine learning device 200 based on the additional data set. In this case, through repeated processing, more appropriate processing conditions can be searched efficiently.

The step of deriving recommended processing conditions may include the following steps: inputting the evaluation conditions of the prediction data into the machine learning device 200; and obtaining the recommended processing conditions derived by the machine learning device 200 based on the plurality of data sets, learning models, and evaluation conditions. . In this case, the search for recommended processing conditions is also performed by the machine learning device 200, so that appropriate processing conditions can be searched more efficiently.

The condition setting support method for substrate processing obtains actual performance data including actual performance values of a plurality of items; and inputs the data set to the machine learning device 200 that generates a learning model including predicted values of the plurality of items in response to the input of processing conditions. The plural model equations output respectively; and the evaluation conditions for evaluating the predicted values of the plural items are input to the machine learning device 200 . In this case, by expanding the evaluation conditions into plural items, the quality of the processing can be evaluated more appropriately, and more appropriate processing conditions can be searched.

The condition setting support method for substrate processing can input evaluation conditions including a condition regarding "differences in predicted values of at least a part of a plurality of items" to the machine learning device 200 . In this case, a plurality of items can be evaluated efficiently, and therefore more appropriate processing conditions can be searched efficiently.

The condition setting support method for substrate processing can obtain performance data including "performance values of post-processing items and in-process items." The post-processing item represents the quality of the wafer W after substrate processing, and the in-processing item represents the substrate processing in progress. The supply status of the processing liquid; and the data set input to the machine learning device 200 is selected based on the actual performance value of the item being processed. In this case, by directly selecting the exception being processed as the data being processed, the search range for recommended processing conditions based on the processed quality can be narrowed. Therefore, appropriate processing conditions can be searched more efficiently.

The substrate processing condition setting support method may further include the following steps: before inputting the data set to the machine learning device 200, components originating from factors different from the substrate processing are excluded from the actual performance data of the data set. In this case, more appropriate processing conditions can be searched.

The substrate treatment may include a development process: a developing solution is supplied to the surface Wa of the wafer W and a photosensitive coating film is overexposed; a pattern including a pattern formed on the surface Wa of the wafer W by the development process may be obtained. The actual performance data of the actual performance value of the line width. When substrate processing includes development processing, it tends to require a lot of labor in order to derive better processing conditions. Therefore, setting the support method based on the above conditions can efficiently search for appropriate processing conditions, and the effectiveness is remarkable.

Substrate processing may include a film forming process: coating a film forming liquid on the surface Wa of the wafer W to form a coating film; the film thickness may include the film formed on the surface Wa of the wafer W by the film forming process. Performance data of actual performance values. When substrate processing includes film formation processing, the quality of substrate processing also depends on the processing conditions. It is very sensitive, so it tends to require a lot of labor in order to derive better processing conditions. Therefore, setting the support method based on the above conditions can efficiently search for appropriate processing conditions, and the effectiveness is remarkable.

The embodiments have been described above. However, the contents disclosed in the present invention are not necessarily limited to the above-described embodiments, and various changes can be made without departing from the gist of the invention. For example, the substrate to be processed is not limited to a semiconductor wafer, but may also be a glass substrate, a mask substrate, an FPD (Flat Panel Display), etc.

S01~S09, S11~S14: steps

Claims (20)

A condition setting support method for substrate processing, including the following steps: inputting a data set including processing conditions for substrate processing performed by a substrate processing device and actual performance data on the quality of the substrate processing into a machine learning device. The processing includes supplying a processing liquid to the substrate; according to a model generated by the machine learning device through machine learning based on the complex data set, prediction data on the quality of the substrate processing is output based on the input of the processing conditions. The learning model is used to derive the recommended processing conditions for the substrate processing; obtain the performance data including "the performance values of the processed items and the items being processed". The processed items represent the quality of the substrate after the substrate is processed. The processed items The item indicates the supply status of the processing liquid to the substrate during processing of the substrate; and a data set to be input to the machine learning device is selected based on the actual performance value of the item in processing. For example, the condition setting support method for substrate processing in item 1 of the patent application further includes the following steps: making the substrate processing device comply with the recommended processing conditions to further perform the substrate processing; and further obtaining information on the substrate processing that follows the recommended processing conditions. additional performance data of the quality; further input the additional data set including the recommended processing conditions and the additional performance data into the machine learning device; and "the learning model updated by the machine learning device based on the additional data set" , update the recommended processing conditions. For example, the substrate processing condition setting support method in item 2 of the patent application further includes steps for evaluating the recommended processing conditions; until the evaluation results of the recommended processing conditions reach a predetermined level, repeat the following steps: make the substrate processing device comply with further execute the substrate processing according to the recommended processing conditions; further obtain the additional performance data; further input the additional data set into the machine learning device; and "the learning model updated by the machine learning device based on the additional data set" , update the recommended processing conditions. For example, the condition setting support method for substrate processing in any one of the patented scope items 1 to 3, wherein the step of deriving the recommended processing conditions includes the following steps: inputting the evaluation conditions of the predicted data into the machine learning device ; and obtain the recommended processing conditions derived by the machine learning device based on the plurality of data sets, the learning model, and the evaluation conditions. For example, the condition setting support method for substrate processing in Item 4 of the patent application scope, wherein the performance data including the performance values of a plurality of items is obtained; the data set is input to the machine learning device that generates the learning model, and the learning model, It includes a complex model equation that outputs the predicted values of the plurality of items respectively in response to the input of the processing condition; and the evaluation condition for evaluating the predicted value of the plurality of items is input to the machine learning device. For example, the condition setting support method for substrate processing in claim 5 of the patent scope includes inputting the evaluation condition including a condition regarding "variation in the predicted values of at least a part of the plurality of items" into the machine learning device. For example, the condition setting support method for substrate processing in Item 1 or 2 of the patent scope further includes the following steps: before inputting the data set into the machine learning device, the actual performance data from the data set will be derived from The substrate handles different factors of composition exclusion. For example, the condition setting support method for substrate processing in claim 1 or 2, wherein the substrate processing includes development processing: supplying a developing solution to the photosensitive coating film that has been overexposed on the surface of the substrate; obtaining The performance data includes the performance value of the line width of the pattern formed on the surface of the substrate by the development process. For example, the condition setting support method for substrate processing in Item 1 or 2 of the patent application scope, wherein the substrate processing includes film forming processing: coating a film forming liquid on the surface of the substrate to form a coating film; The performance data of the film thickness of the coating film formed on the surface of the substrate by film treatment. A condition setting support method for substrate processing, including the following steps: obtaining a data set including set processing conditions for substrate processing and actual performance data on the quality of the substrate processing following the processing conditions, the substrate processing including the substrate processing conditions. The supply of processing liquid; through machine learning based on a plurality of sets of data, a learning model is generated that outputs prediction data about the quality of the substrate processing based on the input of the processing conditions; the acquisition includes "processed items and processing The performance data of "Achievement Value of Medium Project", the processed item represents the quality of the substrate after the substrate is processed, the processed item represents the supply status of the processing liquid to the substrate during the processing of the substrate; and according to the processing Based on the performance value of the winning project, select the complex data set on which the machine learning is based. For example, the condition setting support method for substrate processing in Item 10 of the patent application further includes the following steps: deriving recommended processing conditions for the substrate processing based on the evaluation conditions of the plurality of data sets, the learning model, and the prediction data. For example, the condition setting support method for substrate processing in the patent application scope 11, wherein the step of generating the learning model through the machine learning includes the calculation process of searching the learning model through genetic programming; the step of deriving the recommended processing conditions , including the operation process of searching for the recommended processing conditions through genetic algorithms. For example, the condition setting support method for substrate processing in Item 11 or 12 of the patent application, wherein the performance data includes the data set containing the performance values of a plurality of items is obtained; the learning model is generated, and the learning model includes the processing conditions corresponding to the processing conditions. Input a complex model equation that outputs the predicted values of the plural items respectively; derive the recommended processing conditions based on the complex group data set, the complex model equation, and the evaluation conditions for evaluating the predicted values of the plural items. For example, the condition setting support method for substrate processing in claim 13 of the patent scope, wherein the recommended processing conditions are derived based on the evaluation conditions including conditions regarding "variations in the predicted values of the plurality of items". A substrate processing system includes: a processing unit that performs substrate processing, and the substrate processing includes supply of a processing liquid to the substrate; The processing control unit causes the processing unit to perform the substrate processing according to the preset processing conditions; the data acquisition unit obtains actual performance data on the quality of the substrate processing following the processing conditions; the data input unit includes the processing conditions and The data set of the actual performance data is input to the model generation unit; and the recommendation condition derivation unit outputs prediction data on the quality of the substrate processing in response to the input of the processing conditions. The learning model generated by machine learning of the data set is assembled to derive the recommended processing conditions for the substrate processing; the data acquisition unit obtains the performance data including "the actual performance values of the processed items and the items being processed", and the processed items Indicates the quality of the substrate after the substrate is processed. The in-process item indicates the supply status of the processing liquid to the substrate during the substrate processing. The data input unit selects and inputs to the model based on the actual performance value of the in-process item. Generate the data set of the department. For example, the substrate processing system in Item 15 of the patent application also includes the model generation unit. For example, if the substrate processing system of the patent scope 15 or 16 is applied for, the recommendation condition derivation unit includes: an evaluation condition input unit, which inputs the evaluation conditions of the predicted data into the condition search unit; and a search result acquisition unit, which obtains The condition search unit derives the recommendation processing condition based on the plurality of data sets, the learning model, and the evaluation condition. For example, the substrate processing system in item 17 of the patent scope includes a search part for this condition. A computer-readable recording medium records a program for causing a device to execute a condition setting support method for substrate processing as claimed in Item 1 of the patent application. A learning model that causes an apparatus to execute a step of outputting prediction data on the quality of a substrate process performed by a substrate processing apparatus in response to an input of processing conditions of the substrate process, by using Processing conditions and machine learning of multiple sets of data sets on the quality of the substrate processing that comply with the processing conditions are generated; the substrate processing includes the supply of processing liquid to the substrate; the performance data includes "processed items" and the actual performance value of the item being processed", the item after processing represents the quality of the substrate after processing of the substrate, and the item in process represents the supply status of the processing liquid to the substrate during the processing of the substrate; the machine learning is based on This plural group data set is selected based on the actual performance value of the item being processed.

Images