Substrate processing method and substrate processing device

Abstract

To provide a substrate processing method and a substrate processing device that make it possible to coat a laminate substrate with an appropriate amount of filler.SOLUTION: A substrate processing method includes the steps of inputting data on a laminated substrate of a first substrate and a second substrate bonded to each other and data on a filler into a learned model constructed by machine learning, outputting coating conditions for the filler from the learned model, and applying the filler in a gap between an edge part of the first substrate and an edge part of the second substrate according to the coating conditions while rotating the laminated substrate.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a substrate processing method and a substrate processing apparatus that prevent cracking and chipping of a laminated substrate manufactured by bonding multiple substrates, and more particularly to a technique for applying a filler to gaps formed between the edge portions of multiple substrates that make up the laminated substrate. [Background technology]

[0002] In recent years, in order to achieve even higher density and higher performance in semiconductor devices, development of three-dimensional packaging technology, which stacks multiple substrates to integrate them three-dimensionally, has progressed. In three-dimensional packaging technology, for example, the device surface of a first substrate on which integrated circuits and electrical wiring are formed is bonded to the device surface of a second substrate on which integrated circuits and electrical wiring are formed. Furthermore, after bonding the first substrate to the second substrate, the second substrate is thinned using a polishing or grinding device. In this way, integrated circuits can be stacked in a direction perpendicular to the device surfaces of the first and second substrates.

[0003] In 3D packaging technology, three or more substrates may be bonded together. For example, after a second substrate is bonded to a first substrate and then bonded to the second substrate, a third substrate may be bonded to the second substrate and then bonded to the third substrate. In this specification, the form of multiple substrates bonded together may be referred to as a "laminated substrate."

[0004] Typically, the edges of a substrate are pre-polished to a rounded or chamfered shape to prevent cracks and chipping. When a second substrate having such a shape is ground, a sharp edge is formed on the second substrate. This sharp edge (hereinafter referred to as a knife edge) is formed by the back surface of the ground second substrate and the outer peripheral surface of the second substrate. Such a knife edge is easily chipped by physical contact, which can damage the laminated substrate itself during transportation. Furthermore, if the bonding between the first and second substrates is insufficient, the second substrate may crack during grinding.

[0005] Therefore, to prevent cracking or chipping at the knife edge, a filler is applied to the edge of the laminated substrate before grinding the second substrate. The filler is applied to the gap between the edge of the first substrate and the edge of the second substrate. The filler supports the knife edge formed after grinding the second substrate, preventing cracking or chipping at the knife edge. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Publication No. 05-304062 Summary of the Invention [Problem to be solved by the invention]

[0007] However, if the amount of filler applied is insufficient, voids (air gaps) may form in the filler applied to the gaps between the laminated substrates. Such voids are likely to cause cracks or chips in the knife edge when the second substrate is ground. On the other hand, if the amount of filler applied is excessive, the filler may fall off from the laminated substrate and contaminate the surrounding environment.

[0008] The thinning process of the laminated substrate, which is carried out after the filler is applied, also presents the following challenges. The thinning process of the laminated substrate is carried out by cutting the second substrate with a cutting tool. However, depending on the hardness and type of the applied filler, the cutting tool may be damaged, or the particles (aggregates) contained in the filler may adhere to the cutting tool, causing it to become clogged.

[0009] Therefore, the present invention provides a substrate processing method and substrate processing apparatus that can apply an appropriate amount of filler to a laminated substrate.Furthermore, the present invention provides a substrate processing method and substrate processing apparatus that can improve throughput while preventing clogging of the cutting tool by optimizing the thinning conditions of the laminated substrate. [Means for solving the problem]

[0010] In one aspect, a substrate processing method is provided, which includes inputting data regarding a laminated substrate formed by bonding a first substrate and a second substrate and data regarding a filler into a trained model constructed by machine learning, outputting application conditions for the filler from the trained model, and applying the filler to the gap between the edge portion of the first substrate and the edge portion of the second substrate according to the application conditions while rotating the laminated substrate.

[0011] In one embodiment, the data regarding the laminated substrate includes the materials constituting the surfaces of the first substrate and the second substrate and the shape and size of the gap, and the data regarding the filler includes the composition of the filler. In one embodiment, the application conditions include at least one of a total application amount of the filler, an application amount of the filler per unit time, a temperature of the filler, and a rotation speed of the laminated substrate. In one aspect, in addition to data regarding the laminated substrate and data regarding the filler, data regarding a curing device for curing the filler is input into the trained model, and the application conditions for the filler and the curing conditions for the filler are output from the trained model, and the substrate processing method further includes curing the applied filler using the curing device in accordance with the curing conditions. In one aspect, the data regarding the curing device includes the type of the curing device and the distance between the curing device and the edge of the laminate substrate. In one embodiment, the curing conditions include an output value of the curing device.

[0012] In one aspect, in addition to data regarding the laminated substrate, data regarding the filler, and data regarding the curing device, data regarding a thinning device for thinning the laminated substrate is input into the trained model, and application conditions for the filler, curing conditions for the filler, and thinning conditions for the laminated substrate are output from the trained model, and the substrate processing method further includes thinning the laminated substrate using the thinning device in accordance with the thinning conditions after the filler has hardened. In one embodiment, the data relating to the thinning device includes the type of cutting tool used in the thinning device and the target cutting amount of the laminate substrate. In one embodiment, the thinning conditions include at least one of a pressing force of the cutting tool against the laminated substrate, a rotation speed of the cutting tool, and a rotation speed of the laminated substrate.

[0013] In one embodiment, the trained model is a model constructed by machine learning using training data including data on multiple laminate substrates, data on multiple fillers, and multiple application conditions as correct labels. In one embodiment, the trained model is a model constructed by machine learning using training data including data on multiple laminate substrates, data on multiple fillers, data on multiple curing devices, and multiple application conditions and multiple curing conditions as correct labels. In one aspect, the trained model is a model constructed by machine learning using training data including data on a plurality of laminate substrates, data on a plurality of fillers, data on a plurality of curing devices, data on a plurality of thinning devices, and a plurality of application conditions, a plurality of curing conditions, and a plurality of thinning conditions as correct labels. In one aspect, the substrate processing method further includes updating the training data used for the machine learning by adding the data regarding the laminated substrate and the data regarding the filler input into the trained model and the application conditions output from the trained model, and performing machine learning using the updated training data to update the trained model. In one embodiment, applying the filler to the gap between the edge portion of the first substrate and the edge portion of the second substrate in accordance with the application conditions while rotating the laminated substrate means applying the filler to the gap between the edge portion of the first substrate and the edge portion of the second substrate in accordance with the application conditions while rotating the laminated substrate held vertically.

[0014] In one aspect, a substrate processing method is provided, which inputs data related to a laminated substrate formed by bonding a first substrate and a second substrate, data related to a filler applied to a gap between an edge portion of the first substrate and an edge portion of the second substrate, and data related to a thinning device for thinning the laminated substrate into a trained model constructed by machine learning, outputs thinning conditions for the laminated substrate from the trained model, and thins the laminated substrate using the thinning device in accordance with the thinning conditions.

[0015] In one aspect, a substrate processing apparatus is provided, comprising: a computing system having a trained model constructed by machine learning; and a filler application module that applies filler to a laminated substrate formed by bonding a first substrate and a second substrate while rotating the laminated substrate, wherein the computing system is configured to input data regarding the laminated substrate and data regarding the filler into the trained model, output filler application conditions from the trained model, give instructions to the filler application module, and cause the filler application module to apply the filler to the gap between the edge portion of the first substrate and the edge portion of the second substrate in accordance with the application conditions.

[0016] In one embodiment, the data regarding the laminated substrate includes the materials constituting the surfaces of the first substrate and the second substrate and the shape and size of the gap, and the data regarding the filler includes the composition of the filler. In one embodiment, the application conditions include at least one of a total application amount of the filler, an application amount of the filler per unit time, a temperature of the filler, and a rotation speed of the laminated substrate. In one aspect, the substrate processing apparatus further includes a curing device that cures the applied filler, and the calculation system is configured to input data regarding the curing device in addition to data regarding the laminated substrate and data regarding the filler into the trained model, output the application conditions of the filler and the curing conditions of the filler from the trained model, and give instructions to the curing device to cure the applied filler by the curing device in accordance with the curing conditions. In one aspect, the data regarding the curing device includes the type of the curing device and the distance between the curing device and the edge of the laminate substrate. In one embodiment, the curing conditions include an output value of the curing device.

[0017] In one aspect, the substrate processing apparatus further includes a thinning device that thins the laminated substrate, and the calculation system is configured to input data regarding the laminated substrate, data regarding the filler, and data regarding the curing device, as well as data regarding the thinning device, into the trained model, output filler application conditions, filler curing conditions, and thinning conditions for the laminated substrate from the trained model, and give instructions to the thinning device to thin the laminated substrate using the thinning device in accordance with the thinning conditions after the filler has hardened. In one embodiment, the data relating to the thinning device includes the type of cutting tool used in the thinning device and the target cutting amount of the laminate substrate. In one embodiment, the thinning conditions include at least one of a pressing force of the cutting tool against the laminated substrate, a rotation speed of the cutting tool, and a rotation speed of the laminated substrate.

[0018] In one embodiment, the trained model is a model constructed by machine learning using training data including data on multiple laminate substrates, data on multiple fillers, and multiple application conditions as correct labels. In one embodiment, the trained model is a model constructed by machine learning using training data including data on multiple laminate substrates, data on multiple fillers, data on multiple curing devices, and multiple application conditions and multiple curing conditions as correct labels. In one aspect, the trained model is a model constructed by machine learning using training data including data on a plurality of laminate substrates, data on a plurality of fillers, data on a plurality of curing devices, data on a plurality of thinning devices, and a plurality of application conditions, a plurality of curing conditions, and a plurality of thinning conditions as correct labels. In one aspect, the computing system is configured to update the training data used for the machine learning by adding the data regarding the laminate substrate and the data regarding the filler input into the trained model and the application conditions output from the trained model, and to perform machine learning using the updated training data to update the trained model. In one embodiment, the filler application module includes a substrate holding section or a substrate holding device that holds the laminated substrate in a vertical position.

[0019] In one aspect, a substrate processing apparatus is provided, comprising a computing system having a trained model constructed by machine learning, and a thinning device that thins a laminated substrate formed by bonding a first substrate and a second substrate, wherein the computing system is configured to input data regarding the laminated substrate, data regarding a filler applied to the gap between the edge portion of the first substrate and the edge portion of the second substrate, and data regarding the thinning device into the trained model constructed by machine learning, output thinning conditions for the laminated substrate from the trained model, and give instructions to the thinning device to thin the laminated substrate using the thinning device in accordance with the thinning conditions.

[0020] In one aspect, a computer-readable recording medium is provided that stores a program for causing a computer to execute the steps of: inputting data regarding a laminated substrate in which a first substrate and a second substrate are bonded, and data regarding a filler, into a trained model constructed by machine learning; outputting application conditions for the filler from the trained model; and giving instructions to a filler application module to cause the filler application module to apply the filler to the gap between the edge portion of the first substrate and the edge portion of the second substrate in accordance with the application conditions. [Effects of the Invention]

[0021] According to the present invention, appropriate application conditions are determined by a trained model, and an appropriate amount of filler is applied to a laminated substrate in accordance with these application conditions. As a result, cracks and chipping at the knife edge of the laminated substrate can be prevented, and contamination of the surrounding environment can be prevented. Furthermore, according to the present invention, appropriate thinning conditions are determined by a trained model, and the laminated substrate is appropriately thinned in accordance with these thinning conditions. As a result, clogging of the cutting tool can be prevented, and the throughput of the thinning process can be improved. [Brief explanation of the drawings]

[0022] [Figure 1] Figure 1(a) is a cross-sectional view showing an example of an edge portion of a laminated substrate to be processed, Figure 1(b) is a cross-sectional view showing an example of an edge portion of a laminated substrate to which a filler has been applied, and Figure 1(c) is a cross-sectional view showing an example of an edge portion of a laminated substrate to which a filler has been applied and then thinned. [Figure 2] 1 is a schematic diagram illustrating an embodiment of a substrate processing apparatus. [Figure 3] FIG. 10 is a top view of one embodiment of a filler application module. [Figure 4] FIG. 4 is a side view of the filler application module shown in FIG. 3. [Figure 5] FIG. 1 is a schematic diagram illustrating an embodiment of a coating device. [Figure 6] FIG. 1 is a conceptual diagram illustrating one embodiment of the operation of a computing system. [Figure 7] 1 is a flowchart illustrating one embodiment of a method for building a trained model using machine learning. [Figure 8] 1 is a flowchart illustrating one embodiment of a method for applying filler to a laminate substrate according to application conditions determined using a trained model. [Figure 9] FIG. 1 is a conceptual diagram illustrating one embodiment of the operation of a computing system. [Figure 10] 1 is a flowchart illustrating one embodiment of a method for building a trained model using machine learning. [Figure 11]A flowchart illustrating one embodiment of a method for applying a filler to a laminate substrate and curing the filler according to application and curing conditions determined using a trained model. [Figure 12] FIG. 10 is a schematic view showing another embodiment of the substrate processing apparatus. [Figure 13] FIG. 1 is a schematic diagram illustrating an embodiment of a thinning device. [Figure 14] FIG. 1 is a conceptual diagram illustrating one embodiment of the operation of a computing system. [Figure 15] 1 is a flowchart illustrating one embodiment of a method for building a trained model using machine learning. [Figure 16] A flowchart illustrating one embodiment of a method for applying a filler to a laminate substrate, curing the filler, and thinning the laminate substrate according to application conditions, curing conditions, and thinning conditions determined using a trained model. [Figure 17] FIG. 1 is a conceptual diagram illustrating one embodiment of the operation of a computing system. [Figure 18] 1 is a flowchart illustrating one embodiment of a method for building a trained model using machine learning. [Figure 19] 1 is a flowchart illustrating one embodiment of a method for thinning a laminated substrate in which a filler has hardened according to thinning conditions determined using a trained model. [Figure 20] 1 is a schematic diagram illustrating an embodiment of a substrate processing apparatus. [Figure 21] 1 is a schematic diagram illustrating an embodiment of a substrate processing apparatus. [Figure 22] 1 is a schematic diagram illustrating an embodiment of a substrate processing apparatus. [Figure 23] FIG. 10 is a top view showing another embodiment of the filler application module. [Figure 24] FIG. 24 is a side view of the filler application module shown in FIG. 23. [Figure 25] FIG. 10 is a side view showing yet another embodiment of the filler application module. [Figure 26] 26 is a view seen from the direction indicated by arrow A in FIG. 25. [Figure 27]FIG. 10 is a side view showing yet another embodiment of the filler application module. [Figure 28] 28 is a view seen from the direction indicated by arrow B in FIG. 27. DETAILED DESCRIPTION OF THE INVENTION

[0023] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Fig. 1(a) is a cross-sectional view showing an example of an edge portion of a laminated substrate to be processed. As shown in Fig. 1(a), the laminated substrate Ws has a structure in which a first substrate W1 and a second substrate W2 are bonded together. The first substrate W1 and the second substrate W2 used in this embodiment are circular.

[0024] The edge portion E1 of the first substrate W1 is the outermost side surface inclined with respect to the bonding surface (e.g., device surface) S1 of the first substrate W1. More specifically, the edge portion E1 of the first substrate W1 has a rounded or chamfered shape. The edge portion E2 of the second substrate W2 is also the outermost side surface inclined with respect to the bonding surface (e.g., device surface) S2 of the second substrate W2. More specifically, the edge portion E2 of the second substrate W2 has a rounded or chamfered shape. The edge portions E1 and E2 are sometimes called bevel portions. A gap G is formed between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2.

[0025] 1(b) is a cross-sectional view showing an example of an edge portion of a laminated substrate Ws to which a filler F has been applied. The filler F is applied to a gap G between an edge portion E1 of a first substrate W1 and an edge portion E2 of a second substrate W2. This gap G is formed around the entire periphery of the laminated substrate Ws and has a roughly triangular cross section. The filler F is applied so as to fill this gap G.

[0026] FIG. 1(c) is a cross-sectional view showing an example of an edge portion of a laminated substrate Ws that has been thinned after being coated with filler F. In the example shown in FIG. 1(c), the laminated substrate Ws is thinned by cutting the outer surface of the second substrate W2 using a thinning device (not shown). As a result of this thinning process, a knife edge portion 2 is formed at the edge portion E2 of the second substrate W2. Because the knife edge portion 2 is held (supported) by the filler F, cracking and chipping of the knife edge portion 2 are prevented.

[0027] FIG. 2 is a schematic diagram showing one embodiment of a substrate processing apparatus 5. The substrate processing apparatus 5 of this embodiment includes a calculation system 10 having a trained model constructed by machine learning, and a filler application module 12 that applies a filler to a laminated substrate formed by bonding a first substrate and a second substrate. Although the laminated substrate is not depicted in FIG. 2, it has the configuration described with reference to FIGS. 1(a) to 1(c). As will be described later, the calculation system 10 is configured not only to calculate the filler application conditions according to an algorithm defined by the trained model, but also to control the operation of the filler application module 12.

[0028] The computing system 10 includes a storage device 10a that stores a program and a trained model, and a computing device 10b that executes calculations according to instructions included in the program. The storage device 10a includes a main storage device such as a random access memory (RAM) and an auxiliary storage device such as a hard disk drive (HDD) or a solid state drive (SSD). Examples of the computing device 10b include a CPU (central processing unit) and a GPU (graphics processing unit). However, the specific configuration of the computing system 10 is not limited to these examples.

[0029] The computing system 10 includes at least one computer. The at least one computer may be one server or multiple servers. The computing system 10 may be an edge server connected to the filler application module 12 via a communication line, or may be a cloud server or fog server connected to the filler application module 12 via a communication network such as the Internet or a local area network.

[0030] The computing system 10 may be multiple computers connected via a communication network such as the Internet or a local area network. For example, the computing system 10 may be a combination of an edge server and a cloud server. The storage device 10a and the computing device 10b may be located in multiple computers installed in different locations. Furthermore, the computing system 10 may include a first computer for constructing a trained model through machine learning and a second computer for calculating application conditions using the trained model. The first computer and the second computer may be located in separate locations.

[0031] Fig. 3 is a top view showing one embodiment of the filler application module 12, and Fig. 4 is a side view of the filler application module 12 shown in Fig. 3. The filler application module 12 is configured to apply filler to the edge portion of the laminated substrate Ws while rotating the laminated substrate Ws. The filler application module 12 includes a substrate holding unit 15 that holds the laminated substrate Ws, an application device 16 that applies filler F to the edge portion of the laminated substrate Ws, and a curing device 20 that hardens the applied filler F.

[0032] The substrate holding unit 15 is a stage that holds the back surface of the laminated substrate Ws by vacuum suction. The filler application module 12 further includes a rotation shaft 22 connected to the center of the substrate holding unit 15 and a rotation mechanism 23 that rotates the substrate holding unit 15 and the rotation shaft 22. The laminated substrate Ws is placed on the substrate holding unit 15 so that the center of the laminated substrate Ws coincides with the axis of the rotation shaft 22. The rotation mechanism 23 includes a motor (not shown). As shown in FIG. 3, the rotation mechanism 23 is configured to rotate the substrate holding unit 15 and the laminated substrate Ws together in the direction indicated by the arrow around the central axis Cr of the laminated substrate Ws.

[0033] The coating device 16 is positioned radially outward of the laminated substrate Ws on the substrate holder 15 and is configured to coat the gap G between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 of the laminated substrate Ws with filler F. FIG. 5 is a schematic diagram showing one embodiment of the coating device 16. The coating device 16 includes a syringe 26 for discharging the filler F, a piston 27 reciprocating within the syringe 26, and a horizontal movement mechanism (not shown) for moving the syringe 26 toward or away from the laminated substrate Ws. This horizontal movement mechanism enables the coating device 16 to adjust the distance between the laminated substrate Ws and the filler discharge port 26a of the coating device 16. In one embodiment, the horizontal movement mechanism may be omitted from the coating device 16. In this case, the distance between the laminated substrate Ws and the filler discharge port 26a is predetermined so that the filler F is appropriately injected into the gap G of the laminated substrate Ws.

[0034] Syringe 26 has a hollow structure and is configured to be filled with filler F. Piston 27 is disposed within syringe 26. Syringe 26 has filler outlet 26a at its tip for discharging filler F. The tip of syringe 26 including filler outlet 26a may be configured to be detachable. An appropriate shape for filler outlet 26a is selected depending on the physical properties (e.g., viscosity) of the filler F to be applied. Filler outlet 26a is disposed so as to face gap G between edge portion E1 of first substrate W1 and edge portion E2 of second substrate W2.

[0035] The coating device 16 is connected to a gas supply source via a gas supply line 28. When gas (e.g., dry air or nitrogen gas) is supplied from the gas supply source to the syringe 26, the piston 27 moves forward within the syringe 26. As the piston 27 moves forward, the filler F in the syringe 26 is discharged from the filler discharge port 26a.

[0036] A pressure adjusting device 29 is disposed in the gas supply line 28. By adjusting the pressure of the gas supplied from the gas supply source to the coating device 16, it is possible to adjust the amount of filler F discharged from the filler discharge port 26a per unit time.

[0037] In one embodiment, the application device 16 may include a screw feeder instead of the syringe 26 and piston 27 combination.

[0038] As shown in Figures 3 and 4, the curing device 20 is located radially outward of the laminated substrate Ws. The curing device 20 is disposed downstream of the applicator 16 in the rotation direction of the laminated substrate Ws, and is configured to cure the filler F applied to the laminated substrate Ws by the applicator 16. The curing device 20 cures the filler F while the laminated substrate Ws is rotating. In this embodiment, the filler F is a thermosetting filler. An example of such a filler is a thermosetting resin.

[0039] The curing device 20 is an air heater configured to blow hot air toward the filler F applied to the laminated substrate Ws. The filler F heated by the hot air is cured by a crosslinking reaction. If the filler F contains a solvent, the solvent is volatilized by heating. The curing device 20 is not limited to an air heater, and may be a lamp heater or other configuration as long as it can heat and cure the filler F.

[0040] In this embodiment, the filler F is a thermosetting filler, but in one embodiment, the filler F may be an ultraviolet-curable filler. In this case, the curing device 20 may be a UV irradiation device that irradiates ultraviolet rays to cure the filler F. If the filler F contains a solvent, the filler F may be heated using an air heater or the like to volatilize the solvent.

[0041] The filler application module 12 is configured to apply the filler F to the laminated substrate Ws in accordance with the application conditions determined by the calculation system 10. The calculation system 10 determines the application conditions using a trained model stored in its storage device 10a as follows. That is, as shown in FIG. 6, the calculation system 10 inputs data related to the laminated substrate Ws and data related to the filler F into the trained model, and outputs the application conditions for the filler F from the trained model. The trained model is created in advance by machine learning using training data.

[0042] The data regarding the laminated substrate Ws includes the materials constituting the surfaces of the first substrate W1 and the second substrate W2, and the shape and size of the gap G between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2. In the following description, the gap G between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 may be referred to as the gap G formed in the edge portion of the laminated substrate Ws, or simply as the gap G of the laminated substrate Ws. The shape and size of the gap G in the laminated substrate Ws affect the amount of filler F to be applied.

[0043] Examples of materials that make up the surfaces of the first substrate W1 and the second substrate W2 include silicon, an insulating film, a metal film, or a combination of these. The materials that make up the surfaces of the first substrate W1 and the second substrate W2 affect the ease with which the filler F penetrates into the gap G. Data about the laminated substrate Ws can be obtained from information about the manufacturing process of the laminated substrate Ws, information about previous processes, etc.

[0044] The shape and size of the gap G in the laminated substrate Ws are identified by an image of the gap G or the dimensions of the gap G. The dimensions of the gap G include, for example, the radial width of the gap G, the height of the gap G, and the angle of the inner edge of the gap G (e.g., the angle between the contact points of the edge E1 of the first substrate W1 and the edge E2 of the second substrate W2). The shape and size of the gap G can be measured using a known surface profile measuring device such as a laser scanning device or a confocal microscope. In one embodiment, the substrate processing apparatus 5 may be equipped with a surface profile measuring device for measuring the shape and size (dimensions) of the gap G formed at the edge of the laminated substrate Ws. The shape and size of the gap G are measured before the filler F is applied to the laminated substrate Ws. An image showing the shape and size of the gap G may be used instead of the dimensions of the gap G. In one embodiment, the substrate processing apparatus 5 may be equipped with an imaging device for generating an image showing the shape and size of the gap G.

[0045] Data on Filler F includes its composition. Filler F includes binders, solvents, particles, etc. Particles are dispersed in a binder dissolved in a solvent. For example, the composition of Filler F includes the type of binder, the amount of solvent, the amount of particles, and the particle size. Examples of binders include inorganic binders containing alkali metal silicates, organic binders composed of silicone resins or epoxy resins, and inorganic-organic hybrid binders. Particles are, for example, silica or alumina particles. Particles are mixed into the binder to increase the volume of Filler F and to adjust the viscosity of Filler F. Particles may not be included in Filler F to reduce its viscosity.

[0046] Typically, filler F has a certain degree of viscosity. The viscosity of filler F largely depends on the amount of solvent, the amount of particles, and the binder material. If filler F contains a solvent, the solvent is evaporated by heating. Therefore, when filler F is cured after application, the volume of filler F is likely to decrease due to the evaporation of the solvent. Therefore, to determine the appropriate amount of filler F to fill the gap G formed at the edge portion of the laminated substrate Ws, the viscosity of filler F, i.e., the composition of filler F (e.g., binder material, amount of solvent, amount of particles), is required. The amount of particles also includes the case where the amount of particles is 0, i.e., no particles are included.

[0047] Data relating to the laminated substrate Ws and data relating to the filler F are input to the calculation system 10 and stored in its storage device 10a.

[0048] The application conditions output from the trained model include at least one of the total application amount of filler F, the application amount of filler F per unit time, the temperature of filler F, and the rotation speed of the laminated substrate Ws. The total application amount of filler F is the amount of filler F appropriate for filling the gaps G in the laminated substrate Ws. The application amount of filler F per unit time is, in other words, the application speed, which is the amount of filler F discharged per unit time from the application device 16 shown in FIGS. 3 to 5. The application amount of filler F per unit time can be controlled by the extrusion force of the application device 16 for filler F. The rotation speed of the laminated substrate Ws corresponds to the rotation speed of the substrate holding unit 15 shown in FIGS. 3 and 4 and can be controlled by the rotation mechanism 23.

[0049] The calculation system 10 issues a command to the filler application module 12, causing the filler application module 12 to apply the filler F to the gap G between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 in accordance with the above application conditions.

[0050] The storage device 10a stores a program for constructing a trained model according to a machine learning algorithm. The processing device 10b executes machine learning using training data in accordance with instructions included in the program to construct the trained model. Examples of machine learning algorithms include support vector regression (SVR), partial least squares (PLS), deep learning, random forests, and decision trees. In one example, the trained model is composed of a neural network constructed by deep learning. Constructing the trained model by machine learning includes optimizing parameters such as the weights of the neural network.

[0051] The training data used for machine learning includes data on multiple laminate substrates, data on multiple fillers, and multiple application conditions as correct labels. The multiple laminate substrates are multiple laminate substrates actually used to apply fillers. In the following description, these laminate substrates are referred to as training laminate substrates. The multiple training laminate substrates include multiple laminate substrates with different configurations. The different configurations refer to configurations where at least one of the size of the gap at the edge, the shape of the gap at the edge, and the surface constituent material is different. For example, the training laminate substrates include multiple training laminate substrates with different shapes and sizes (dimensions) of the gap formed at the edge, and multiple training laminate substrates with different materials constituting the surfaces of the first and second substrates.

[0052] The plurality of fillers are actually applied to the plurality of learning laminate substrates. In the following description, these fillers are referred to as learning fillers. The plurality of learning fillers includes a plurality of fillers having different compositions. Different compositions refer to compositions in which at least one of the components constituting the filler is different in material and / or amount. For example, the learning fillers include a plurality of learning fillers with different binder materials, a plurality of learning fillers with different amounts of solvent, and a plurality of learning fillers with different amounts of particles (including no particles).

[0053] The multiple application conditions as correct labels are multiple application conditions under which good application results for the multiple fillers are obtained. More specifically, the multiple application conditions as correct labels are application conditions under which good application results are obtained when multiple learning fillers are actually applied to multiple learning laminate substrates. The multiple application conditions as correct labels include at least one of the total application amount of the learning filler, the application amount of the learning filler per unit time, the temperature of the learning filler, and the rotation speed of the learning laminate substrate. The multiple application conditions as correct labels include multiple application conditions that are different from each other. For example, the multiple different application conditions include at least one of different total application amounts of the learning filler, different application amounts of the learning filler per unit time, different temperatures of the learning filler, and different rotation speeds of the learning laminate substrate.

[0054] As described above, the correct label indicates the application conditions that resulted in a good application result. Whether the application condition is good or not is determined by observing the edge of the learning laminate substrate to which the learning filler has actually been applied. Specifically, the learning laminate substrate is thinned after the learning filler has been applied, and the application condition can be determined based on the number or size of cracks or chippings that occur on the knife edge portion of the learning laminate substrate (see reference numeral 2 in Figure 1(c)). For example, if the number of chips on the knife edge portion is less than a threshold value or if the chippings are smaller than the threshold value (including cases where no chips occur on the knife edge portion), the application condition of the learning filler is determined to be good.

[0055] In another example, an image of the edge of a learning laminate substrate coated with learning filler is generated using an infrared microscope, and the coating condition can be determined based on the size of the learning filler within a target area on the image. More specifically, if the area or width of the learning filler within the target area is larger than a threshold, the coating condition of the learning filler is determined to be good. The target area is a predetermined observation area on the image.

[0056] The infrared microscope is configured to irradiate the edge portion of the learning laminate substrate with infrared light and generate an image from the infrared light transmitted through or reflected from the edge portion of the learning laminate substrate. The infrared light passes through the first and second substrates made of silicon and is reflected by the filler. As a result, the image generated by the infrared microscope shows the filler applied to the gaps in the learning laminate substrate. Therefore, the application state of the learning filler can be determined based on the image generated by the infrared microscope.

[0057] In yet another example, the state of application of the learning filler can be determined by cutting the learning laminated substrate to which the learning filler has been applied and visually observing the edge portion of the learning laminated substrate.

[0058] The coating conditions that produced a good coating state (i.e., a good coating result) are training coating conditions, and are associated (linked) as correct labels with the data on the corresponding training laminate substrate and the data on the corresponding training filler. The correct labels are included in the training data along with the data on the corresponding training laminate substrate and the data on the corresponding training filler. The training data is also called learning data or teacher data.

[0059] In machine learning, when data on the training laminated substrate and data on the training filler are input to the model, the parameters of the model (e.g., weights) are adjusted so that the model outputs the application conditions as correct labels. The data on the training laminated substrate and the data on the training filler are explanatory variables, and the application conditions are objective variables. The computing system 10 stores the trained model constructed by machine learning in the storage device 10a.

[0060] FIG. 7 is a flowchart illustrating one embodiment of a method for building a trained model using machine learning. In step 101, the calculation system 10 acquires data on a plurality of learning laminated substrates, data on a plurality of learning fillers, and a plurality of application conditions as correct labels via an input device or signal communication (not shown). The calculation system 10 stores the acquired data and correct labels in the storage device 10a. In step 102, the computing system 10 associates data on a plurality of training laminate substrates and data on a plurality of training fillers with a plurality of application conditions that are corresponding ground truth labels to create training data. In step 103, the computing system 10 executes machine learning using the training data to construct (create) a trained model. The computing system 10 stores the trained model in the storage device 10a.

[0061] Next, the calculation system 10 uses the trained model to determine optimal application conditions for the laminated substrate Ws to which the filler F is to be applied. More specifically, as shown in FIG. 6, the calculation system 10 inputs data about the laminated substrate Ws to which the filler F is to be applied (e.g., the shape and size of the gap G at the edge portion) and data about the filler F to be applied to the laminated substrate Ws (e.g., the composition of the filler F) into the trained model, and performs calculations according to an algorithm defined by the trained model, thereby outputting application conditions from the trained model. Then, the calculation system 10 issues a command to the filler application module 12, causing the filler application module 12 to apply the filler F to the gap G (see FIG. 1(a)) between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 that constitute the laminated substrate Ws in accordance with the application conditions output from the trained model.

[0062] FIG. 8 is a flowchart illustrating one embodiment of a method for applying filler F to a laminated substrate Ws according to application conditions determined using a trained model. In step 201, the calculation system 10 inputs data relating to the laminated substrate Ws to which the filler F is to be applied and data on the filler F to be applied to the laminated substrate Ws into the trained model. In step 202, the calculation system 10 performs calculations according to the algorithm defined by the trained model, thereby outputting the application conditions from the trained model. In step 203, the calculation system 10 issues a command to the filler application module 12 to apply filler F to the gap G between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 that constitute the laminated substrate Ws. The filler application module 12 applies the filler F to the gap G at the edge portion of the laminated substrate Ws in accordance with the application conditions determined in step 202. Thereafter, the filler F is hardened by the hardening device 20 shown in FIGS. 3 and 4, and the laminated substrate Ws with the hardened filler F is thinned by a thinning device (not shown).

[0063] According to this embodiment, appropriate application conditions are determined by the trained model, and an appropriate amount of filler F is applied to the laminated substrate Ws in accordance with these application conditions. As a result, cracks and chipping at the knife edge of the laminated substrate Ws can be prevented, and contamination of the surrounding environment can also be prevented.

[0064] The coating conditions obtained in step 202 are expected to produce good coating results, and can therefore be used as correct labels. Therefore, in one embodiment, the computing system 10 may update the training data by adding the data about the laminated substrate Ws and the data about the filler F input to the trained model in step 201, and the coating conditions output from the trained model in step 202 as correct labels, and then update the trained model by performing machine learning again using the updated training data. Updating the trained model in this manner can improve the accuracy of the trained model.

[0065] Next, a description will be given of another embodiment of the substrate processing method and the substrate processing apparatus 5. The configuration and operation of this embodiment that are not particularly described are the same as those of the embodiment described with reference to FIGS.

[0066] In this embodiment, as shown in Figure 9, the calculation system 10 is configured to input data regarding the laminated substrate Ws and data regarding the filler F, as well as data regarding the curing device 20 (see Figures 3 and 4) for curing the filler F, into the trained model, and output the application conditions for the filler F and the curing conditions for the filler F from the trained model.

[0067] The data related to the curing device 20 includes the type of the curing device 20 and the distance between the curing device 20 and the edge of the laminate substrate Ws. Examples of the type of the curing device 20 include a lamp heater, an air heater, and a UV irradiation device. The type of the curing device 20 is predetermined based on the material of the filler F.

[0068] The curing conditions include the output value of the curing device 20. The output value of the curing device 20 varies depending on the type of the curing device 20. For example, if the curing device 20 is a lamp heater, the output value of the curing device 20 is the lamp power [W]; if the curing device 20 is an air heater, the output value of the curing device 20 is the hot air temperature or the heat source output [W]; and if the curing device 20 is a UV irradiation device, the output value is the irradiance of the UV source.

[0069] The calculation system 10 is configured to give instructions to the filler application module 12, cause the filler application module 12 to apply filler F to the gap G (see Figure 1(a)) between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 that constitute the laminated substrate Ws in accordance with the application conditions output from the trained model, and further cause the curing device 20 to harden the filler F in accordance with the curing conditions output from the trained model.

[0070] The training data used for machine learning to build the trained model includes data on multiple training laminate substrates, data on multiple training fillers, data on multiple curing devices, and multiple application conditions and multiple curing conditions as correct labels.

[0071] The data for the plurality of curing devices in the training data includes data for different curing devices, the data for the different curing devices including at least one of different types of curing devices and different distances between the curing devices and the edge of the laminate substrate.

[0072] The multiple application conditions and multiple curing conditions for the correct label are the multiple application conditions and multiple curing conditions under which good application results are obtained for the multiple fillers. More specifically, the multiple application conditions and multiple curing conditions for the correct label are the application conditions and multiple curing conditions under which good application results are obtained when multiple learning fillers are actually applied to multiple learning laminated substrates and the applied learning fillers are cured by a curing device. As described in the above embodiment, whether the application state is good or not is determined by observing the edge portion of the learning laminated substrate to which the learning filler has actually been applied.

[0073] The multiple curing conditions used as correct labels include the output value of the curing device. The output value of the curing device varies depending on the type of curing device. For example, if the curing device is a lamp heater, the output value of the curing device is the lamp power [W]. If the curing device is an air heater, the output value of the curing device is the temperature of the hot air or the output of the heat source [W]. If the curing device is a UV irradiation device, the output value of the UV source is the irradiance.

[0074] FIG. 10 is a flowchart illustrating one embodiment of a method for building a trained model using machine learning. In step 301, the calculation system 10 acquires data on a plurality of learning laminated substrates, data on a plurality of learning fillers, data on a plurality of curing devices, and a plurality of application conditions and a plurality of curing conditions as correct labels via an input device or signal communication (not shown). The calculation system 10 stores the acquired data and correct labels in the storage device 10a. In step 302, the computing system 10 associates data on a plurality of training laminate substrates, data on a plurality of training fillers, and data on a plurality of curing devices with corresponding correct labels of a plurality of application conditions and a plurality of curing conditions to create training data. In step 303, the computing system 10 executes machine learning using the training data to construct (create) a trained model. The computing system 10 stores the trained model in the storage device 10a.

[0075] Next, the calculation system 10 uses the trained model to determine optimal application conditions for the laminated substrate Ws to which the filler F is to be applied and optimal curing conditions for curing the filler F applied to the laminated substrate Ws. More specifically, as shown in Fig. 9, the calculation system 10 inputs data about the laminated substrate Ws to which the filler F is to be applied (e.g., the shape and size of the gap G at the edge portion), data about the filler F to be applied to the laminated substrate Ws (e.g., the composition of the filler F), and data about the curing device 20 (e.g., the type of the curing device 20) into the trained model, and performs calculations according to an algorithm defined by the trained model, thereby outputting the application conditions and curing conditions from the trained model. Then, the calculation system 10 gives an instruction to the filler application module 12, causing the filler application module 12 to apply filler F to the gap G (see Figure 1(a)) between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 that make up the laminated substrate Ws in accordance with the application conditions output from the trained model, and further causing the curing device 20 to harden the filler F in accordance with the curing conditions output from the trained model.

[0076] FIG. 11 is a flowchart illustrating one embodiment of a method for applying filler F to a laminated substrate Ws and curing the filler F according to application conditions and curing conditions determined using a trained model. In step 401, the calculation system 10 inputs data regarding the laminated substrate Ws to which the filler F is to be applied, data regarding the filler F to be applied to the laminated substrate Ws, and data regarding the curing device 20 that hardens the applied filler F into the trained model. In step 402, the calculation system 10 performs calculations according to the algorithm defined by the trained model, thereby outputting the application conditions and curing conditions from the trained model.

[0077] In step 403, the calculation system 10 issues a command to the filler application module 12 to apply filler F to the gap G between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 that constitute the laminated substrate Ws. The filler application module 12 applies the filler F to the gap G at the edge portion of the laminated substrate Ws in accordance with the application conditions determined in step 402 above. In step 404, the computing system 10 issues a command to the filler application module 12 to cause the curing device 20 to harden the filler F filled in the gaps G of the laminated substrate Ws. The hardening device 20 of the filler application module 12 hardens the filler F in accordance with the hardening conditions determined in step 402. Steps 403 and 404 may overlap in time. Thereafter, the laminated substrate Ws with the hardened filler F is thinned by a thinning device (not shown).

[0078] According to this embodiment, an appropriate amount of filler F is applied to the laminated substrate Ws according to the application conditions, and the filler F is appropriately cured according to the curing conditions. As a result, cracks and chipping at the knife edge of the laminated substrate Ws can be prevented, and contamination of the surrounding environment can also be prevented.

[0079] The coating conditions and curing conditions obtained in step 402 are expected to produce good coating results, and can therefore be used as correct labels. Therefore, in one embodiment, the computing system 10 may update the training data by adding the data about the laminated substrate Ws, the data about the filler F, and the data about the curing device 20 input to the trained model in step 401, and the coating conditions and curing conditions output from the trained model in step 402 as correct labels, and then update the trained model by performing machine learning again using the updated training data. Updating the trained model in this way can improve the accuracy of the trained model.

[0080] Next, a description will be given of another embodiment of the substrate processing method and the substrate processing apparatus 5. The configuration and operation of this embodiment, which will not be specifically described, are the same as those of the embodiment described with reference to Figures 9 to 11, and therefore, redundant description will be omitted.

[0081] 12 is a schematic diagram showing another embodiment of the substrate processing apparatus 5. As shown in FIG. 12, the substrate processing apparatus 5 of this embodiment further includes a thinning device 40 that thins the laminated substrate Ws to which the filler F has been applied and cured by the filler application module 12. The computing system 10 is configured to control the operations of the filler application module 12 and the thinning device 40. The laminated substrate Ws is transported from the filler application module 12 to the thinning device 40 by a transport device (not shown). The filler application module 12 and / or the thinning device 40 may be located at a location remote from the computing system 10.

[0082] 13 is a schematic diagram showing one embodiment of a thinning apparatus 40. The thinning apparatus 40 includes a holding stage 41 that holds the laminated substrate Ws, a stage rotation device 44 that rotates the holding stage 41, a cutting tool 47 that cuts the second substrate W2 that constitutes the laminated substrate Ws on the holding stage 41, a cutting tool rotation device 48 that rotates the cutting tool 47, and a cutting tool pressing device 51 that presses the cutting tool 47 against the laminated substrate Ws on the holding stage 41. The cutting tool 47 has a cutting surface 47a to which cutting particles such as diamond particles are fixed. The holding stage 41 is configured to be able to hold the laminated substrate Ws on its stage surface 41a by vacuum suction or the like.

[0083] The laminated substrate Ws is placed on the stage surface 41a of the holding stage 41 with the second substrate W2 facing the cutting surface 47a of the cutting tool 47. The first substrate W1 of the laminated substrate Ws is held on the stage surface 41a of the holding stage 41 by vacuum suction or the like. When the stage rotation device 44 rotates the holding stage 41, the laminated substrate Ws on the holding stage 41 rotates. While the cutting tool rotation device 48 rotates the cutting tool 47, the cutting tool pressing device 51 presses the cutting surface 47a of the cutting tool 47 against the second substrate W2 of the laminated substrate Ws. The second substrate W2 is cut by the cutting tool 47, thereby thinning the laminated substrate Ws (see FIG. 1(c)). The rotational speed of the cutting tool 47 is adjusted by the cutting tool rotation device 48, the pressing force of the cutting tool 47 against the laminated substrate Ws is adjusted by the cutting tool pressing device 51, and the rotational speed of the laminated substrate Ws is adjusted by the stage rotation device 44.

[0084] FIG. 14 is a conceptual diagram illustrating the operation of the computing system 10 of this embodiment. As shown in FIG. 14, the computing system 10 is configured to input data related to the laminated substrate Ws, data related to the filler F, data related to the curing device 20, and data related to the thinning device 40 into a trained model, and output from the trained model application conditions for the filler F, curing conditions for the filler F, and thinning conditions for the laminated substrate Ws. Furthermore, the computing system 10 is configured to issue a command to the filler application module 12, causing the filler application module 12 to apply the filler F to the gap G (see FIG. 1(a)) between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 that constitute the laminated substrate Ws in accordance with the application conditions output from the trained model, and further to harden the filler F using the curing device 20 in accordance with the curing conditions output from the trained model. Furthermore, the computing system 10 is configured to issue a command to the thinning device 40, causing the laminated substrate Ws containing the cured filler F to be thinned in accordance with the thinning conditions output from the trained model.

[0085] The data related to the thinning apparatus 40 includes the type of cutting tool 47 used in the thinning apparatus 40 and the target cutting amount of the laminated substrate Ws. The type of cutting tool 47 is, for example, the coarseness of the cutting particles that make up the cutting surface 47a. The thinning conditions are, in other words, the operating conditions of the thinning apparatus 40, and include, for example, the rotation speed of the cutting tool 47, the pressing force of the cutting tool 47 against the laminated substrate Ws, and the rotation speed of the laminated substrate Ws (the rotation speed of the holding stage 41).

[0086] The training data used for machine learning to build the trained model includes data on multiple training laminate substrates, data on multiple training fillers, data on multiple curing devices, data on multiple thinning devices, and multiple application conditions, multiple curing conditions, and multiple thinning conditions as correct labels.

[0087] The data for the plurality of thinning devices in the training data includes data for a plurality of different thinning devices, the data for the plurality of different thinning devices including at least one of different types of cutting tools and different target cutting amounts of the laminate substrate.

[0088] The multiple application conditions, multiple curing conditions, and multiple thinning conditions as the correct label are the multiple application conditions, multiple curing conditions, and multiple thinning conditions under which good application results and good thinning results are obtained. More specifically, the multiple application conditions, multiple curing conditions, and multiple thinning conditions as the correct label are the application conditions, multiple curing conditions, and multiple thinning conditions under which good application results and good thinning results are obtained when multiple learning fillers are actually applied to multiple learning laminated substrates, the applied learning fillers are cured by a curing device, and the laminated substrates are then thinned by a thinning device.

[0089] As described in the above-described embodiment, whether the application condition is satisfactory is determined by observing the edge of the learning laminate substrate to which the learning filler has actually been applied. A satisfactory thinning result is determined based on whether the cutting tool has thinned the laminate substrate without clogging and whether the time from the start of cutting to reaching the target removal amount is within the allowable time. Typically, clogging of the cutting tool is caused by particles contained in the filler. The time from the start of cutting to reaching the target removal amount varies depending on the hardness of the hardened filler, which in turn depends on the filler material. Therefore, the thinning condition for the correct label is the thinning condition when the cutting tool does not clog and thinning is completed in a short time.

[0090] FIG. 15 is a flowchart illustrating one embodiment of a method for building a trained model using machine learning. In step 501, the calculation system 10 acquires data on a plurality of learning laminated substrates, data on a plurality of learning fillers, data on a plurality of curing devices, data on a plurality of thinning devices, and a plurality of application conditions, a plurality of curing conditions, and a plurality of thinning conditions as correct labels via an input device or signal communication (not shown). The calculation system 10 stores the acquired data and correct labels in the storage device 10a. In step 502, the computing system 10 associates data relating to a plurality of training laminate substrates, data relating to a plurality of training fillers, data relating to a plurality of curing devices, and data relating to a plurality of thinning devices with corresponding correct labels of a plurality of application conditions, a plurality of curing conditions, and a plurality of thinning conditions to create training data. In step 503, the computing system 10 executes machine learning using the training data to construct (create) a trained model. The computing system 10 stores the trained model in the storage device 10a.

[0091] Next, the calculation system 10 uses the trained model to determine optimal application conditions for the laminated substrate Ws to be applied with filler F, optimal curing conditions for curing the filler F applied to the laminated substrate Ws, and optimal thinning conditions for thinning the laminated substrate Ws on which the filler F has been cured. More specifically, as shown in Fig. 14, the calculation system 10 inputs data about the laminated substrate Ws to be applied with filler F (e.g., the shape and size of the gap G at the edge), data about the filler F to be applied to the laminated substrate Ws (e.g., the composition of the filler F), data about the curing device 20 (e.g., the type of curing device 20), and data about the thinning device 40 (e.g., the type of cutting tool) into the trained model, and performs calculations according to an algorithm defined by the trained model to output the application conditions, curing conditions, and thinning conditions from the trained model.

[0092] Then, the calculation system 10 issues a command to the filler application module 12, causing the filler application module 12 to apply filler F to the gap G (see FIG. 1(a)) between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 that constitute the laminated substrate Ws in accordance with the application conditions output from the trained model, and further causes the curing device 20 to harden the filler F in accordance with the curing conditions output from the trained model. Furthermore, the calculation system 10 issues a command to the thinning device 40, causing the thinning device 40 to thin the laminated substrate Ws on which the filler F has hardened in accordance with the thinning conditions output from the trained model.

[0093] Figure 16 is a flowchart illustrating one embodiment of a method for applying filler F to a laminated substrate Ws, curing the filler F, and thinning the laminated substrate Ws according to application conditions, curing conditions, and thinning conditions determined using a trained model. In step 601, the calculation system 10 inputs data regarding the laminated substrate Ws to which the filler F is to be applied, data regarding the filler F to be applied to the laminated substrate Ws, data regarding the curing device 20 that hardens the applied filler F, and data regarding the thinning device 40 that thins the laminated substrate Ws on which the filler F has hardened into the trained model. In step 602, the calculation system 10 performs calculations according to the algorithm defined by the trained model, thereby outputting the application conditions, curing conditions, and thinning conditions from the trained model.

[0094] In step 603, the calculation system 10 issues a command to the filler application module 12 to apply filler F to the gap G between the edge portion E1 of the first substrate W1 and the edge portion E2 of the second substrate W2 that constitute the laminated substrate Ws. The filler application module 12 applies the filler F to the gap G at the edge portion of the laminated substrate Ws in accordance with the application conditions determined in step 602 above. In step 604, the computing system 10 issues a command to the filler application module 12 to harden the filler F filled in the gap G of the laminated substrate Ws by the hardening device 20. The hardening device 20 of the filler application module 12 hardens the filler F in accordance with the hardening conditions determined in step 602. Steps 603 and 604 may overlap in time. In step 605, the calculation system 10 issues a command to the thinning device 40 to operate the thinning device 40 in accordance with the thinning conditions output from the trained model, thereby thinning the laminated substrate Ws. The thinning device 40 thins the laminated substrate Ws having the hardened filler F in accordance with the thinning conditions determined in step 602.

[0095] According to this embodiment, an appropriate amount of filler F is applied to the laminated substrate Ws according to the application conditions, the filler F is appropriately cured according to the curing conditions, and the laminated substrate Ws is appropriately thinned according to the thinning conditions. As a result, cracks and chipping at the knife edge of the laminated substrate Ws can be prevented, and contamination of the surrounding environment can be prevented. In addition, because the thinning conditions are optimized, clogging of the cutting tool can be prevented, and the throughput of the thinning process can be improved.

[0096] The coating conditions, curing conditions, and thinning conditions obtained in step 602 are expected to produce good coating and thinning results, and can therefore be used as correct labels. Therefore, in one embodiment, the computing system 10 may update the training data by adding the data about the laminated substrate Ws, the data about the filler F, the data about the curing device 20, and the data about the thinning device 40 input to the trained model in step 601, and the coating conditions, curing conditions, and thinning conditions output from the trained model in step 602 as correct labels, and then update the trained model by performing machine learning again using the updated training data. Updating the trained model in this manner can improve the accuracy of the trained model.

[0097] Next, another embodiment of the substrate processing method will be described. The configuration and operation of this embodiment, which will not be specifically described, are the same as those of the embodiment described with reference to Figures 12 to 16, and therefore, redundant description will be omitted.

[0098] In this embodiment, the trained model is configured to calculate thinning conditions for thinning a laminated substrate Ws that has already been coated with filler F and cured. FIG. 17 is a conceptual diagram illustrating the operation of the calculation system 10 of this embodiment. As shown in FIG. 17, the calculation system 10 is configured to input data related to the laminated substrate Ws, data related to the filler F, and data related to the thinning device 40 into the trained model and output the thinning conditions for the laminated substrate Ws from the trained model. Furthermore, the calculation system 10 is configured to give a command to the thinning device 40 to thin the laminated substrate Ws that has the cured filler F in accordance with the thinning conditions output from the trained model.

[0099] The training data used in the machine learning to build the trained model includes data on a plurality of training laminate substrates, data on a plurality of training fillers, data on a plurality of thinning devices, and a plurality of thinning conditions as correct labels. The data on the plurality of thinning devices in the training data includes data on a plurality of different thinning devices. The data on the plurality of different thinning devices includes at least one of different types of cutting tools and different target cutting amounts of the laminate substrate.

[0100] The multiple thinning conditions that are labeled as correct are those under which good thinning results are obtained. More specifically, the multiple thinning conditions that are labeled as correct are those under which good thinning results are obtained when a laminated substrate having hardened filler in gaps at its edge is thinned by a thinning device. The thinning conditions under which good thinning results are obtained are those under which the cutting tool does not clog and thinning is completed in a short time.

[0101] FIG. 18 is a flowchart illustrating one embodiment of a method for building a trained model using machine learning. In step 701, the calculation system 10 acquires data on a plurality of learning laminated substrates, data on a plurality of learning fillers, data on a plurality of thinning devices, and a plurality of thinning conditions as correct labels via an input device or signal communication (not shown). The calculation system 10 stores the acquired data and correct labels in the storage device 10a. In step 702, the computing system 10 associates data on a plurality of training laminate substrates, data on a plurality of training fillers, and data on a plurality of thinning devices with a plurality of thinning conditions that are corresponding ground truth labels to create training data. In step 703, the computing system 10 executes machine learning using the training data to construct (create) a trained model. The computing system 10 stores the trained model in the storage device 10a.

[0102] Next, the calculation system 10 uses the trained model to determine optimal thinning conditions for thinning the laminated substrate Ws on which the filler F has been cured. More specifically, as shown in FIG. 17 , the calculation system 10 inputs data about the laminated substrate Ws on which the filler F has been applied (e.g., the shape and size of the gap G at the edge), data about the filler F applied to the laminated substrate Ws (e.g., the composition of the filler F), and data about the thinning device 40 (e.g., the type of cutting tool 47) into the trained model, and performs calculations according to an algorithm defined by the trained model, thereby outputting thinning conditions from the trained model. The calculation system 10 then issues a command to the thinning device 40, causing the thinning device 40 to thin the laminated substrate Ws on which the filler F has been cured, according to the thinning conditions output from the trained model.

[0103] FIG. 19 is a flowchart illustrating one embodiment of a method for thinning a laminated substrate Ws in which filler F has hardened, according to thinning conditions determined using a trained model. In step 801, the calculation system 10 inputs data regarding the laminated substrate Ws to which filler F has been applied, data regarding the filler F applied to the laminated substrate Ws, and data regarding the thinning device 40 that thins the laminated substrate Ws on which the filler F has hardened into the trained model. In step 802, the calculation system 10 outputs the thinning conditions from the learned model by performing calculations according to the algorithm defined by the learned model. In step 803, the calculation system 10 issues a command to the thinning device 40, causing the thinning device 40 to operate in accordance with the thinning conditions output from the trained model, thereby thinning the laminated substrate Ws.

[0104] According to this embodiment, the thinning conditions are optimized, so that clogging of the cutting tool can be prevented and the throughput of the thinning process can be improved.

[0105] The thinning conditions obtained in step 802 are expected to produce good thinning results, and can therefore be used as correct labels. Therefore, in one embodiment, the computing system 10 may update the training data by adding the data about the laminated substrate Ws, the data about the filler F, and the data about the thinning device 40 input to the trained model in step 801, and the thinning conditions output from the trained model in step 802 as correct labels, and then update the trained model by performing machine learning again using the updated training data. Updating the trained model in this manner can improve the accuracy of the trained model.

[0106] The arrangement of the elements constituting the substrate processing apparatus 5 according to each embodiment described with reference to FIGS. 3 to 19 is not particularly limited. For example, as shown in FIG. 20, the computing system 10, the filler application module 12, and the thinning apparatus 40 may be located within a single factory. Alternatively, as shown in FIG. 21, the filler application module 12 and the thinning apparatus 40 may be located within a single factory, while the computing system 10 may be located outside the factory. The computing system 10 is connected to the filler application module 12 and the thinning apparatus 40 via a communication network such as the Internet. In another example, as shown in FIG. 22, the computing system 10 may be connected to a plurality of remote filler application modules 12 and a plurality of remote thinning apparatuses 40 via a communication network such as the Internet.

[0107] The operation of the substrate processing apparatus 5 in each of the above-described embodiments is controlled by a computing system 10. The computing system 10 is configured with at least one computer. The computing system 10 operates according to instructions contained in a program electronically stored in a storage device 10a. That is, the computing system 10 issues commands to the filler application module 12 and / or the thinning device 40 to execute the substrate processing method of any of the above-described embodiments. The program for causing the computing system 10 to execute such operations is recorded on a computer-readable recording medium, which is a non-transitory tangible object, and provided to the computing system 10 via the recording medium. Alternatively, the program may be input to the computing system 10 from a communication device via a communication network such as the Internet or a local area network.

[0108] Figure 23 is a top view showing another embodiment of the filler application module 12. Figure 24 is a side view of the filler application module 12 shown in Figure 23. The configuration of this embodiment not specifically described is similar to the configuration of the embodiment described with reference to Figures 3 and 4, so redundant description will be omitted. In this embodiment, the filler application module 12 includes a substrate holding device 60 instead of the substrate holding unit 15, the rotation shaft 22, and the rotation mechanism 23.

[0109] The substrate holding device 60 includes three or more (four in this embodiment) rollers 61 that can come into contact with the peripheral edge of the laminated substrate Ws, a roller rotation mechanism (not shown) that rotates each roller 61 around its axis, and a roller movement mechanism (not shown) that moves each roller 61. In this embodiment, the substrate holding device 60 includes four rollers 61, but the substrate holding device 60 may include three, or five or more rollers.

[0110] The four rollers 61 are arranged around the central axis Cr of the laminated substrate Ws. The rollers 61 are configured to contact the peripheral edge of the laminated substrate Ws and hold the laminated substrate Ws horizontally. The roller rotation mechanism may have any configuration as long as it can rotate three or more rollers 61 in the same direction at the same speed, and any known rotation mechanism can be used as the roller rotation mechanism. Examples of roller rotation mechanisms include a combination of a motor, pulleys (and / or gears), and a rotating belt.

[0111] The roller rotation mechanism is connected to the four rollers 61 and is configured to rotate the four rollers 61 in the same direction at the same speed. The roller movement mechanism can move the four rollers 61 between a holding position (see solid lines in FIG. 23) where the peripheral edge of the laminated substrate Ws is held by the rollers 61 and a release position (see dotted lines in FIG. 23) where the laminated substrate Ws is released from the rollers 61.

[0112] The roller movement mechanism may have any configuration as long as it can move the four rollers 61 between the holding position and the release position, and any known movement mechanism can be used as the roller movement mechanism. Examples of the roller movement mechanism include a piston-cylinder mechanism and a combination of a ball screw and a motor (stepping motor).

[0113] The laminated substrate Ws is transported by a transport device (not shown) to a position where the axis of the laminated substrate Ws coincides with the central axis Cr of the laminated substrate Ws. At this time, the rollers 61 are in the release position. Next, the roller moving mechanism moves the four rollers 61 to the holding position, causing the peripheral edge of the laminated substrate Ws to be held by the four rollers 61. This operation causes the laminated substrate Ws to be held in a horizontal position by the four rollers 61. The four rollers 61 moved to the holding position are rotated by the roller rotation mechanism, causing the laminated substrate Ws to rotate around its axis.

[0114] When the roller movement mechanism moves the four rollers 61 from the holding position to the release position, the four rollers 61 move away from the peripheral edge of the laminated substrate Ws, releasing the laminated substrate Ws from the four rollers 61. The released laminated substrate Ws is transported by a transport device (not shown) for subsequent processing. The application of filler F by the application device 16 and the curing of filler F by the curing device 20 are performed while rotating the laminated substrate Ws, which is held horizontally by the substrate holding device 60. The rotation speed of the laminated substrate Ws depends on the rotation speed of the rollers 61 of the substrate holding device 60 and can be controlled by the roller rotation mechanism.

[0115] In one embodiment, the roller rotation mechanism may be configured to rotate only some of the rollers 61. For example, the roller rotation mechanism may be connected to two of the four rollers 61 and configured to rotate the two rollers 61 in the same direction at the same speed. In this case, the other two rollers 61 are configured to rotate freely. When the four rollers 61 are placed in the holding position, as the two rollers 61 connected to the roller rotation mechanism rotate, the other two rollers 61 rotate in response to the rollers 61 connected to the roller rotation mechanism via the laminated substrate Ws.

[0116] In one embodiment, the roller moving mechanism may be configured to move only some of the rollers 61. For example, the roller moving mechanism may be coupled to two of the four rollers 61 and move these two rollers 61 between a holding position and a release position. In this case, the other two rollers 61 are fixed in advance to the holding position. The transport device transports the laminated substrate Ws to a position where the peripheral edge of the laminated substrate Ws contacts the two fixed rollers 61. The roller moving mechanism moves the two rollers 61 coupled to the roller moving mechanism to the holding position, thereby holding the laminated substrate Ws in a horizontal position. The roller moving mechanism moves the two rollers 61 coupled to the roller moving mechanism to the release position, thereby releasing the laminated substrate Ws.

[0117] In the above-described embodiment, the substrate holding unit 15 and the rollers 61 of the substrate holding device 60 are configured to hold the laminated substrate Ws horizontally. That is, the laminated substrate Ws is held in a horizontal position by the substrate holding unit 15 or the rollers 61 of the substrate holding device 60. The application device 16 applies the filler F while rotating the horizontally placed laminated substrate Ws by the substrate holding unit 15 or the rollers 61 of the substrate holding device 60. However, as long as the filler F can be applied to the gap G, the method of holding the laminated substrate Ws is not limited to the above-described embodiment. For example, the filler application module 12 may have a substrate holding unit or a substrate holding device configured to hold the laminated substrate Ws vertically. When the laminated substrate Ws is held in a vertical position, the upper and lower surfaces of the laminated substrate Ws are located within imaginary planes extending vertically perpendicular to the horizontal direction.

[0118] FIG. 25 is a side view showing yet another embodiment of the filler application module 12. FIG. 26 is a view from the direction indicated by arrow A in FIG. 25. Configurations of this embodiment that are not particularly described are similar to those of the embodiment described with reference to FIGS. 3 and 4, so redundant description will be omitted. FIG. 25 is a view from the back surface side of the laminated substrate Ws. In this embodiment, the filler application module 12 includes a substrate holding unit 65, a rotation shaft 66, and a rotation mechanism 68 instead of the substrate holding unit 15, the rotation shaft 22, and the rotation mechanism 23.

[0119] The substrate holding unit 65 is configured to hold the back surface of the multilayer substrate Ws by vacuum suction. As shown in Fig. 26, a holding surface 65a of the substrate holding unit 65 that holds the back surface of the multilayer substrate Ws is a surface perpendicular to the horizontal plane. The multilayer substrate Ws is held so that it is perpendicular to the horizontal plane. In other words, the multilayer substrate Ws is held in a vertically oriented state by the substrate holding unit 65.

[0120] The rotation shaft 66 is connected to the center of the substrate holding part 65. The laminated substrate Ws is held by the substrate holding part 65 so that the center of the laminated substrate Ws coincides with the axis of the rotation shaft 66. The rotation mechanism 68 includes a motor (not shown), and as shown in FIG. 25 , the rotation mechanism 68 is configured to rotate the substrate holding part 65 and the laminated substrate Ws together in the direction indicated by the arrow around the central axis Cr of the laminated substrate Ws.

[0121] The applicator 16 is disposed above the laminated substrate Ws held by the substrate holding unit 65, facing the gap G between the laminated substrates Ws. When the applicator 16 dispenses the filler F, the filler F falls toward the gap G between the laminated substrates Ws, thereby coating the gap G between the laminated substrates Ws. The curing device 20 is located radially outward of the laminated substrate Ws held by the substrate holding unit 65. The curing device 20 is disposed downstream of the applicator 16 in the rotation direction of the laminated substrate Ws and is configured to cure the filler F applied to the laminated substrate Ws by the applicator 16. The application of the filler F by the applicator 16 and the curing of the filler F by the curing device 20 are performed while the laminated substrate Ws, held vertically by the substrate holding unit 65, is rotated. The rotation speed of the laminated substrate Ws corresponds to the rotation speed of the substrate holding unit 65 and can be controlled by the rotation mechanism 68.

[0122] Figure 27 is a side view showing yet another embodiment of the filler application module 12. Figure 28 is a view seen from the direction indicated by arrow B in Figure 27. Configurations of this embodiment that are not particularly described are similar to those of the embodiment described with reference to Figures 25 and 26, so redundant description will be omitted. In this embodiment, the filler application module 12 includes a substrate holding device 70 instead of the substrate holding unit 65, the rotation shaft 66, and the rotation mechanism 68.

[0123] The substrate holding device 70 includes three or more (four in this embodiment) rollers 71 that can come into contact with the peripheral edge of the laminated substrate Ws, a roller rotation mechanism (not shown) that rotates each roller 71 around its axis, and a roller movement mechanism (not shown) that moves each roller 71. In this embodiment, the substrate holding device 70 includes four rollers 71, but the substrate holding device 70 may include three, or five or more rollers.

[0124] The four rollers 71 are arranged around the central axis Cr of the laminated substrate Ws. The rollers 71 are configured to contact the peripheral portion of the laminated substrate Ws and hold the laminated substrate Ws vertically. That is, the laminated substrate Ws is held in a vertically oriented state by the rollers 71 of the substrate holding device 70. As shown in FIG. 27 , when the laminated substrate Ws is held in a vertically oriented state by the rollers 71 of the substrate holding device 70, the upper and lower surfaces of the laminated substrate Ws are each within an imaginary plane extending in the vertical direction.

[0125] The roller rotation mechanism is connected to the four rollers 71 and is configured to rotate the four rollers 71 in the same direction at the same speed. The roller rotation mechanism may be configured in any way as long as it can rotate three or more rollers 71 in the same direction at the same speed, and any known rotation mechanism can be used as the roller rotation mechanism. Examples of roller rotation mechanisms include a combination of a motor, pulleys (and / or gears), and a rotating belt.

[0126] The roller movement mechanism is connected to four rollers 71 and is configured to move each roller 71 in a direction toward the central axis Cr of the laminated substrate Ws and a direction away from the central axis Cr. The roller movement mechanism can move the four rollers 71 between a holding position (see solid lines in FIG. 27 ) where the peripheral edge of the laminated substrate Ws is held by the rollers 71 and a release position (see dotted lines in FIG. 27 ) where the laminated substrate Ws is released from the rollers 71. The roller movement mechanism may be configured in any way as long as it can move the four rollers 71 between the holding position and the release position, and any known movement mechanism can be used as the roller movement mechanism. Examples of roller movement mechanisms include a piston-cylinder mechanism and a combination of a ball screw and a motor (stepping motor).

[0127] The laminated substrate Ws is transported by a transport device (not shown) to a position where the axis of the laminated substrate Ws coincides with the central axis Cr of the laminated substrate Ws. At this time, the rollers 71 are in the release position. Next, the roller moving mechanism moves the four rollers 71 to the holding position, causing the peripheral edge of the laminated substrate Ws to be held by the four rollers 71. This operation causes the laminated substrate Ws to be held in a vertical position by the four rollers 71. The four rollers 71 moved to the holding position are rotated by the roller rotation mechanism, causing the laminated substrate Ws to rotate around its axis.

[0128] When the roller movement mechanism moves the four rollers 71 from the holding position to the release position, the four rollers 71 move away from the peripheral edge of the laminated substrate Ws, releasing the laminated substrate Ws from the four rollers 71. The released laminated substrate Ws is transported by a transport device (not shown) for subsequent processing. The application of filler F by the application device 16 and the curing of filler F by the curing device 20 are performed while rotating the laminated substrate Ws held vertically by the substrate holding device 70. The rotation speed of the laminated substrate Ws depends on the rotation speed of the rollers 71 of the substrate holding device 70 and can be controlled by the roller rotation mechanism.

[0129] In one embodiment, the roller rotation mechanism may be configured to rotate only some of the rollers 71. For example, the roller rotation mechanism may be connected to two of the four rollers 71 and rotate the two rollers in the same direction at the same speed. In this case, the other two rollers 71 are configured to rotate freely. When the four rollers 71 are positioned in the holding position, as the two rollers 71 connected to the roller rotation mechanism rotate, the other two rollers 71 rotate in response to the two rollers 71 connected to the roller rotation mechanism via the laminate substrate Ws.

[0130] In one embodiment, the roller moving mechanism may be configured to move only some of the rollers 71. For example, the roller moving mechanism may be coupled to two of the four rollers 71 and move these two rollers 71 between a holding position and a release position. In this case, the other two rollers 71 are fixed in advance to the holding position. The transport device transports the laminated substrate Ws to a position where the peripheral edge of the laminated substrate Ws contacts the two fixed rollers 71. The roller moving mechanism can hold the laminated substrate Ws in a vertical position by moving the two rollers 71 coupled to the roller moving mechanism to the holding position. The roller moving mechanism can release the laminated substrate Ws by moving the two rollers 71 coupled to the roller moving mechanism to the release position.

[0131] The embodiments described with reference to FIGS. 23 to 28 may be applied to the embodiments described with reference to FIGS. 9 to 11, 12 to 16, and 17 to 19.

[0132] The above-described embodiments have been described for the purpose of enabling a person of ordinary skill in the art to practice the present invention. Various modifications of the above-described embodiments would be obvious to a person skilled in the art, and the technical concept of the present invention may be applied to other embodiments. Therefore, the present invention is not limited to the described embodiments, but is to be interpreted in the broadest scope in accordance with the technical concept defined by the claims. [Explanation of symbols]

[0133] 2 Knife edge 5. Substrate processing equipment 10. Computing Systems 12 Filler application module 15 Board holding part 16 Coating equipment 20 Curing equipment 22 Rotation axis 23 Rotation mechanism 26 syringe 26a Filler outlet 27 Piston 28 Gas supply line 29 Pressure Regulating Device 40 Thinning device 41 Holding stage 44 Stage rotation device 47 Cutting tools 48 Cutting tool rotation device 51 Cutting tool pressing device 60 Substrate holding device 61 Roller 65 Board holding part 66 Rotation axis 68 Rotation mechanism 70 Substrate holding device 71 Roller Ws laminated substrate F filler G Gap

Claims

1. Data related to a laminated substrate in which a first substrate and a second substrate are joined, and data related to a filler are input into a trained model constructed by machine learning, the coating conditions of the filler are output from the trained model, while rotating the laminated substrate, the filler is applied to the gap between the edge portion of the first substrate and the edge portion of the second substrate according to the coating conditions, a substrate processing method.

2. The data related to the laminated substrate includes the materials constituting the surfaces of the first substrate and the second substrate, and the shape and size of the gap, The data related to the filler includes the composition of the filler, the substrate processing method according to claim 1.

3. The coating conditions include at least one of the total coating amount of the filler, the coating amount of the filler per unit time, the temperature of the filler, and the rotation speed of the laminated substrate, the substrate processing method according to claim 1 or 2.

4. In addition to the data related to the laminated substrate and the data related to the filler, data related to a curing device for curing the filler is input into the trained model, the coating conditions of the filler and the curing conditions of the filler are output from the trained model, the substrate processing method further includes curing the applied filler by the curing device according to the curing conditions, the substrate processing method according to claim 1 or 2.

5. The data related to the curing device includes the type of the curing device and the distance between the curing device and the edge portion of the laminated substrate, the substrate processing method according to claim 4.

6. The curing conditions include the output value of the curing device, the substrate processing method according to claim 4.

7. In addition to the data related to the laminated substrate, the data related to the filler, and the data related to the curing device, data related to a thinning device for thinning the laminated substrate is input into the trained model, the coating conditions of the filler, the curing conditions of the filler, and the thinning conditions of the laminated substrate are output from the trained model, the substrate processing method further includes thinning the laminated substrate by the thinning device according to the thinning conditions after curing of the filler, the substrate processing method according to claim 4.

8. The data related to the thinning device includes the type of cutting tool used in the thinning device and the target cutting amount of the laminated substrate, the substrate processing method according to claim 7.

9. The substrate processing method according to claim 8, wherein the thinning conditions include at least one of the pressing force of the cutting tool against the laminated substrate, the rotational speed of the cutting tool, and the rotational speed of the laminated substrate.

10. The substrate processing method according to claim 1, wherein the learned model is a model constructed by machine learning using training data including data on a plurality of laminated substrates, data on a plurality of fillers, and a plurality of coating conditions as correct labels.

11. The substrate processing method according to claim 4, wherein the learned model is a model constructed by machine learning using training data including data on a plurality of laminated substrates, data on a plurality of fillers, data on a plurality of curing devices, and a plurality of coating conditions and a plurality of curing conditions as correct labels.

12. The substrate processing method according to claim 7, wherein the learned model is a model constructed by machine learning using training data including data on a plurality of laminated substrates, data on a plurality of fillers, data on a plurality of curing devices, data on a plurality of thinning devices, and a plurality of coating conditions, a plurality of curing conditions, and a plurality of thinning conditions as correct labels.

13. Updating the training data by adding the data on the laminated substrate and the data on the filler input to the learned model and the coating conditions output from the learned model to the training data used for the machine learning, The substrate processing method according to claim 1, further comprising updating the learned model by performing machine learning using the updated training data.

14. Applying the filler to the gap between the edge portion of the first substrate and the edge portion of the second substrate according to the coating conditions while rotating the laminated substrate is to apply the filler to the gap between the edge portion of the first substrate and the edge portion of the second substrate according to the coating conditions while rotating the laminated substrate held vertically. The substrate processing method according to claim 1.

15. Inputting data on a laminated substrate in which a first substrate and a second substrate are joined, data on a filler applied to a gap between an edge portion of the first substrate and an edge portion of the second substrate, and data on a thinning device for thinning the laminated substrate into a learned model constructed by machine learning, Outputting the thinning conditions of the laminated substrate from the learned model, A substrate processing method of thinning the laminated substrate by the thinning device according to the thinning conditions.

16. An arithmetic system having a learned model constructed by machine learning, A filler application module that applies a filler to the laminated substrate while rotating the laminated substrate in which the first substrate and the second substrate are joined, The arithmetic system, Inputs data related to the laminated substrate and data related to the filler into the learned model, Outputs the application conditions of the filler from the learned model, A substrate processing apparatus configured to give a command to the filler application module to apply the filler to the gap between the edge portion of the first substrate and the edge portion of the second substrate by the filler application module according to the application conditions.

17. The data related to the laminated substrate includes the materials constituting the surfaces of the first substrate and the second substrate, and the shape and size of the gap, The data related to the filler includes the composition of the filler. The substrate processing apparatus according to claim 16.

18. The application conditions include at least one of the total application amount of the filler, the application amount of the filler per unit time, the temperature of the filler, and the rotation speed of the laminated substrate. The substrate processing apparatus according to claim 16 or 17.

19. The substrate processing apparatus further includes a curing device for curing the applied filler, The arithmetic system, In addition to the data related to the laminated substrate and the data related to the filler, inputs data related to the curing device into the learned model, Outputs the application conditions of the filler and the curing conditions of the filler from the learned model, A substrate processing apparatus configured to give a command to the curing device to cure the applied filler by the curing device according to the curing conditions. The substrate processing apparatus according to claim 16 or 17.

20. The data related to the curing device includes the type of the curing device and the distance between the curing device and the edge portion of the laminated substrate. The substrate processing apparatus according to claim 19.

21. The curing conditions include the output value of the curing device. The substrate processing apparatus according to claim 19.

22. The substrate processing apparatus further includes a thinning device for thinning the laminated substrate, The arithmetic system, In addition to the data related to the laminated substrate, the data related to the filler, and the data related to the curing device, input the data related to the thinning device into the learned model, Output the application conditions of the filler, the curing conditions of the filler, and the thinning conditions of the laminated substrate from the learned model, The substrate processing apparatus according to claim 19, wherein a command is given to the thinning device to thin the laminated substrate by the thinning device according to the thinning conditions after curing of the filler.

23. The substrate processing apparatus according to claim 22, wherein the data related to the thinning device includes the type of cutting tool used in the thinning device and the target cutting amount of the laminated substrate.

24. The substrate processing apparatus according to claim 23, wherein the thinning conditions include at least one of the pressing force of the cutting tool against the laminated substrate, the rotational speed of the cutting tool, and the rotational speed of the laminated substrate.

25. The substrate processing apparatus according to claim 16, wherein the learned model is a model constructed by machine learning using training data including data related to a plurality of laminated substrates, data related to a plurality of fillers, and a plurality of application conditions as correct labels.

26. The substrate processing apparatus according to claim 19, wherein the learned model is a model constructed by machine learning using training data including data related to a plurality of laminated substrates, data related to a plurality of fillers, data related to a plurality of curing devices, and a plurality of application conditions and a plurality of curing conditions as correct labels.

27. The substrate processing apparatus according to claim 22, wherein the learned model is a model constructed by machine learning using training data including data related to a plurality of laminated substrates, data related to a plurality of fillers, data related to a plurality of curing devices, data related to a plurality of thinning devices, and a plurality of application conditions, a plurality of curing conditions, and a plurality of thinning conditions as correct labels.

28. The arithmetic system is Update the training data by adding the data related to the laminated substrate and the data related to the filler input to the learned model and the application conditions output from the learned model to the training data used for the machine learning, The substrate processing apparatus according to claim 16, wherein machine learning is executed using the updated training data to update the learned model.

29. The substrate processing apparatus according to claim 16, wherein the filler application module includes a substrate holding unit or a substrate holding device that holds the laminated substrate in a vertically placed state.

30. An arithmetic system having a learned model constructed by machine learning, A thinning device for thinning a laminated substrate in which a first substrate and a second substrate are joined, The arithmetic system, Inputs data on the laminated substrate, data on a filler applied to a gap between an edge portion of the first substrate and an edge portion of the second substrate, and data on the thinning device into a learned model constructed by machine learning, Outputs thinning conditions for the laminated substrate from the learned model, A substrate processing apparatus configured to give an instruction to the thinning device to thin the laminated substrate by the thinning device according to the thinning conditions.

31. A step of inputting data on a laminated substrate in which a first substrate and a second substrate are joined and data on a filler into a learned model constructed by machine learning, A step of outputting application conditions for the filler from the learned model, A computer-readable recording medium recording a program for causing a computer to execute a step of giving an instruction to a filler application module to apply the filler to a gap between an edge portion of the first substrate and an edge portion of the second substrate by the filler application module according to the application conditions.

Images