Method for predicting the number of substrates that can be plated and constructing a prediction model thereof and a selection model for components that are expected to cause abnormalities

Abstract

The present invention provides a method for predicting the number of substrates that can be plated and constructing a prediction model thereof and a selection model for components that are expected to cause abnormalities, which can correctly predict the period when a substrate holder needs to be repaired. The method uses a substrate holder (18) to plate a plurality of substrates, determines the total number of substrates plated by the substrate holder (18) before the substrate holder (18) causes an abnormality, determines a first processable number of substrates and a second processable number of substrates, creates a first data set consisting of a combination of first condition data and the first processable number of substrates, the first condition data representing the state of components of the substrate holder (18), creates a second data set consisting of a combination of second condition data and the second processable number of substrates, the second condition data representing the state of the components, and uses training data containing the first data set and the second data set to optimize the parameters of the prediction model formed by a neural network.

Description

Technical Field

[0001] The present invention relates to a method for constructing a prediction model for predicting the number of substrates that can be plated before an abnormality occurs on a substrate holder used in a plating device, and more particularly to a method for constructing the prediction model by machine learning such as deep learning. In addition, the present invention also relates to a method for predicting the number of substrates that can be plated using such a prediction model. Background Art

[0002] The plating device immerses a substrate (e.g., a wafer) held by a substrate holder in a plating solution, applies a voltage between the substrate and an anode, and deposits a conductive film on the surface of the substrate. The substrate holder is composed of multiple components such as a plurality of electrical contacts for establishing electrical connection between the substrate and a power source, a seal for isolating the electrical contacts from the plating solution, and a seal holder for holding the seal.

[0003] If an abnormality is caused in the substrate holder, it will have an adverse effect on the plating of the substrate. For example, when the seal is deformed, the plating solution penetrates into the interior of the substrate holder and contacts the electrical contacts. As a result, a conductive film with a target thickness cannot be formed on the substrate. The substrate holder has a component that contacts the substrate to be plated, so the condition of the substrate holder will have a greater impact on the plating result. Therefore, in order to achieve a good plating result, the maintenance of the substrate holder is very important.

[0004] Patent Document 1: Japanese Patent Application Publication No. 2018-3102

[0005] It is preferred that the substrate holder be repaired before the substrate holder causes an abnormality. However, it is difficult to correctly determine whether the substrate holder needs repair. Therefore, in the past, there was a situation where the substrate holder was repaired after the substrate holder caused an abnormality, but in this case, it was impossible to prepare for the repair in advance, resulting in a time-consuming repair. In addition, Patent Document 1 describes a substrate holder inspection device that has a substrate holder appearance inspection device and cleans the substrate holder as needed, but it only determines whether the appearance of the substrate holder has an abnormality. Summary of the invention

[0006] Therefore, the present invention provides a method for constructing a prediction model that can correctly predict the period when a substrate holder needs maintenance. In addition, the present invention provides a method for constructing a selection model that can predict the constituent elements that cause the abnormality of the substrate holder from among a plurality of constituent elements of the substrate holder that may be the cause of the abnormality of the substrate holder. In addition, the present invention also provides a method for predicting the maintenance period of the substrate holder using such a prediction model.

[0007] In one embodiment, a method is provided for constructing a prediction model for predicting the number of substrates that can be plated before a substrate holder causes an abnormality. In the method, a plurality of substrates are plated using a substrate holder, the total number of substrates plated using the substrate holder before the substrate holder causes an abnormality is determined, the number of substrates that can be plated before the substrate holder causes an abnormality, i.e., a first processable number of substrates and a second processable number of substrates are determined, a first data set is prepared consisting of a combination of first condition data corresponding to the first processable number of substrates and the first processable number of substrates, the first condition data representing a state of a constituent element of the substrate holder, a second data set is prepared consisting of a combination of second condition data corresponding to the second processable number of substrates and the second processable number of substrates, the second condition data representing a state of the constituent element, and the parameters of the prediction model formed by a neural network are optimized using training data including the first data set and the second data set.

[0008] In one embodiment, the first processable number of sheets is 0, the first status data is defect status data indicating the state of the constituent elements of the substrate holder when the abnormality is caused, and the first data set is a defect data set consisting of a combination of the defect status data and 0.

[0009] In one embodiment, the second processable number of wafers is the number of processable wafers obtained by subtracting the number of intermediate wafers that is less than the total number of wafers from the total number of wafers, the second status data is intermediate status data representing the state of the constituent elements at the moment when the substrate is coated with the intermediate number of wafers, and the second data set is an intermediate data set composed of a combination of the intermediate status data and the second processable number of wafers.

[0010] In one embodiment, the first processable number of wafers is the processable number obtained by subtracting the first intermediate number of wafers less than the total number of wafers from the total number of wafers, the first condition data is the first intermediate condition data indicating the state of the constituent element at the moment when the substrate is coated with the first intermediate number of wafers, the first data set is the first intermediate data set consisting of a combination of the first intermediate condition data and the first processable number of wafers, the second processable number of wafers is the processable number obtained by subtracting the second intermediate number of wafers less than the first intermediate number of wafers from the total number of wafers, the second condition data is the second intermediate condition data indicating the state of the constituent element at the moment when the substrate is coated with the second intermediate number of wafers, and the second data set is the second intermediate data set consisting of a combination of the second intermediate condition data and the second processable number of wafers.

[0011] In one embodiment, the first status data and the second status data are each composed of one of image data and surface shape data of the substrate holder.

[0012] In one embodiment, the prediction model is composed of a neural network having an input layer, at least two intermediate layers, and an output layer.

[0013] In one embodiment, the prediction model is updated by repeatedly performing steps from plating a plurality of substrates using a substrate holder to optimizing parameters of the prediction model.

[0014] In one embodiment, the method further includes the following steps: preparing a selection data set, the selection data set including a numerical value set indicating that the cause of the abnormality of the substrate holder is the constituent element, reference condition data indicating the state of other constituent elements of the substrate holder when the abnormality is caused, and defect condition data indicating the state of the constituent element when the abnormality is caused, and using the selection data set to optimize the parameters of the selection model formed by the neural network.

[0015] In one embodiment, a method is provided, wherein a prediction model constructed using the method described above is prepared, the latest status data representing the status of components of a substrate holder currently in use is input into the prediction model, and a predicted number of processable wafers is output from the prediction model.

[0016] In one embodiment, the method further comprises writing the estimated number of processable sheets into an electronic tag attached to the substrate holder currently in use.

[0017] In one embodiment, a method is provided for constructing a selection model for predicting a constituent element that may cause an abnormality of the substrate holder from among a plurality of constituent elements of the substrate holder that may be the cause of an abnormality of the substrate holder. In the method, a plurality of substrates are plated before an abnormality of the first substrate holder is caused by a first constituent element of the first substrate holder, and a first selection data set is prepared. The first selection data set includes a first value set indicating that the cause of the abnormality of the first substrate holder is the first constituent element, first reference status data indicating the state of a second constituent element of the first substrate holder when the abnormality is caused, and first reference status data indicating the state of the first constituent element when the abnormality is caused. The method comprises the following steps: optimizing the parameters of a selection model constituted by a neural network using the first defect condition data of the state of a constituent element, coating a plurality of substrates before an abnormality of the second substrate holder is caused by a second constituent element of the second substrate holder, and producing a second selection data set, the second selection data set comprising a second numerical set indicating that the cause of the abnormality of the second substrate holder lies in the second constituent element, second reference condition data indicating the state of the first constituent element of the second substrate holder when the abnormality is caused, and second defect condition data indicating the state of the second constituent element when the abnormality is caused, and further optimizing the parameters using the second selection data set.

[0018] In one embodiment, the first reference condition data and the first defect condition data are respectively composed of one of the image data and the surface shape data of the first substrate holder, and the second reference condition data and the second defect condition data are respectively composed of one of the image data and the surface shape data of the second substrate holder.

[0019] In one embodiment, a method is provided, which includes the following steps: inputting the latest status data of the first component and the second component of the substrate holder representing the status of the components of the substrate holder currently in use into a selection model constructed by the method described in Technical Solution 11 or 12, and when the first confidence corresponding to the first component output from the above-mentioned selection model is higher than the second confidence corresponding to the above-mentioned second component, inputting the latest status data of the above-mentioned first component into a prediction model corresponding to the above-mentioned first component, and outputting the predicted number of processable pieces of the above-mentioned substrate holder from the above-mentioned prediction model, and the above-mentioned prediction model is a prediction model constructed by the method described in any one of Technical Solutions 1 to 10.

[0020] According to the present invention, a prediction model can be constructed that can correctly predict the number of substrates that can be plated before the substrate holder causes an abnormality. In addition, according to the present invention, a selection model can be constructed that can predict the constituent elements that cause the abnormality of the substrate holder from among a plurality of constituent elements of the substrate holder that may cause the abnormality of the substrate holder. In addition, according to the present invention, the number of substrates that can be plated before the substrate holder causes an abnormality can be correctly predicted using the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a diagram of the overall configuration of the plating equipment.

[0022] Figure 2 Yes means Figure 1 A schematic perspective view of a substrate holder is shown.

[0023] Figure 3 Yes means Figure 1 A schematic top view of a substrate holder is shown.

[0024] Figure 4 Yes means Figure 1 A schematic right side view of the substrate holder is shown.

[0025] Figure 5 yes Figure 4 Enlarged view of part A.

[0026] Figure 6 This is a table listing the types of state changes of the components of the substrate holder that may cause leakage and electrical failure.

[0027] Figure 7 is a schematic diagram showing an example of a prediction model.

[0028] Figure 8 is a flow chart showing one embodiment of a method for optimizing parameters of a prediction model.

[0029] Fig. 9 is a schematic diagram showing an example of a selection model.

[0030] Fig.10 is a flow chart illustrating one embodiment of a method for optimizing parameters of a selection model.

[0031] Fig.11 This is a flowchart showing one embodiment of a process of determining whether a substrate holder can be used.

[0032] Fig.12 This is a flowchart showing one embodiment of a process of determining whether a substrate holder can be used.

[0033] Fig.13 (a) and Fig.13 (b) is a schematic diagram showing an example of a method for photographing components of a substrate holder and a method for measuring a surface shape.

[0034] Description of Reference Numerals

[0035] 10...cassette; 12...cassette station; 14...aligner; 16...spin rinse dryer; 18...substrate holder; 20...fixing station; 22...substrate conveyor; 24...temporary storage box; 26...pre-wetting tank; 28...pre-treatment tank; 30a...first water washing tank; 30b...second water washing tank; 32...spraying tank; 34...plating tank; 36...overflow tank; 38...plating chamber; 40...substrate holder conveyor; 42...first conveyor; 44...second conveyor; 46...paddle drive; 50...track; 52...carrier; 54...first holding element; 54a...through hole; 56...hinge; 58...second holding element; 58a...opening; 60...base; 62...seal holder; 64...slide plate; 64a...convex portion; 64b...protrusion; 65... spacer; 66...seal (first sealing protrusion); 68...seal (second sealing protrusion); 69a...stopper; 69b...stopper; 70a...first fixing ring; 70b...second fixing ring; 72...pressing plate; 74...fixing clamp; 80...supporting surface; 82...protruding strip; 84...recess; 86...second electric contact; 88...first electric contact; 89...stopper; 90...arm; 91...external electric contact; 92...electric wire; 95...electronic tag; 101...photographing device; 102...three-dimensional measuring device; 110...computing system; 110a...storage device; 110b...processing device; 115...controller; 117...leakage inspection device; 118...power-on inspection device; 201, 301...input layer; 202, 302...intermediate layer; 203, 303...output layer. DETAILED DESCRIPTION

[0036] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Figure 1 This is the overall configuration diagram of the plating device. Figure 1 As shown, the plating device comprises: two cassette tables 12 carrying cassettes 10 for storing substrates such as wafers, an aligner 14 for aligning the positions of orientation planes and notches such as grooves of the substrates with a specified direction, and a spin rinse dryer 16 for rotating the substrates at high speed and drying them after plating treatment.

[0037] A fixing station 20 is provided near the spin-rinsing dryer 16, and the fixing station 20 places the substrate holder 18 and performs loading and unloading of the substrate to the substrate holder 18. A substrate conveying device 22 composed of a conveying robot for conveying the substrate between the cassette 10, the aligner 14, the spin-rinsing dryer 16, and the fixing station 20 is further provided. The fixing station 20 is provided with an imaging device 101 for imaging the substrate holder 18, and a three-dimensional measuring device 102 for measuring the surface shape of the substrate holder 18.

[0038] The imaging device 101 and the three-dimensional measuring device 102 are electrically connected to a computing system 110 that performs machine learning. The imaging device 101 and the three-dimensional measuring device 102 are configured to be able to transmit the generated image data and surface shape data to the computing system 110. The computing system 110 is composed of at least one computer. The computing system 110 includes a storage device 110a for storing the above-mentioned image data and surface shape data. The computing system 110 also includes a processing device 110b such as a CPU (central processing unit) or a GPU (graphics processing unit).

[0039] exist Figure 1 In the figure, the computing system 110 is schematically depicted. The computing system 110 can be an edge server connected to the plating device by a communication line, a cloud server connected to the plating device through a network such as the Internet, or a fog computing device (gateway, fog server, router, etc.) arranged in a network connected to the plating device. The computing system 110 can also be a combination of multiple servers (computers). For example, the computing system 110 can also be a combination of an edge server configured near the plating device and a cloud server away from the plating device. The multiple servers (computers) constituting the computing system 110 can be interconnected via a network such as the Internet, or they may not be interconnected.

[0040] Furthermore, the following are arranged in order: a temporary storage box 24 for storing and temporarily placing the substrate holder 18; a pre-wetting tank 26 for hydrophilizing the surface of the substrate; a pre-treatment tank 28 for etching away the oxide film on the surface of the conductive film such as the seed layer formed on the surface of the substrate; a first water washing tank 30a for washing the substrate after the pre-treatment; a blowing tank 32 for draining the washed substrate; a second water washing tank 30b for washing the substrate after plating; and a plating tank 34. The plating tank 34 is composed of a plurality of plating chambers 38 housed inside an overflow tank 36, each of which houses one substrate, and performs copper plating, metal plating (tin, gold, silver, nickel, ruthenium, indium plating), and alloy plating (tin / silver alloy, tin / indium alloy, etc.).

[0041] Furthermore, the plating device includes a substrate holder conveying device 40, for example, a linear motor method, which conveys the substrate holder 18 together with the substrate. The substrate holder conveying device 40 includes: a first conveyor 42 for conveying the substrate between the fixed station 20, the temporary storage box 24, and the pre-wetting tank 26; and a second conveyor 44 for conveying the substrate between the temporary storage box 24, the pre-wetting tank 26, the pre-treatment tank 28, the first water washing tank 30a, the second water washing tank 30b, the blowing tank 32, and the plating tank 34. It is also possible to have only the first conveyor 42 without the second conveyor 44. In this case, the first conveyor 42 is configured to be able to convey the substrate between the fixed station 20, the temporary storage box 24, the pre-wetting tank 26, the pre-treatment tank 28, the first water washing tank 30a, the second water washing tank 30b, the blowing tank 32, and the plating tank 34.

[0042] In addition, the plating apparatus includes a controller 115. The imaging device 101, the three-dimensional measuring device 102, the computing system 110, the fixing station 20, the substrate conveying device 22, and the substrate holder conveying device 40 are electrically connected to the controller 115. The operations of the imaging device 101, the three-dimensional measuring device 102, the fixing station 20, the substrate conveying device 22, and the substrate holder conveying device 40 are controlled by the controller 115. The computing system 110 transmits the prediction result of the maintenance period to the controller 115, and the controller 115 controls the imaging device 101, the three-dimensional measuring device 102, and the substrate holder conveying device 40 according to the prediction result of the maintenance period.

[0043] A paddle driving device 46 is disposed adjacent to the overflow tank 36 of the plating tank 34 to drive a paddle (not shown) that is located inside each plating chamber 38 and serves as a stirring rod for stirring the plating solution.

[0044] The fixed station 20 includes a loading plate 52 that is slidable in the horizontal direction along the rail 50. Two substrate holders 18 are placed side by side in a horizontal state on the loading plate 52, and after the substrate is transferred between the one substrate holder 18 and the substrate conveyor 22, the loading plate 52 is slid in the horizontal direction, and the substrate is transferred between the other substrate holder 18 and the substrate conveyor 22.

[0045] like Figure 2 to Figure 5 As shown, the substrate holder 18 includes: a first holding element (base holding element) 54 made of, for example, polyvinyl chloride and in the shape of a rectangular flat plate; and a second holding element (movable holding element) 58 mounted to the first holding element 54 so as to be openable and closable via a hinge 56. In this example, an example in which the second holding element 58 is configured to be openable and closable via the hinge 56 is shown, but, for example, the second holding element 58 may be arranged at a position opposite to the first holding element 54 and opened and closed by advancing the second holding element 58 toward the first holding element 54.

[0046] The second holding element 58 has a base 60 and a seal holder 62. The seal holder 62 is made of, for example, polyvinyl chloride, and promotes sliding with a slide plate 64 described below. A seal (first seal protrusion) 66 is installed on the upper surface of the seal holder 62 in a manner protruding inward. When the substrate W is held by the substrate holder 18, the seal 66 is pressed against the outer peripheral portion of the surface of the substrate W to seal the gap between the substrate W and the second holding element 58. In addition, a seal (second seal protrusion) 68 is installed on the surface of the seal holder 62 that is opposite to the first holding element 54. When the substrate W is held by the substrate holder 18, the seal 68 is pressed against the first holding element 54 to seal the gap between the first holding element 54 and the second holding element 58. The seal 68 is located outside the seal 66.

[0047] The seal (first sealing protrusion) 66 and the seal (second sealing protrusion) 68 are endless seals. The seal 66 and the seal 68 can also be sealing elements such as O-rings. In one embodiment, the second retaining element 58 including the seal 66 and the seal 68 itself can be composed of a material having a sealing function. In this embodiment, the seal 66 and the seal 68 are annular and are arranged concentrically. When plating the substrate W, the substrate holder 18 holding the substrate W is arranged in a vertical posture in the plating chamber 38. When the substrate holder 18 is arranged in a horizontal posture in the plating chamber, the seal 68 can also be omitted.

[0048] like Figure 5 As shown, the seal 66 is mounted on the seal holder 62 by being sandwiched between the seal holder 62 and the first fixing ring 70a. The first fixing ring 70a is mounted on the seal holder 62 via a stopper 69a such as a bolt. The seal 68 is mounted on the seal holder 62 by being sandwiched between the seal holder 62 and the second fixing ring 70b. The second fixing ring 70b is mounted on the seal holder 62 via a stopper 69b such as a bolt.

[0049] A step portion is provided on the outer periphery of the seal holder 62 of the second holding element 58, and a slide plate 64 is rotatably mounted on the step portion via a spacer 65. The slide plate 64 is supported by a pressure plate 72 ( Figure 3 The slide plate 64 is made of a material such as titanium that is excellent in corrosion resistance to acids and alkalis and has sufficient rigidity. The spacer 65 is made of a material with a low friction coefficient such as PTFE (polytetrafluoroethylene) so that the slide plate 64 can rotate smoothly.

[0050] Located outside the slide plate 64 and in the first holding element 54, inverted L-shaped fixing clips 74 having protrusions protruding inward are erected at equal intervals along the circumferential direction. In addition, a protrusion 64b protruding outward is provided at a position along the circumferential direction of the slide plate 64 and opposite to the fixing clip 74. Moreover, the lower surface of the inner protrusion of the fixing clip 74 and the upper surface of the protrusion 64b of the slide plate 64 are tapered surfaces inclined in opposite directions to each other along the rotation direction. Protrusions 64a protruding upward are provided at multiple locations (for example, 3 locations) along the circumferential direction of the slide plate 64. Thus, the slide plate 64 can be rotated by rotating the rotating pin (not shown) of the fixing station 20 and pushing the protrusion 64a from the side.

[0051] With the second holding element 58 opened, the substrate W is placed in the center of the first holding element 54. Next, the second holding element 58 is closed via the hinge 56, and the slide plate 64 is rotated clockwise and the protrusion 64b of the slide plate 64 is slid into the inner protrusion of the fixing clip 74, so that the first holding element 54 and the second holding element 58 are clamped and locked to each other via the tapered surfaces provided on the slide plate 64 and the fixing clip 74, and the lock can be released by rotating the slide plate 64 counterclockwise and removing the protrusion 64b of the slide plate 64 from the inverted L-shaped fixing clip 74.

[0052] When the second holding element 58 is locked in this way (i.e., when the substrate holder 18 holds the substrate W), the lower end of the lower protrusion on the inner peripheral surface side of the seal 66 is uniformly pressed against the outer peripheral portion of the surface of the substrate W, and the seal 66 seals the gap between the second holding element 58 and the outer peripheral portion of the surface of the substrate W. Similarly, the lower end of the lower protrusion on the outer peripheral side of the seal 68 is uniformly pressed against the surface of the first holding element 54, and the seal 68 seals the gap between the first holding element 54 and the second holding element 58.

[0053] The substrate holder 18 holds the substrate W by sandwiching the substrate W between the first holding element 54 and the second holding element 58. The second holding element 58 has a circular opening 58a. The opening 58a is slightly smaller than the size of the substrate W. When the substrate W is sandwiched between the first holding element 54 and the second holding element 58, the processed surface of the substrate W is exposed through the opening 58a. Therefore, various processing liquids such as the pre-wetting liquid, the pre-processing liquid, and the plating liquid described later can contact the exposed surface of the substrate W held by the substrate holder 18. The exposed surface of the substrate W is surrounded by a seal (first sealing protrusion) 66.

[0054] When the substrate W is held by the substrate holder 18, as shown in FIG. Figure 5As shown in FIG. 1 , an internal space R1 is formed inside the substrate holder 18, the inner peripheral side of which is sealed by the seal 66 and the outer peripheral side of which is sealed by the seal 68. A protrusion 82 is provided at the center of the first holding element 54. The protrusion 82 protrudes in an annular shape in accordance with the size of the substrate W, abuts against the outer peripheral portion of the substrate W, and has a supporting surface 80 for supporting the substrate W. A recess 84 is provided at a predetermined position along the circumferential direction of the protrusion 82.

[0055] Moreover, if Figure 3 As shown, a plurality of (12 in the figure) second electric contacts 86 are arranged in each recess 84, and these second electric contacts 86 are respectively connected to a plurality of electric wires 92 extending from external electric contacts 91 provided on the arm 90. When the substrate W is placed on the support surface 80 of the first holding element 54, the end of the second electric contact 86 is exposed on the surface of the first holding element 54 in an elastic state at the side of the substrate W, and is connected to the first holding element 54. Figure 5 The lower contact of the first electrical contact 88 is shown.

[0056] The first electric contact 88 electrically connected to the second electric contact 86 is fixed to the seal holder 62 of the second holding element 58 via a stopper 89 such as a bolt. The first electric contact 88 has a leaf spring shape. The first electric contact 88 is located outside the seal 66, and has a contact portion protruding in the shape of a leaf spring inside. The contact portion has elasticity and can be easily bent due to its elastic force. When the substrate W is held by the first holding element 54 and the second holding element 58, the contact portion of the first electric contact 88 is elastically in contact with the outer peripheral surface of the substrate W supported on the support surface 80 of the first holding element 54.

[0057] The second holding element 58 is opened and closed by an air cylinder (not shown) and the self-weight of the second holding element 58. That is, a through hole 54a is provided in the first holding element 54, and an air cylinder is provided at a position opposite to the through hole 54a when the substrate holder 18 is placed on the fixing station 20. Thus, by extending the piston rod, the seal holder 62 of the second holding element 58 is pushed upward through the through hole 54a by a pressing rod (not shown), thereby opening the second holding element 58, and by contracting the piston rod, the second holding element 58 is closed by its own weight.

[0058] A pair of substantially T-shaped arms 90 are provided at the end of the first holding element 54 of the substrate holder 18, which serve as a support portion when the substrate holder 18 is transported or suspended. In the temporary storage box 24, the arms 90 are hooked on the upper surface of the peripheral wall of the temporary storage box 24, so that the substrate holder 18 is vertically suspended. The arms 90 of the suspended substrate holder 18 are held by the conveyor 42 or 44 of the substrate holder transport device 40 to transport the substrate holder 18. In addition, in the pre-wetting tank 26, the pre-treatment tank 28, the first water washing tank 30a, the second water washing tank 30b, the blowing tank 32, and the plating tank 34, the substrate holder 18 is also suspended from the peripheral walls thereof via the arms 90.

[0059] The substrate W used in this embodiment is a circular substrate such as a wafer, but the present invention can also be applied to a square substrate. Each component of the substrate holder for holding a square substrate has a shape suitable for the shape of the substrate. For example, the above-mentioned opening 58a becomes a square opening smaller than the size of the entire square substrate. Other components such as seals 66 and 68 are also formed into a shape suitable for the shape of the square substrate. The shapes of other components can also be appropriately changed within the scope of the above-mentioned technical ideas.

[0060] As described above, the substrate holder 18 is a composite assembly consisting of a plurality of components, such as the seals 66, 68, the first electric contact 88, the second electric contact 86, the external electric contact 91, the seal holder 62, and the first holding element 54. The above-mentioned components may be deformed or corroded as the substrate holder 18 is used to plate a plurality of substrates. For example, when the first electric contact 88 is deformed, it is impossible to make an appropriate current flow to the substrate. In other examples, when the seals 66, 68 are deformed, the plating solution may penetrate into the internal space R1 of the substrate holder 18, causing the plating solution to contact the electric contacts 86, 88. As a result, poor plating is caused. Therefore, it is important to perform maintenance on the substrate holder 18 before poor plating is caused.

[0061] In the present embodiment, a prediction model constructed by machine learning is used to predict the maintenance period of the substrate holder 18. The prediction model is a model for predicting the number of substrates that can be plated using the substrate holder 18 before the substrate holder 18 causes an abnormality. In this specification, machine learning is learning using a neural network, and deep learning is included in machine learning. Machine learning is performed by an operation system 110 composed of at least one computer. The operation system 110 includes: a storage device 110a storing a program for causing the operation system 110 to perform machine learning; and a processing device 110b that performs operations based on the program.

[0062] The state of each component of the substrate holder 18 gradually changes as the plating of the substrate is repeatedly performed. The change in the state of each component may become the cause of the abnormality of the substrate holder 18. Examples of abnormalities of the substrate holder 18 include leakage and poor power supply. Leakage is caused by insufficient sealing function, resulting in the plating liquid infiltrating into the internal space R1 of the substrate holder 18. Poor power supply is the absence of the desired current flowing in the substrate held in the substrate holder 18. When leakage or poor power supply occurs, the plating device cannot form a film of the desired thickness on the surface of the substrate.

[0063] There are various reasons for leakage and poor electrical conduction. Figure 6 1 is a table listing the types of state changes of the components of the substrate holder 18 that may cause leakage and electrical failure. Figure 6 In the figure, the circle symbol (○) indicates that the state change of the component is likely to cause leakage or poor electrical conduction, and the triangle symbol (Δ) indicates that it is unclear whether the state change of the component will cause leakage or poor electrical conduction. Figure 6 The table shows data types used to detect each state change. For example, the discoloration of the seal holder 62 is a state change that may cause leakage, and the discoloration of the seal holder 62 is detected based on the image data. The deformation of the electrical contacts 86 and 88 is a state change that may cause a malfunction in the electrical conduction, and the deformation of the electrical contacts 86 and 88 is detected based on the surface shape data.

[0064] Specific examples of the state change of each component are as follows. However, the state change of each component is not limited to the following specific examples.

[0065] Seal holder 62: discoloration, deformation, crystal residue

[0066] First holding element 54: discoloration, deformation, and residual crystals

[0067] First electrical contact 88: deformation, peeling of the gold (Au) surface film, metal precipitation, crystallization of copper sulfate contained in the plating solution

[0068] Second electrical contact 86: deformation, peeling of the gold surface film, metal precipitation, crystallization of copper sulfate contained in the plating solution

[0069] External electrical contact 91: deformation, peeling of the gold surface film, metal precipitation, crystallization of copper sulfate contained in the plating solution

[0070] Seals 66, 68: deformation, discoloration

[0071] Skateboard 64: Transformation

[0072] Specific examples of deformation of the above-mentioned components include deformation due to application of external force, strain due to internal stress, and corrosion of the components.

[0073] At least one of the image data and the surface shape data of each component is used to construct a prediction model for predicting abnormality of the substrate holder 18. For example, the surface shape data is used to construct a prediction model for predicting abnormality of the substrate holder 18 based on deformation of the seals 66 and 68, and the image data is used to construct a prediction model for predicting abnormality of the substrate holder 18 based on discoloration of the seals 66 and 68.

[0074] Before holding the substrate to be plated, the image data and surface shape data of each component of the substrate holder 18 are generated by the imaging device 101 and the three-dimensional measuring device 102. The computing system 110 obtains the image data and the surface shape data from the imaging device 101 and the three-dimensional measuring device 102, and stores the image data and the surface shape data in the storage device 110a. In one embodiment, the image data and the surface shape data of each component of the substrate holder 18 can also be generated after the plated substrate is taken out from the substrate holder 18.

[0075] The imaging device 101 is a camera having an image sensor such as a CCD or a CMOS. The three-dimensional measuring device 102 is a device capable of measuring the surface shape of an object, for example, using a laser displacement meter. More specifically, the three-dimensional measuring device 102 measures the positions of a plurality of measurement points on the surface of a component, and outputs the X coordinate, Y coordinate, and Z coordinate of each measurement point as a measured value of the position.

[0076] The computing system 110 counts the cumulative number of substrates plated using a certain substrate holder 18 from the beginning of using the substrate holder 18. Specifically, each time a substrate is plated using the substrate holder 18, the computing system 110 counts the cumulative number of substrates plated using the substrate holder 18. Furthermore, each time image data and surface shape data are obtained from the imaging device 101 and the three-dimensional measuring device 102, the computing system 110 associates the image data and the surface shape data with the current cumulative number of plated substrates. Furthermore, the computing system 110 stores the image data and the surface shape data in the storage device 110a in combination with the current cumulative number of plated substrates.

[0077] After holding the substrate to be plated, the substrate holder 18 is connected to the leakage inspection device 117. The leakage inspection device 117 checks whether the seals 66 and 68 of the substrate holder 18 function normally. The leakage inspection device 117 forms a positive pressure or a negative pressure in the internal space R1 formed in the substrate holder 18 by the seals 66 and 68. When the pressure (positive pressure or negative pressure) of the internal space R1 exceeds the allowable value and changes within a specified time, an alarm signal indicating that the substrate holder 18 has caused an abnormality is issued. In the case where the pressure in the internal space R1 changes greatly, it is possible that the plating liquid penetrates into the internal space R1 during the plating of the substrate. This means that an abnormality (i.e., liquid leakage) has occurred in the substrate holder 18. The leakage inspection device 117 is electrically connected to the controller 115.

[0078] In addition, the substrate holder 18 is connected to the current inspection device 118 while holding the substrate to be plated. The current inspection device 118 allows a predetermined current to flow through the substrate via the external electrical contact 91 of the substrate holder 18, and measures the internal resistance of the substrate holder 18. The internal resistance of the substrate holder 18 is a composite resistance of the external electrical contact 91, the wire 92, the first electrical contact 88, the second electrical contact 86, and the substrate held by the substrate holder 18. The internal resistance of the substrate holder 18 may change depending on the contact state between the first electrical contact 88 and the substrate, and the contact state between the first electrical contact 88 and the second electrical contact 86. For example, when the first electrical contact 88 corrodes, the contact resistance between the first electrical contact 88 and the substrate changes. As a result, the internal resistance of the substrate holder 18 changes.

[0079] When the measured value of the internal resistance exceeds the predetermined setting range, the power-on inspection device 118 issues an alarm signal indicating that the substrate holder 18 has caused an abnormality. When the measured value of the internal resistance exceeds the setting range, it is inferred that at least one of the external electrical contact 91, the first electrical contact 88, and the second electrical contact 86 has an abnormality. As a result, a film of a desired thickness cannot be deposited on the substrate. This means that an abnormality has occurred in the substrate holder 18 (i.e., a power-on failure). The power-on inspection device 118 is electrically connected to the controller 115.

[0080] In one embodiment, the internal resistance of the substrate holder 18 may be measured for a substrate holder 18 that holds a dummy substrate instead of a substrate to be plated. Examples of dummy substrates include blanket boards with no pattern formed on the surface, such as substrates with a conductive film such as copper on the surface.

[0081] The image data and the surface shape data are sent to the operation system 110 and stored in the storage device 110a. The storage device 110a stores a plurality of prediction models for predicting the number of substrates that can be plated before the substrate holder 18 causes an abnormality. When any one of the image data and the surface shape data is input to each prediction model, the prediction model outputs the predicted number of substrates that can be plated using the substrate holder 18.

[0082] A plurality of prediction models are provided at least corresponding to a plurality of constituent elements of the substrate holder 18 (seal holder 62, electrical contacts 86, 88, seals 66, 68, etc.). Furthermore, a plurality of prediction models are provided for each type of abnormality of the substrate holder 18 (leakage, poor electrical conduction). This is because the predicted number of substrates that can be plated using the substrate holder 18 may vary for each constituent element of the substrate holder 18, and may vary for each type of abnormality of the substrate holder 18. In one embodiment, a plurality of prediction models may also be provided for each type of state change of the constituent elements of the substrate holder 18. That is, a plurality of prediction models may also be provided. Figure 6 The number of prediction models of the circle symbols (○) and triangle symbols (Δ) shown in the table is . In one embodiment, only one prediction model may be provided for the substrate holder 18 .

[0083] The plurality of prediction models are models each composed of a neural network. The operation system 110 constructs a prediction model by learning parameters (weights, etc.) of each prediction model using training data including at least one pre-selected from image data and surface shape data. The parameters of the prediction model include bias in addition to weights.

[0084] In this embodiment, the data representing the state of each component of the substrate holder 18 is composed of image data and surface shape data. In one embodiment, as the data representing the state of each component of the substrate holder 18, only one of the image data and the surface shape data may be used.

[0085] Figure 7 is a schematic diagram showing an example of a prediction model. Figure 7 As shown, the prediction model is a neural network having an input layer 201 , a plurality of intermediate layers (also called hidden layers) 202 , and an output layer 203 . Figure 7 The prediction model shown has four intermediate layers 202, but the structure of the prediction model is not limited to Figure 7 Example shown. Machine learning performed using a neural network having multiple intermediate layers 202 is also referred to as deep learning.

[0086] In the prediction model using image data, the numerical values ​​representing the red, green, and blue colors of each pixel constituting the image data are input to the input layer 201. In the prediction model using surface shape data, the X-coordinate value, Y-coordinate value, and Z-coordinate value representing the position of the measurement point on the surface of the constituent element of the substrate holder 18 are input to the input layer 201. In either case, the output layer 203 outputs the number of substrates that can be plated before abnormality occurs in the substrate holder 18. In the following description, the number of substrates output from the prediction model is referred to as the predicted processable number of substrates.

[0087] The operation system 110 optimizes the parameters (weights, etc.) of the prediction model through machine learning using the training data, thereby improving the accuracy of the prediction model. Figure 8 is a flow chart showing one embodiment of a method for optimizing parameters of a prediction model. In step 1, a plurality of substrates are plated using a new substrate holder 18. The substrates are plated by Figure 1 The plating device shown in the figure is used for execution. In the example of "new substrate holder 18", in addition to unused substrate holders, it also includes substrate holders that have been repaired. Image data and surface shape data are generated by the imaging device 101 and the three-dimensional measuring device 102 before each substrate is held in the substrate holder 18. The computing system 110 obtains the image data and surface shape data from the imaging device 101 and the three-dimensional measuring device 102, and stores them together with the cumulative number of substrates that have been plated in the storage device 110a of the computing system 110.

[0088] The above-mentioned new substrate holder 18 is used to plate a plurality of substrates until the substrate holder 18 causes an abnormality. The user can know that the substrate holder 18 causes an abnormality based on the alarm signal sent from the leakage inspection device 117 or the power-on inspection device 118. When the substrate holder 18 causes an abnormality, the user takes out the substrate holder 18 from the plating device and decomposes the substrate holder 18 to determine the cause of the abnormality of the substrate holder 18. In addition, the user uses an input device and a communication device not shown in the figure to provide the information that the substrate holder 18 causes an abnormality due to a change in the state of the constituent element to the operation system 110. Specifically, the user teaches the operation system 110 the constituent element that should be used to construct (learn) the prediction model. For example, in the case where the substrate holder 18 causes an abnormality due to the deformation of the first electrical contact 88, the user provides the information that the constituent element that should be used to construct (learn) the prediction model is the first electrical contact 88 to the operation system 110.

[0089] In step 2, the operation system 110 determines the total number of substrates plated using the substrate holder 18 before the above-mentioned abnormality occurs, and stores the determined total number of substrates in the storage device 110a. The total number of substrates is the cumulative number of substrates plated using the substrate holder 18 from the start of use of the substrate holder 18 until the abnormality occurs on the substrate holder 18. The total number of substrates is equivalent to the latest cumulative number stored in the storage device 110a of the operation system 110. And, the number of substrates that can be plated using the substrate holder 18 when the abnormality occurs is 0.

[0090] In step 3, the operation system 110 determines defect status data indicating the state of the constituent elements of the substrate holder 18 that is the cause of the abnormality of the substrate holder 18. The defect status data is status data indicating the state of the constituent elements of the substrate holder 18 that caused the abnormality of the substrate holder 18, and more specifically, status data indicating the state of the above-mentioned constituent elements when the abnormality of the substrate holder 18 was caused. The defect status data constitutes a part of the image data and surface shape data stored in the storage device 110a. The defect status data is composed of one of the image data and surface shape data of the constituent elements. For example, in the case where the abnormality of the substrate holder 18 is caused by the deformation of the first electric contact 88, the defect status data is the image data of the first electric contact 88 when the abnormality of the substrate holder 18 is caused, that is, the latest image data of the first electric contact 88.

[0091] In step 4, the computing system 110 creates a defect data set consisting of a combination of defect status data and 0. The value 0 is the number of substrates that can be plated using the substrate holder 18, that is, the number of processable substrates. The defect status data is status data corresponding to the value 0.

[0092] In step 5, the operation system 110 determines the processable number of sheets by subtracting the number of intermediate sheets less than the total number of sheets from the total number of sheets obtained in step 2. For example, when the total number of sheets of the substrate when the substrate holder 18 causes an abnormality is 500 sheets and the number of intermediate sheets is 200 sheets, the processable number of sheets is 300 sheets (500 sheets - 200 sheets).

[0093] In step 6, the operation system 110 determines the intermediate status data corresponding to the processable number of sheets determined in step 5. The intermediate status data is status data indicating the state of the constituent elements of the substrate holder 18 which is the cause of the abnormality of the substrate holder 18, and more specifically, status data indicating the state of the above constituent elements at the moment when the above intermediate number of substrates are plated. Similar to the defect status data, the intermediate status data constitutes a part of the image data and the surface shape data stored in the storage device 110a. The intermediate status data is data of the same type as the defect status data, which is composed of one of the image data and the surface shape data of the constituent elements. For example, when the defect status data is the surface shape data, the intermediate status data is also the surface shape data.

[0094] In step 7 , the computing system 110 creates an intermediate data set consisting of a combination of the intermediate status data and the processable number of slices determined in step 5 above.

[0095] The above steps 5, 6, and 7 are repeated a specified number of times. More specifically, the operation system 110 changes the number of intermediate sheets, and repeatedly determines the number of processable sheets and produces intermediate data sets, thereby producing multiple intermediate data sets. For example, when the total number of plated substrates when the substrate holder 18 causes an abnormality is 500 sheets, the number of intermediate sheets is set to 100 sheets, 200 sheets, 300 sheets, and 400 sheets. Preferably, the above-mentioned multiple intermediate sheet numbers are evenly distributed between 0 and the total number of sheets. In this embodiment, the processable number of sheets is determined to be 400 sheets (500 sheets-100 sheets), 300 sheets (500 sheets-200 sheets), 200 sheets (500 sheets-100 sheets), and 100 sheets (500 sheets-400 sheets).

[0096] The operation system 110 determines intermediate status data corresponding to 400 processable sheets (i.e., intermediate status data of the constituent elements when 100 substrates are plated), intermediate status data corresponding to 300 processable sheets (i.e., intermediate status data of the constituent elements when 200 substrates are plated), intermediate status data corresponding to 200 processable sheets (i.e., intermediate status data of the constituent elements when 300 substrates are plated), and intermediate status data corresponding to 100 processable sheets (i.e., intermediate status data of the constituent elements when 400 substrates are plated).

[0097] Furthermore, the operation system 110 creates an intermediate data set consisting of a combination of intermediate status data of the constituent elements when 100 substrates are plated and 400 processable sheets, an intermediate data set consisting of a combination of intermediate status data of the constituent elements when 200 substrates are plated and 300 processable sheets, an intermediate data set consisting of a combination of intermediate status data of the constituent elements when 300 substrates are plated and 200 processable sheets, and an intermediate data set consisting of a combination of intermediate status data of the constituent elements when 400 substrates are plated and 100 processable sheets.

[0098] In step 8, the computing system 110 generates training data including the defective data set and the plurality of intermediate data sets. In one embodiment, the training data may not include the defective data set. In this case, the training data only includes the plurality of intermediate data sets.

[0099] In step 9, the operation system 110 uses the training data to construct a prediction model composed of a neural network. More specifically, the operation system 110 uses the training data and optimizes the parameters (weights, etc.) of the prediction model through deep learning. Defect condition data (e.g., image data of a discolored seal retainer 62) is input to the input layer 201 of the prediction model. The operation system 110 determines the optimal parameters of the prediction model that can minimize the difference between the output value from the output layer 203 and 0. Similarly, the intermediate condition data is input to the input layer 201 of the prediction model. The operation system 110 determines the optimal parameters of the prediction model that can minimize the difference between the output value from the output layer 203 and the number of processable pieces (e.g., 300). In this way, the operation system 110 uses the training data to perform deep learning and optimizes the parameters of the prediction model. The prediction model constructed by deep learning is stored in the storage device 110a of the operation system 110.

[0100] Before inputting the defect condition data and the intermediate condition data into the prediction model, the defect condition data and the intermediate condition data may be pre-processed. Specifically, the operation system 110 reduces the capacity of the defect condition data and the intermediate condition data by deleting a portion of the defect condition data and a portion of the intermediate condition data that are not helpful in predicting the number of substrates that can be processed. For example, the seals 66 and 68 are annular, so the operation system 110 deletes the image data of the area inside the seals 66 and 68 from the image data sent from the self-photographing device 101. Through such pre-processing, the capacity of the defect condition data and the intermediate condition data can be reduced, and the load on the operation system 110 and the learning speed can be improved.

[0101] All the processes from step 1 to step 9 are repeated multiple times, and the prediction model is updated using new training data. That is, a new substrate holder is prepared (for example, another substrate holder having the same structure as the above-mentioned substrate holder 18, or a substrate holder 18 that has been repaired), and before the new substrate holder causes an abnormality, the new substrate holder is used to plate multiple substrates. The operation system 110 creates new training data and further optimizes the parameters of the above-mentioned prediction model to update the prediction model. The constituent elements that cause the abnormality of the new substrate holder are the same as the constituent elements that cause the abnormality of the substrate holder 18 previously used for substrate plating. The types of defect condition data and intermediate condition data contained in the new training data are also the same as the types of defect condition data and intermediate condition data contained in the previously created training data.

[0102] Similarly, multiple prediction models corresponding to multiple components of the substrate holder that may be the cause of the abnormality are constructed. Specifically, multiple training data are prepared using multiple substrate holders having the same structure, and the parameters of the multiple prediction models are optimized using the multiple training data. The operation system 110 stores the prediction models in the storage device 110a. Multiple prediction models can also be constructed. Figure 6 The numbers of circle symbols (○) and triangle symbols (Δ) in the table are shown. In one embodiment, only one prediction model may be constructed.

[0103] Figure 1 The plating device shown in the figure uses another substrate holder having the same structure as the substrate holder 18 used in the construction of the prediction model to plate the substrate. Before plating the substrate, the imaging device 101 and the three-dimensional measuring device 102 generate image data and surface shape data of each component of the substrate holder, and the power inspection device 118 measures the internal resistance of the substrate holder. The image data and surface shape data are sent to the computing system 110 and stored in the storage device 110a.

[0104] The operation system 110 uses all the prediction models (learning completed models) that have been prepared to predict the number of substrates that can be plated before the substrate holder currently used by the plating device causes an abnormality. That is, the operation system 110 inputs the latest status data representing the status of each component of the substrate holder currently in use into all the prediction models. The latest status data is composed of image data or surface shape data of each component of the substrate holder currently in use. Each prediction model outputs the number of substrates that can be plated before the substrate holder currently used by the plating device causes an abnormality, that is, the predicted number of processable pieces.

[0105] In this embodiment, the operation system 110 uses all prediction models to determine the predicted processable number of slices, so the multiple predicted processable numbers of slices obtained may be different. Therefore, the operation system 110 selects the smallest predicted processable number of slices from the multiple predicted processable numbers of slices.

[0106] In the plating apparatus, a plurality of substrate holders having the same structure are used. Each of these substrate holders is equipped with an electronic tag using RFID (radio frequency identification) technology (see Figure 3 95 in the figure). The electronic tag is called an RFID tag or an RF tag. The electronic tag has a circuit that can send and receive information in a non-contact manner by communicating with a reader and a writer (not shown). The identification number of the substrate holder on which the electronic tag is installed is pre-written in the electronic tag. The computing system 110 transmits the predicted number of processable sheets (i.e., the above-mentioned minimum predicted number of processable sheets) to the writer, and the writer writes the predicted number of processable sheets into the electronic tag of the substrate holder.

[0107] As the substrate holder is used for substrate plating, the predicted number of processable sheets output from the prediction model decreases. Whenever the computing system 110 outputs multiple predicted numbers of processable sheets from all the prediction models, it determines the minimum predicted number of processable sheets and transmits the minimum predicted number of processable sheets to the writer. The writer rewrites the predicted number of processable sheets already written into the electronic tag of the substrate holder.

[0108] The reader reads the predicted number of processable sheets from the electronic tag of the substrate holder, and displays the predicted number of processable sheets on a display device (not shown). In this way, the user can know the predicted number of processable sheets of each substrate holder. As a result, each substrate holder can be repaired at an appropriate time.

[0109] The electronic tag has the identification number of the substrate holder and the predicted number of processable sheets written on it. Therefore, even after the substrate holder is taken out of the plating device, the predicted number of processable sheets inherent to the substrate holder can be known without referring to the predicted number of processable sheets stored in the computing system 110 or the controller 115, making it easy to manage the operation of the substrate holder.

[0110] In addition to the above-mentioned multiple prediction models, the operation system 110 also has a selection model for predicting the constituent elements that may cause the abnormality of the substrate holder 18 from among the multiple constituent elements of the substrate holder 18 that may be the cause of the abnormality of the substrate holder 18. The selection model is constructed by calculating the confidence for each constituent element. The confidence is an index value that represents the probability that each constituent element causes the abnormality of the substrate holder 18, and is represented by a numerical value from 0 to 100. The confidence can also be represented by a numerical value from 0 to 1. The confidence is sometimes also referred to as a score. The predicted number of processable pieces output from the prediction model constructed for the constituent element with high confidence is a predicted number of processable pieces with high reliability.

[0111] In the present embodiment, two selection patterns are provided corresponding to two types of abnormalities of the substrate holder 18 (liquid leakage and electrical failure). Fig. 9 is a schematic diagram showing an example of a selection model. Fig. 9 As shown, the selection model is a neural network having an input layer 301 , a plurality of intermediate layers (also referred to as hidden layers) 302 , and an output layer 303 . Fig. 9 The selection model shown has four intermediate layers 302, but the structure of the selection model is not limited to Fig. 9 Example shown.

[0112] Data representing the states of the plurality of components of the substrate holder 18 when the substrate holder 18 causes an abnormality is input to the input layer 301 of the selection model. Figure 6 The image data and surface shape data of the seal holder 62, the first holding element 54, the first electric contact 88, the second electric contact 86, the external electric contact 91, the seals 66, 68, and the slide plate 64 recorded in the table are input to the input layer 301 of the selection model. The output layer 303 of the selection model outputs a plurality of confidences corresponding to the plurality of constituent elements. The output layer 303 outputs the confidences that are the same as or greater than the number of constituent elements of the substrate holder 18 that may be the cause of the abnormality of the substrate holder 18.

[0113] The operation system 110 optimizes the parameters (weights, etc.) of the selection model through deep learning using the training data, thereby constructing the selection model. Fig.10 is a flow chart illustrating one embodiment of a method for optimizing parameters of a selection model.

[0114] In step 1, a plurality of substrates are plated using a new substrate holder 18. The plating of a plurality of substrates using the new substrate holder 18 is carried out until an abnormality occurs in the substrate holder 18. Examples of the "new substrate holder 18" include not only unused substrate holders but also repaired substrate holders.

[0115] When the substrate holder 18 causes an abnormality, the user removes the substrate holder 18 from the plating device, disassembles the substrate holder 18, and determines the cause of the abnormality of the substrate holder 18. In addition, the user uses an input device and a communication device (not shown) to provide the operation system 110 with information that determines that the substrate holder 18 has caused an abnormality due to a change in the state of the Nth component. Specifically, the user teaches the operation system 110 the Nth component that should be used to construct (learn) the selection model. The Nth component is one of a plurality of components that may be the cause of the abnormality of the substrate holder 18. Figure 6 In the example shown, the Nth component is any one of the seal holder 62, the first holding element 54, the first electric contact 88, the second electric contact 86, the external electric contact 91, the seals 66, 68, and the slide plate 64. For example, when the substrate holder 18 is abnormal due to deformation of the first electric contact 88, the user provides the computing system 110 with information that the component to be selected for model construction (learning) is the first electric contact 88.

[0116] In step 2, the operation system 110 determines defect status data indicating the state of the Nth component that caused the abnormality of the substrate holder 18. More specifically, the defect status data is status data indicating the state of the Nth component when the abnormality of the substrate holder 18 was caused. The defect status data is composed of one of image data and surface shape data of the Nth component.

[0117] In step 3, the operation system 110 creates a selection data set, which includes: a value set indicating that the cause of the abnormality of the substrate holder 18 is the Nth component; reference status data indicating the state of other components of the substrate holder 18 when the abnormality of the substrate holder 18 is caused; and the above-mentioned defect status data. The other components of the substrate holder 18 are components that may become the cause of the abnormality of the substrate holder 18 and are components other than the Nth component.

[0118] The numerical value set indicating that the cause of the abnormality of the substrate holder 18 is the Nth constituent element is composed of a combination of 100 and 0 indicating the degree of certainty. The numerical value set is a numerical value set indicating the degree of certainty that the change in the state of the Nth constituent element is the cause of the abnormality determined by the user in step 1. Specifically, the numerical value corresponding to the Nth constituent element determined to be the cause of the abnormality is 100, and the numerical values ​​corresponding to the other constituent elements are all 0.

[0119] There are also cases where multiple components are the cause of the abnormality. Therefore, in one embodiment, the numerical values ​​corresponding to two or more components may be set to 100. When it is impossible to determine that the cause of the abnormality is the Nth component, the numerical value corresponding to the Nth component may be set to a number smaller than 100. For example, when it is highly likely that the cause of the abnormality of the substrate holder 18 is the Nth component but it cannot be determined, the numerical value corresponding to the Nth component may be set to 80.

[0120] exist Figure 6 In the example shown, as the constituent elements that may cause the abnormality of the substrate holder 18, the seal holder 62, the first holding element 54, the first electric contact 88, the second electric contact 86, the external electric contact 91, the seals 66, 68, and the slide plate 64 are listed. For example, when the Nth constituent element is the seal holder 62, the other constituent elements are the first holding element 54, the first electric contact 88, the second electric contact 86, the external electric contact 91, the seals 66, 68, and the slide plate 64. Therefore, the reference state data of the other constituent elements is the reference state data indicating the states of the first holding element 54, the first electric contact 88, the second electric contact 86, the external electric contact 91, the seals 66, 68, and the slide plate 64. The reference state data includes at least one of the image data and the surface shape data of the other constituent elements.

[0121] In step 4, the computing system 110 uses the selected data set generated in step 3 above and transforms the selected data set into Fig. 9 The parameters (weights, etc.) of the selection model shown are optimized. Specifically, the defect condition data and the reference condition data are input to the input layer 301 of the selection model. For example, when the seal holder 62 is discolored when the substrate holder 18 is abnormal, the image data of the seal holder 62 is input to the input layer 301 as the defect condition data, and the image data and surface shape data of the components other than the seal holder 62 are input to the input layer 301 as the reference condition data. The operation system 110 determines the optimal parameters of the selection model that minimize the difference between the set of output values ​​from the output layer 303 and the set of values ​​prepared in the above step 3.

[0122] from Fig.10The process of step 1 to step 4 is repeated until all the multiple components of the substrate holder 18 that may cause the abnormality of the substrate holder 18 are formed into a selection data set, and the parameters of the selection model are optimized using all the prepared selection data sets. That is, a new substrate holder (for example, another substrate holder having the same structure as the above-mentioned substrate holder 18, or a substrate holder 18 that has been repaired) is prepared, and before the new substrate holder causes an abnormality due to another component different from the Nth component, the new substrate holder is used to plate multiple substrates. The operation system 110 creates a new selection data set to further optimize the parameters of the above-mentioned selection model. Fig.10 The initial value of N shown is 1, and N=N+1 indicates that a certain component of the substrate holder is changed to another component. Figure 6 In the embodiment shown, the process from step 1 to step 4 is repeated until a selection data set is created for all of the seal holder 62, the first holding element 54, the first electrical contact 88, the second electrical contact 86, the external electrical contact 91, the seals 66, 68, and the slide plate 64, and the parameters of the selection model are optimized using all the created selection data sets. In this way, a selection model of the constituent elements that can predict the abnormality of the substrate holder 18 is constructed. The constructed selection model is stored in the storage device 110a of the operation system 110. In addition, Fig.10 The schematic diagram shows the selected data set corresponding to the constituent elements of the substrate holder 18 that may be the cause of the abnormality, and the learning (model optimization) using the selected data set. In fact, when making a high-precision prediction, it is necessary to use a sufficient amount of selected data sets for one constituent element to perform repeated learning.

[0123] Figure 1 The plating device shown in the figure uses a substrate holder having the same structure as the substrate holder 18 used for constructing the prediction model and the selection model to plate the substrate. Before plating the substrate, the imaging device 101 and the three-dimensional measuring device 102 generate image data and surface shape data of each component of the substrate holder, and the power inspection device 118 measures the internal resistance of the substrate holder. The image data and surface shape data are sent to the computing system 110 and stored in the storage device 110a.

[0124] The computing system 110 inputs all the latest status data representing the status of each of the multiple components of the substrate holder currently used by the plating device into the input layer 301 of the selection model. Each latest status data is composed of image data or surface shape data of each component of the substrate holder currently used. Figure 6In the example shown, image data and surface shape data of all constituent elements are input to the selection model input layer 301. The selection model outputs a plurality of confidences corresponding to a plurality of constituent elements.

[0125] The operation system 110 selects at least one component based on the plurality of confidences output from the selection model, and basically selects the component with the highest confidence.

[0126] The operation system 110 stores the reference value in its storage device 110a, and selects all components having a confidence level greater than the reference value. In the following example, component A and component B are selected.

[0127] Baseline value: 65%

[0128] Component A: Confidence 80%

[0129] Component B: Confidence 70%

[0130] Component C: 10% confidence

[0131] Component D: Confidence 5%

[0132] Component E: Confidence 1%

[0133] In the above example, when the reference value is 85%, all components are not selected. In this case, the first M components are selected (M is a natural number smaller than the total number of all components). Alternatively, when the difference between the confidence of the previous Kth component and the confidence of the previous K+1th component is greater than a preset value, the first K components are selected (K is a natural number smaller than M). For example, in the above example, when the preset value is 40%, since the difference between the confidence of component B and the confidence of component C is 60%, component A and component B are selected.

[0134] The operation system 110 uses the prediction model constructed for at least one selected constituent element to calculate the predicted number of substrates that can be plated before the substrate holder currently used by the plating device causes an abnormality. That is, the operation system 110 inputs image data and surface shape data into the prediction model constructed for at least one selected constituent element. The prediction model outputs the predicted number of substrates that can be plated before the substrate holder causes an abnormality, that is, the predicted number of processable sheets. When multiple constituent elements are selected, multiple prediction models corresponding to the above-mentioned multiple constituent elements respectively output the predicted number of processable sheets. The operation system 110 selects the smallest predicted number of processable sheets from multiple predicted numbers of processable sheets.

[0135] In the calculation of the predicted processable number of pieces using only the above-mentioned prediction model, because it is unknown which constituent element of the substrate holder in use will eventually cause an abnormality, it is necessary to input the latest status data corresponding to each constituent element into the prediction model corresponding to each constituent element. On the other hand, in the calculation of the predicted processable number of pieces using the above-mentioned selection model and the prediction model, first, all the latest status data of the states of the multiple constituent elements of the substrate holder currently used by the plating device are input into the input layer 301 of the selection model. The confidence output from the output layer of the selection model may change depending on the constituent element. That is, the confidence corresponding to the constituent element with the sign of the cause of the abnormality is high, and the confidence corresponding to the constituent element with the sign of the cause of the abnormality is low. In this way, by using the selection model, the constituent element that is the cause of the abnormality of the substrate holder can be predicted, and therefore, the predicted processable number of pieces can be calculated using only the latest status data and the prediction model corresponding to the constituent element. In one embodiment, the above-mentioned reference value can also be set to be low at the initial setting of the use of the substrate holder, and the reference value is increased as the number of processed pieces of the substrate holder increases.

[0136] In one embodiment, the operation system 110 may also include a first server (first computer) and a second server (second computer). The storage device 110a and the processing device 110b of the operation system 110 are composed of the first storage device and the first processing device in the first server, and the second storage device and the second processing device in the second server. For example, the construction and update of the above-mentioned prediction model and selection model may be performed by the first server, and the calculation of the predicted number of processable pieces and the calculation of the confidence using the prediction model and the selection model may be performed by the second server.

[0137] In one embodiment, the prediction model and the selection model may be constructed in advance by another system (server or computer) different from the operation system 110, and the constructed prediction model and the selection model may be installed in the operation system 110, and the operation system 110 only executes the prediction model and the selection model to calculate the number of processable slices and the calculation confidence.

[0138] Next, regarding one embodiment of the process of determining whether the substrate holder 18 mounted on the plating device can be used, according to Fig.11 as well as Fig.12 In step 1, the substrate holder 18 contained in the temporary storage box 24 is grasped by the first conveyor 42 of the substrate holder conveying device 40 and conveyed to the fixing station 20. Then, the substrate holder 18 is lowered to a horizontal state, thereby placing the substrate holder 18 on the placing plate 52 of the fixing station 20.

[0139] In step 2, the air cylinder of the fixing station 20 is operated to open the second holding element 58 of the substrate holder 18, and the image data and surface shape data of each component of the substrate holder 18 are generated by the imaging device 101 and the three-dimensional measuring device 102.

[0140] Fig.13 (a) and Fig.13 (b) is a schematic diagram showing an example of a method of photographing components of the substrate holder 18 and a method of measuring the surface shape. Fig.13 (a) is a schematic diagram showing an example of a method of photographing the seal holder 62 , the first electric contact 88 , the seal 66 , and the seal 68 , and a method of measuring the surface shape. Fig.13 (b) is a schematic diagram showing an example of a method for photographing the first holding element 54 and the second electrical contact 86 and a method for measuring the surface shape. The photographing device 101 photographs a plurality of components of the substrate holder 18 and generates image data of each component. The three-dimensional measuring device 102 measures the surface shapes of the plurality of components of the substrate holder 18 and generates surface shape data of each component.

[0141] Back to Fig.11 In step 3, the imaging device 101 and the three-dimensional measuring device 102 transmit the image data and surface shape data generated in step 2 to the computing system 110. In step 4, the controller 115 issues a command to the fixing station 20 to make the substrate holder 18 hold the substrate to be plated. Thereafter, the substrate holder 18 is connected to the leakage inspection device 117 (step 5).

[0142] In step 6, the leakage inspection device 117 inspects whether the seals 66 and 68 of the substrate holder 18 function normally. The leakage inspection device 117 forms a positive pressure or a negative pressure in the internal space R1 formed in the substrate holder 18 by the seals 66 and 68. When the pressure (positive pressure or negative pressure) in the internal space R1 changes beyond the allowable value within a prescribed time, an alarm signal indicating that an abnormality has occurred in the substrate holder 18 is issued. The alarm signal is transmitted to the controller 115. When receiving the alarm signal, the controller 115 issues an instruction to the substrate holder conveying device 40 to hold the substrate holder 18 by the first conveyor 42 of the substrate holder conveying device 40 and return it to a prescribed location of the temporary storage box 24 (step 7).

[0143] In the case where the leakage inspection device 117 issues an alarm signal, the user may also perform step 8. In step 8, the user may also remove the substrate holder 18 from the plating device, disassemble the substrate holder 18, and determine the cause of the abnormality of the substrate holder 18. In addition, the user may also use an input device and a communication device (not shown) to provide the operation system 110 with information that the abnormality of the substrate holder 18 is caused by the change in the state of the constituent element. That is, the user may also teach the operation system 110 to construct (learn) the constituent elements that should be used in the prediction model.

[0144] When the seals 66 and 68 of the substrate holder 18 function normally, the substrate holder 18 is connected to the power inspection device 118 while holding the substrate to be plated (step 9). In step 10, the power inspection device 118 measures the internal resistance of the substrate holder 18 and compares the measured resistance value with a predetermined setting range. When the measured resistance value exceeds the setting range, the power inspection device 118 issues an alarm signal indicating that an abnormality has occurred on the substrate holder 18. The alarm signal is transmitted to the controller 115. When receiving the alarm signal, the controller 115 issues an instruction to the substrate holder conveying device 40, uses the first conveyor 42 of the substrate holder conveying device 40 to hold the substrate holder 18, and returns it to a specified location of the temporary storage box 24 (step 11).

[0145] When the power-on inspection device 118 issues an alarm signal, the user may also perform step 12. In step 12, the user may also remove the substrate holder 18 from the plating device, disassemble the substrate holder 18, and determine the cause of the abnormality of the substrate holder 18. In addition, the user may also use an input device and a communication device (not shown) to provide the information that the abnormality of the substrate holder 18 is caused by the change in the state of the constituent element to the operation system 110. That is, the user may also teach the operation system 110 to construct (learn) the constituent element that the prediction model should use.

[0146] When the resistance value measured in step 10 is within the set range, the operation system 110 inputs the data transmitted in step 3 into the prediction model and outputs the predicted number of processable chips (step 13).

[0147] In step 14, the operation system 110 compares the predicted number of processable sheets output in step 12 with a predetermined threshold value. When the predicted number of processable sheets is greater than the threshold value, the process of determining whether the substrate holder 18 can be used is terminated, and the operation system 110 transmits a use permission signal to the controller 115. The controller 115 causes the plating device to perform the plating process described later.

[0148] When the predicted number of processable sheets is less than the threshold value, the operation system 110 issues an alarm signal indicating that maintenance of the substrate holder 18 is required. The alarm signal is transmitted to at least the controller 115. Upon receiving the alarm signal, the controller 115 issues an instruction to the substrate holder conveying device 40 to hold the substrate holder 18 by the first conveyor 42 of the substrate holder conveying device 40 and return it to a predetermined location of the temporary storage box 24 (step 15).

[0149] Hereinafter, an embodiment of the plating process will be described. The substrate is held in the substrate holder 18 in a state where the plated surface thereof is exposed from the opening 58a of the substrate holder 18. The substrate is electrically connected to a plurality of first electrical contacts 88 at a portion not in contact with the plating solution. The external electrical contacts 91 of each substrate holder 18 are in contact with a power supply electrode (not shown) disposed at the edge of each plating chamber 38. The power supply electrode is electrically connected to a conductive film such as a seed layer of the substrate through the external electrical contacts 91, the wire 92, the second electrical contact 86, and the first electrical contact 88. The power supply electrode is electrically connected to a power supply (not shown).

[0150] After the substrate holder 18 is suspended in the plating chamber 38 filled with the plating solution, a plating voltage is applied between the anode (not shown) in the plating chamber 38 and the substrate, thereby plating the surface of the substrate. After the plating is completed, the substrate holder 18 is transported to the second water washing tank 30b to clean the surface of the substrate. After the substrate is cleaned, the substrate and the substrate holder 18 are dried by the spray layer 32. The first conveyor 42 of the substrate holder conveying device 40 holds the dried substrate holder 18 and places it on the mounting plate 52 of the fixing station 20. The dried substrate is taken out from the substrate holder 18 and returned to the cassette 10.

[0151] The above-mentioned embodiments are recorded for the purpose of enabling the present invention to be implemented by persons with common knowledge in the technical field to which the present invention belongs. As long as a person skilled in the art can certainly implement various variations of the above-mentioned embodiments, the technical idea of ​​the present invention can also be applied to other embodiments. Therefore, the present invention is not limited to the described embodiments, but is interpreted in the widest range of the technical idea defined by the claims.

Claims

1. A method for constructing a prediction model for predicting the number of substrates that can be plated before a substrate holder causes an abnormality, In the method, Plating multiple substrates using a substrate holder, determining the total number of substrates plated using the substrate holder before the substrate holder causes an abnormality, The number of substrates that can be plated before the substrate holder causes an abnormality, i.e., the first processable number of substrates and the second processable number of substrates are determined. creating a first data set consisting of a combination of first status data corresponding to the first processable number of wafers and the first processable number of wafers, wherein the first status data indicates a state of a component of the substrate holder, creating a second data set consisting of a combination of second status data corresponding to the second processable number of sheets and the second processable number of sheets, wherein the second status data indicates a state of the constituent element; Optimizing the parameters of the prediction model composed of the neural network using the training data including the first data set and the second data set, The abnormality of the substrate holder is a leakage in the substrate holder or a power failure of the substrate holder. The first status data and the second status data are each composed of one of image data and surface shape data of the substrate holder.

2. The method according to claim 1, wherein: The first processable number of slices is 0, The first status data is defect status data indicating the status of the constituent element of the substrate holder when the abnormality occurs. The first data set is a defect data set consisting of a combination of the defect status data and 0.

3. The method according to claim 2, wherein: The second processable number of sheets is the processable number obtained by subtracting the number of intermediate sheets less than the total number of sheets from the total number of sheets. The second status data is intermediate status data indicating the status of the constituent element at the time when the intermediate number of substrates are plated. The second data set is an intermediate data set consisting of a combination of the intermediate status data and the second processable number of slices.

4. The method according to claim 1, wherein: The first processable number of sheets is the processable number of sheets obtained by subtracting a first intermediate number of sheets that is smaller than the total number of sheets from the total number of sheets. The first status data is first intermediate status data indicating the status of the constituent element at the time when the first intermediate number of substrates are plated. The first data set is a first intermediate data set consisting of a combination of the first intermediate status data and the first processable number of slices. The second processable number of sheets is the processable number of sheets obtained by subtracting the second intermediate number of sheets which is smaller than the first intermediate number of sheets from the total number of sheets. The second status data is second intermediate status data indicating the state of the constituent element at the time when the second intermediate number of substrates are plated. The second data set is a second intermediate data set composed of a combination of the second intermediate status data and the second processable number of slices.

5. The method according to any one of claims 1 to 4, wherein: The prediction model is composed of a neural network having an input layer, at least two intermediate layers and an output layer.

6. The method according to any one of claims 1 to 4, wherein: The prediction model is updated by repeating the steps from plating a plurality of substrates using a substrate holder to optimizing the parameters of the prediction model.

7. The method according to any one of claims 1 to 4, wherein: It also includes the following processes: creating a selection data set, the selection data set comprising: a value set indicating that the cause of the abnormality of the substrate holder is the constituent element; reference condition data indicating the state of other constituent elements of the substrate holder when the abnormality was caused; and defect condition data indicating the state of the constituent element when the abnormality was caused, Using the selection data set, the parameters of the selection model composed of the neural network are optimized.

8. A method for predicting the number of substrates that can be plated before a substrate holder causes an abnormality, wherein: preparing a prediction model constructed using the method according to any one of claims 1 to 7, inputting into the prediction model the latest status data representing the status of the constituent elements of the substrate holder currently in use, The prediction model outputs a predicted number of processable slices.

9. The method according to claim 8, wherein: The method further includes writing the estimated number of processable sheets into an electronic tag attached to the substrate holder currently in use.

10. A method for constructing a selection model for predicting a component that may cause an abnormality of a substrate holder from among a plurality of components of the substrate holder that may cause an abnormality of the substrate holder, In the method, before the first component of the first substrate holder causes an abnormality in the first substrate holder, plating a plurality of substrates; A first selected data set is prepared, the first selected data set comprising: a first value set indicating that the cause of the abnormality of the first substrate holder is the first constituent element; first reference condition data indicating the state of the second constituent element of the first substrate holder when the abnormality is caused; and first defect condition data indicating the state of the first constituent element when the abnormality is caused, Using the first selection data set, the parameters of the selection model composed of the neural network are optimized. before the second substrate holder fails due to the second component of the second substrate holder, the plurality of substrates are plated. creating a second selected data set, the second selected data set comprising: a second value set indicating that the cause of the abnormality of the second substrate holder is the second constituent element; second reference condition data indicating the state of the first constituent element of the second substrate holder when the abnormality was caused; and second defect condition data indicating the state of the second constituent element when the abnormality was caused, Using the second selected data set, the parameters are further optimized, The abnormality of the first substrate holder and the second substrate holder is liquid leakage in the first substrate holder and the second substrate holder or poor power supply of the first substrate holder and the second substrate holder. The first reference condition data and the first defect condition data are respectively composed of one of the image data and the surface shape data of the first substrate holder, and the second reference condition data and the second defect condition data are respectively composed of one of the image data and the surface shape data of the second substrate holder.

11. A method for predicting the number of substrates that can be plated before a substrate holder causes an abnormality, wherein: The process includes the following steps: inputting the latest status data of the first component and the second component of the substrate holder, which indicates the status of the components of the substrate holder currently in use, into the selection model constructed by the method according to claim 10, When the first certainty corresponding to the first constituent element output from the selection model is higher than the second certainty corresponding to the second constituent element, the latest status data of the first constituent element is input into the prediction model corresponding to the first constituent element, outputting the predicted processable number of wafers of the substrate holder from the prediction model, The prediction model is a prediction model constructed by the method according to any one of claims 1 to 9.

Images