Substrate processing apparatus

Abstract

A substrate processing apparatus includes at least one sensor that detects a physical quantity of an object in polishing and / or cleaning and / or drying of a substrate; a conversion section that converts a sensor value in polishing and / or cleaning and / or drying detected by the sensor into a feature quantity for each processing step for a learned machine learning model; and an inference section that outputs a prediction value of at least one of a number of defects, a size of defects, and a position of defects in the substrate of the object by inputting object data including the feature quantity into the learned machine learning model, the learned machine learning model being learned using a learning data group whose input data includes a feature quantity obtained by converting a sensor value in polishing and / or cleaning detected by a sensor for each processing step in a production line of the object or a production line of the same kind as the production line of the object, and whose output data is at least one of a number of defects, a size of defects, and a position of defects in the substrate.

Description

Technical Field

[0001] This invention relates to a substrate processing apparatus and an information processing system. Background Technology

[0002] The cleaning performance of a substrate processing apparatus (e.g., a polishing apparatus) is evaluated by measuring the substrates (specifically, wafers) that have been discharged from the apparatus after processing, such as polishing, cleaning, and drying, using a dedicated defect inspection device (e.g., see Patent Document 1). Defect inspection is costly (mainly time-consuming), so it is difficult to inspect all substrates after processing (e.g., polishing, cleaning, drying) at the manufacturing site, and sampling inspection is instead implemented.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent Application Publication No. 2002-257533

[0006] The technical problem that the invention aims to solve

[0007] However, there is also the possibility of defective products being produced due to insufficient cleaning of uninspected substrates (specifically, wafers). Summary of the Invention

[0008] The present invention was made in view of the above-mentioned problems, and its object is to provide a substrate processing apparatus and an information processing system that can determine whether a substrate after substrate processing is a defective product without inspection by a defect inspection device.

[0009] Technical means for solving technical problems

[0010] An embodiment of the substrate processing apparatus of the present invention comprises: at least one sensor that detects physical quantities of an object during polishing and / or cleaning and / or drying of the substrate; a conversion unit that, for a learned machine learning model, converts sensor values ​​detected by the sensor during polishing and / or cleaning and / or drying into feature quantities at each processing step; and an inference unit that outputs a predicted value of at least one of the number of defects, the size of defects, and the location of defects in the substrate of the object by inputting object data containing the feature quantities into the learned machine learning model, wherein the learned machine learning model is learned using a learning data set, the input data of which includes feature quantities obtained by converting sensor values ​​detected by the sensor during polishing and / or cleaning in the production line of the object or a production line of the same type as the production line of the object at each processing step, and the output data of the learning data set is at least one of the number of defects, the size of defects, and the location of defects in the substrate.

[0011] According to this structure, at least one of the following can be predicted: the number of defects, the size of defects, and the location of defects in the substrate after substrate processing, without inspection by a defect inspection device. Therefore, it is possible to presume whether the substrate after substrate processing is a defective product without inspection by a defect inspection device.

[0012] Alternatively, in the aforementioned substrate processing apparatus, the input data for learning the machine learning model may also include dwell time, which is the dwell time in each unit counted by the substrate processing apparatus. The substrate processing apparatus may also include a unit dwell time counting unit, which counts the dwell time in each unit contained in the substrate processing apparatus. The object data input to the learned machine learning model may also include the dwell time in each unit counted by the unit dwell time counting unit.

[0013] Alternatively, in the aforementioned substrate processing apparatus, the input data during the learning of the machine learning model may also include a second feature quantity, which is obtained by converting the position of the component used for grinding or cleaning. The conversion unit converts the position of the component used for grinding or cleaning into the second feature quantity, and the object data input to the learned machine learning model may also include the second feature quantity of each component after conversion by the conversion unit.

[0014] Alternatively, in the aforementioned substrate processing apparatus, the input data for the learning of the machine learning model may also include process information, which includes instruction values ​​for the units contained in the substrate processing apparatus, and the object data input to the learned machine learning model may also include process information, which includes instruction values ​​for the units contained in the substrate processing apparatus.

[0015] Alternatively, the substrate processing apparatus described above may include: a regression analysis unit that, according to a predetermined regression analysis algorithm, outputs a correlation parameter representing the relationship between each of a plurality of sensor values ​​and any one of the following: the number of defects, the size of defects, or the location of defects in the substrate; a receiving unit that receives at least one sensor whose output is a sensor value that forms the basis of the feature quantity contained in the input data of a machine learning model; and a learning unit that enables the machine learning model to be relearned using the feature quantity obtained by converting the sensor values ​​of the received sensor, and the inference unit uses the machine learning model relearned by the learning unit to output the predicted value.

[0016] The information processing system of another aspect of the present invention includes: a conversion unit that, for a learned machine learning model, converts sensor values ​​detected by sensors of a substrate processing apparatus during grinding and / or cleaning and / or drying of the substrate into feature quantities according to each processing step; and an inference unit that, by inputting object data containing the feature quantities, outputs a predicted value of at least one of the number of defects in the substrate, the size of defects, and the location of defects in the substrate, wherein the learned machine learning model is learned using a learning data set, the input data of which includes feature quantities obtained by converting sensor values ​​detected by sensors during grinding and / or cleaning in the production line of the object or a production line of the same type as the production line of the object according to each processing step, and the output data of which is at least one of the number of defects in the substrate, the size of defects, and the location of defects.

[0017] The effects of the invention

[0018] According to one aspect of the present invention, at least one of the following predicted values—the number of defects, the size of defects, and the location of defects—can be obtained without inspection by a defect inspection device. Therefore, it is possible to presume whether the substrate after substrate processing is a defective product without inspection by a defect inspection device. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating an example of the schematic structure of the substrate processing apparatus of this embodiment.

[0020] Figure 2 This is an example of a block diagram illustrating the general structure of the control unit in this embodiment.

[0021] Figure 3 This is a graph showing examples of dwell time in each unit of this embodiment.

[0022] Figure 4 It is an example of a data structure containing the features of the input data to a machine learning model.

[0023] Figure 5 This is an example of a data structure that represents the dwell time of each unit in the input data fed into a machine learning model.

[0024] Figure 6A This is a schematic diagram illustrating an example of the learning process in a machine learning model.

[0025] Figure 6B This is a schematic diagram illustrating an example of the inference process in a machine learning model.

[0026] Figure 7This is an example of a graph comparing the measured number of defects with the predicted number of defects.

[0027] Figure 8A This is a schematic diagram illustrating a variation of the learning process of a machine learning model, as shown in Example 1.

[0028] Figure 8B This is a schematic diagram illustrating a variation of Example 1 of the inference process of a machine learning model.

[0029] Figure 9A This is a schematic diagram of Variation 2 used to illustrate the learning process of a machine learning model.

[0030] Figure 9B This is a schematic diagram of variation 2 used to illustrate the inference process of a machine learning model.

[0031] Figure 10A This is a schematic diagram of variation 3 used to illustrate the learning process of a machine learning model.

[0032] Figure 10B This is a schematic diagram of variation 3 used to illustrate the inference process of a machine learning model.

[0033] Figure 11A This is a schematic structural diagram of a variation of this embodiment, Example 1.

[0034] Figure 11B This is a schematic structural diagram of a variation of this embodiment, Example 2.

[0035] Figure 12A This is a schematic structural diagram of a variation of this embodiment, Example 3.

[0036] Figure 12B This is a schematic structural diagram of variation 4 of this embodiment. Detailed Implementation

[0037] Hereinafter, various embodiments will be described with reference to the accompanying drawings. However, there may be instances where unnecessary details are omitted. For example, detailed descriptions of matters that are already clearly known or repeated descriptions of substantially the same structures may be omitted. This is to avoid making the following description unnecessarily lengthy and to facilitate understanding by those skilled in the art. In this embodiment, as an example of a substrate processing apparatus, a polishing apparatus for flattening the surface of a substrate W by performing chemical mechanical polishing (CMP) will be described.

[0038] Figure 1 This is a diagram illustrating an example of the schematic structure of the substrate processing apparatus according to this embodiment. (See diagram for example.) Figure 1As shown, substrate processing apparatus 1 is a polishing apparatus that performs chemical mechanical polishing (CMP) on the surface of a substrate W, such as a silicon wafer (hereinafter referred to as a wafer), to make it flat.

[0039] like Figure 1 As shown, the substrate processing apparatus 1 includes, for example, a rectangular box-shaped housing 2. The housing 2 is formed to be approximately rectangular in plan view. A substrate transport path 3 extending along its long side is provided in the center of the housing 2. A loading / unloading section 10 is provided at one end of the substrate transport path 3 in the long side direction. A polishing section 20 is provided on one side of the substrate transport path 3 in the width direction, that is, in the direction orthogonal to the long side direction in plan view, and a cleaning section 30 is provided on the other side. A substrate transport section 40 for transporting substrate W is provided in the substrate transport path 3. In addition, the substrate processing apparatus 1 includes a control section 50, which controls the operation of the loading / unloading section 10, the polishing section 20, the cleaning section 30, and the substrate transport section 40 in a unified manner.

[0040] The loading / unloading unit 10 includes a front loading section 11 for housing the substrate W. Multiple front loading sections 11 are provided on one side of the housing 2 along its long side. These multiple front loading sections 11 are arranged along the width direction of the housing 2. The front loading sections 11 may be equipped with, for example, an open box, an SMIF (Standard Manufacturing Interface) carrier, or a FOUP (FrontOpening Unified Pod). Both SMIF and FOUP are sealed containers that internally house the substrate W and are covered by partitions, maintaining an environment independent of the external space.

[0041] Furthermore, the loading / unloading unit 10 includes two transport robots 12 that receive and place substrates W from the front loading unit 11, and a travel mechanism 13 that moves each transport robot 12 along the arrangement direction of the front loading unit 11. Each transport robot 12 has two arms, an upper arm and a lower arm, which are used separately before and after processing the substrate W. For example, when returning the substrate W to the front loading unit 11, each transport robot 12 uses the upper arm, and when removing the substrate W from the front loading unit 11 before processing, each transport robot 12 uses the lower arm.

[0042] The polishing unit 20 includes multiple polishing devices 21 (21A, 21B, 21C, 21D) for polishing the substrate W. The multiple polishing devices 21 are arranged along the long side of the substrate transport path 3. Each polishing device 21 includes: a polishing table 23 for rotating a polishing pad 22 with a polishing surface; a top ring 24 for holding the substrate W and polishing the substrate W while pressing it against the polishing pad 22 on the polishing table 23; a polishing liquid supply nozzle 25 for supplying polishing liquid, dressing liquid, etc. to the polishing pad 22; a dressing device 26 for dressing the polishing surface of the polishing pad 22; and an atomizer 27 for spraying a mixture of liquid such as pure water and gas such as nitrogen, or a liquid such as pure water, into a mist onto the polishing surface.

[0043] The polishing apparatus 21 polishes the substrate W by supplying polishing slurry to the polishing pad 22 from the polishing slurry supply nozzle 25 and pressing the substrate W against the polishing pad 22 via the top ring 24. This causes the top ring 24 to move relative to the polishing table 23, thereby polishing the substrate W and flattening its surface. Furthermore, the top ring 24 has multiple concentrically arranged pressure-reducing zones (e.g., air cushions). The pressure of the top ring 24 on the polishing pad 22 is adjusted by regulating the pressure within these multiple pressure-reducing zones.

[0044] The dressing device 26 has hard particles such as diamond particles and ceramic particles fixed on the rotating part at the top end that contacts the grinding pad 22. By rotating and swinging the rotating part, the grinding surface of the grinding pad 22 is uniformly dressed to form a flat grinding surface.

[0045] The atomizer 27 performs the purification of the grinding surface and the reshaping operation of the grinding surface based on the dresser 26, i.e., the regeneration of the grinding surface, by flushing the grinding debris, abrasive particles and other residues on the grinding surface of the grinding pad 22 with high-pressure fluid.

[0046] The cleaning unit 30 includes multiple cleaning devices 31 (31A, 31B) for cleaning the substrate W and a substrate drying device 32 for drying the cleaned substrate W. The multiple cleaning devices 31 and the substrate drying device 32 are arranged along the long side of the substrate transport path 3. A first transport chamber 33 is provided between the cleaning devices 31A and 31B. A transport robot 35 is provided in the first transport chamber 33 to transport the substrate W between the substrate transport unit 40, the cleaning devices 31A and 31B.

[0047] The transport robot 35 has two arms, an upper arm and an lower arm, and these two arms are used separately before and after cleaning in the cleaning device 31A of the substrate W. For example, when taking the substrate W out of the temporary placement table (also called the wafer station) 47 (described later) before cleaning and transporting it to the cleaning device 31A, the transport robot 35 uses the lower arm, and when taking the substrate W out of the cleaning device 31A after cleaning and transporting it to the cleaning device 31A, the transport robot 35 uses the upper arm.

[0048] Additionally, a second transport chamber 34 is provided between the cleaning device 31B and the substrate drying device 32. A transport robot 36 for transporting the substrate W between the cleaning device 31B and the substrate drying device 32 is provided in the second transport chamber 34.

[0049] The cleaning apparatus 31 includes a roller-shaped sponge (hereinafter also referred to as a roller) type cleaning component, which is used to clean the substrate W. Furthermore, cleaning apparatus 31A and cleaning apparatus 31B can be of the same type or different types of cleaning components. Alternatively, cleaning apparatus 31A and cleaning apparatus 31B can replace the roller-shaped sponge type cleaning component with, for example, a pen-shaped sponge (hereinafter also referred to as a pen) type cleaning component or a two-fluid nozzle type cleaning component. Here, in the case of a two-fluid nozzle type cleaning component, the two fluids are, for example, a mixture of nitrogen (N2) and pure water.

[0050] The substrate drying apparatus 32 includes drying components for, for example, nitrogen (N2) drying and rotagonal drying using isopropyl alcohol (IPA). After the substrate W is dried using rotagonal drying, the gate 1a of the partition between the substrate drying apparatus 32 and the loading / unloading unit 10 is opened, and the substrate W is transported out of the substrate drying apparatus 32 by the transport robot 12.

[0051] The substrate transport unit 40 includes: a lift 41, a first linear transport device 42, a second linear transport device 43, and a swing transport device 44. The substrate transport path 3 is provided with a first transport position TP1, a second transport position TP2, a third transport position TP3, a fourth transport position TP4, a fifth transport position TP5, a sixth transport position TP6, and a seventh transport position TP7, sequentially arranged from the loading / unloading section 10 side.

[0052] The elevator 41 is a mechanism for vertically transporting the substrate W at the first transport position TP1. The elevator 41 receives the substrate W from the transport robot 12 of the loading / unloading unit 10 at the first transport position TP1. Then, the elevator 41 hands the substrate W received from the transport robot 12 to the first linear conveying device 42. A gate 1b is provided in the partition wall between the first transport position TP1 and the loading / unloading unit 10. During the transport of the substrate W, the gate 1b is opened, thereby transferring the substrate W from the transport robot 12 to the elevator 41.

[0053] The first linear conveying device 42 is a mechanism for conveying substrate W between two of the first conveying position TP1, the second conveying position TP2, the third conveying position TP3, and the fourth conveying position TP4. The first linear conveying device 42 includes a plurality of conveying arms 45 (45A, 45B, 45C, 45D) and a linear guide mechanism 46 that moves each conveying arm 45 at a plurality of heights in the horizontal direction.

[0054] The conveying arm 45A moves between the first conveying position TP1 and the fourth conveying position TP4 via the linear guide mechanism 46. The conveying arm 45A is a transfer arm used to receive the substrate W from the elevator 41 and hand the substrate W over to the second linear conveying device 43.

[0055] The conveyor arm 45B moves between a first conveying position TP1 and a second conveying position TP2 via a linear guide mechanism 46. At the first conveying position TP1, the conveyor arm 45B receives the substrate W from the elevator 41 and delivers the substrate W to the grinding apparatus 21A at the second conveying position TP2. The conveyor arm 45B is equipped with a lifting drive unit that rises when delivering the substrate W to the top ring 24 of the grinding apparatus 21A and descends after delivering the substrate W to the top ring 24. Furthermore, the same lifting drive unit is also provided in the conveyor arms 45C and 45D.

[0056] The conveying arm 45C moves between a first conveying position TP1 and a third conveying position TP3 via a linear guide mechanism 46. At the first conveying position TP1, the conveying arm 45C receives the substrate W from the elevator 41 and delivers the substrate W to the polishing apparatus 21B at the third conveying position TP3. Additionally, the conveying arm 45C also functions as a retrieval arm, receiving the substrate W from the top ring 24 of the polishing apparatus 21A at the second conveying position TP2 and delivering the substrate W to the polishing apparatus 21B at the third conveying position TP3.

[0057] The conveying arm 45D moves between the second conveying position TP2 and the fourth conveying position TP4 via the linear guide mechanism 46. The conveying arm 45D functions as a retrieval arm for receiving the substrate W from the top ring 24 of the grinding device 21A or the grinding device 21B at the second conveying position TP2 or the third conveying position TP3 and handing the substrate W over to the swing conveying device 44 at the fourth conveying position TP4.

[0058] The oscillating conveyor 44 has an arm capable of moving between a fourth conveying position TP4 and a fifth conveying position TP5, transferring the substrate W from the first linear conveyor 42 to the second linear conveyor 43. Additionally, the oscillating conveyor 44 transfers the substrate W, after being ground by the grinding unit 20, to the cleaning unit 30. A temporary placement stage 47 for the substrate W is provided on the side of the oscillating conveyor 44. The oscillating conveyor 44 reverses the substrate W received at the fourth conveying position TP4 or the fifth conveying position TP5 and places it on the temporary placement stage 47. The substrate W placed on the temporary placement stage 47 is then transported to the first conveying chamber 33 by the conveying robot 35 of the cleaning unit 30.

[0059] The second linear conveying device 43 is a mechanism for conveying substrate W between two of the fifth conveying position TP5, the sixth conveying position TP6, and the seventh conveying position TP7. The second linear conveying device 43 includes multiple conveying arms 48 (48A, 48B, 48C) and a linear guide mechanism 49 that moves each conveying arm 48 at multiple heights in a horizontal direction. Conveyor arms 48A move between the fifth conveying position TP5 and the sixth conveying position TP6 via the linear guide mechanism 49. Conveyor arms 48A function as access arms that receive substrate W from the oscillating conveying device 44 and deliver the substrate W to the polishing device 21C.

[0060] The conveying arm 48B moves between the sixth conveying position TP6 and the seventh conveying position TP7. The conveying arm 48B functions as a retrieval arm for receiving the substrate W from the polishing device 21C and handing the substrate W over to the polishing device 21D. The conveying arm 48C moves between the seventh conveying position TP7 and the fifth conveying position TP5. The conveying arm 48C functions as a retrieval arm for receiving the substrate W from the top ring 24 of the polishing device 21C or the polishing device 21D at the sixth conveying position TP6 or the seventh conveying position TP7 and handing the substrate W over to the oscillating conveyor 44 at the fifth conveying position TP5. Furthermore, although the description is omitted, the operation of the conveying arm 48 when handing over the substrate W is the same as the operation of the first linear conveyor 42 described above.

[0061] The substrate processing apparatus 1 includes at least one sensor (not shown) that detects physical quantities of the object during the polishing and / or cleaning and / or drying of the substrate (e.g., a wafer). In the case of polishing, the physical quantities of the object are, for example, the following parameters.

[0062] The rotational speed and / or torque of the grinding table 23

[0063] Speed ​​and / or torque of top ring 24

[0064] Air cushion pressure of top ring 24

[0065] Speed ​​and / or load of the dressing device 26

[0066] Slurry / water flow rate

[0067] Atomizer 27 flow rate

[0068] The flow rate of nitrogen (N2) in atomizer 27

[0069] In the case of cleaning, the physical quantities of the object are, for example, the following parameters.

[0070] Roller speed and / or torque and / or load

[0071] Pen speed and / or torque and / or load

[0072] Flow rate of medicine / water

[0073] wafer rotation speed

[0074] Nitrogen (N2) flow rate

[0075] In the case of drying, the physical quantities of the object are, for example, the following parameters.

[0076] Nitrogen (N2) flow rate

[0077] Isopropyl alcohol (IPA) flow rate

[0078] As an example of the aforementioned sensors, the substrate processing apparatus 1 may include: a sensor for detecting stage speed and / or torque, a sensor for detecting top ring speed and / or torque, a sensor for detecting top ring air cushion pressure, a sensor for detecting trimmer speed and / or load, a sensor for detecting slurry / water flow rate, and a sensor for detecting slurry / water flow rate.

[0079] In addition, the substrate processing apparatus 1 may also include: a sensor for detecting the roller speed and / or torque and / or load of the cleaning apparatus 31, a sensor for detecting the pen speed and / or torque and / or load of the cleaning apparatus 31, a sensor for detecting the flow rate of the cleaning solution / water in the cleaning apparatus 31, and a sensor for detecting the wafer speed of the cleaning apparatus 31.

[0080] In addition, the substrate processing apparatus 1 may also include a sensor for detecting the flow rate of nitrogen (N2) and a sensor for detecting the flow rate of isopropanol (IPA).

[0081] Hereinafter, the components constituting the loading / unloading unit 10, the grinding unit 20, the cleaning unit 30, and the substrate transport unit 40 will be referred to as units.

[0082] Figure 2 This is a block diagram illustrating an example of the schematic structure of the control unit in this embodiment. For example... Figure 2 As shown, the control unit 50 includes: a unit control unit 51, a processor 6, and a storage unit 7.

[0083] The unit control unit 51 centrally controls the operation of each unit of the loading / unloading unit 10, the grinding unit 20, the cleaning unit 30, and the substrate transport unit 40.

[0084] Storage unit 7 stores a learned machine learning model 71. This learned machine learning model 71 is a model obtained by learning using a learning data set. The input data of this learning data set includes feature quantities obtained by converting sensor values ​​detected by sensors during grinding and / or cleaning in the production line of the object or a production line similar to the object's production line according to each processing step. The output data is at least one of the number of defects in the substrate, the size of the defects, and the location of the defects. In this embodiment, as an example, the output data of the learning data set is described as the number of defects in the substrate. Here, the feature quantities are the average, maximum, minimum, total, median, standard deviation, variance, kurtosis, or skewness of the time series values ​​of the sensors, or the average, maximum, minimum, total, median, standard deviation, variance, kurtosis, or skewness of the time series data of the differential values ​​of the sensor values.

[0085] The processor 6 functions as the unit dwell time counting unit 61, conversion unit 62, learning unit 63, inference unit 64, regression analysis unit 65, and processing unit 66 by reading and executing the prescribed program from the storage unit 7.

[0086] The conversion unit 62 converts the sensor values ​​detected by the sensors during grinding and / or cleaning and / or drying into feature quantities for each processing step according to the learned machine learning model.

[0087] The learning unit 63 uses a learning data set to enable the machine learning model 71 to learn. The input data of the learning data set includes feature quantities obtained by converting sensor values ​​detected by sensors during grinding and / or cleaning in the production line of the object or a production line of the same type as the object according to each processing step, and the output data is at least one of the number of defects in the substrate, the size of the defects, and the location of the defects in the substrate.

[0088] The inference unit 64 inputs object data containing the said feature quantities into a learned machine learning model, and outputs a predicted value for at least one of the number of defects, the size of defects, and the location of defects in the substrate of the object.

[0089] The regression analysis unit 65, based on a pre-determined regression analysis algorithm, outputs a correlation parameter (e.g., correlation coefficient) for each of multiple sensor values, representing the correlation between that sensor value and any one of the following: the number of defects, the size of the defects, or the location of the defects in the substrate. Here, the regression analysis algorithm can be: LASSO (least absolute shrinkage and selection operator) regression, Ridge regression, Support Vector Regression (SVR), Random Forest Regression (RFR), or Light GBM. The regression analysis unit 65 can also display the correlation parameter (e.g., correlation coefficient) on the display device 8. Thus, the operator or user of the substrate processing device 1 can confirm the correlation parameter (e.g., correlation coefficient).

[0090] The receiving unit 66 receives at least one sensor from an operator or user who has confirmed the aforementioned output correlation parameters (e.g., correlation coefficients). This sensor outputs sensor values ​​that form the basis of the feature quantities contained in the input data of the machine learning model. At this time, the operator or user inputs or selects, for example, a sensor whose output is based on a highly correlated feature quantity. Then, the learning unit 63 causes the machine learning model 71 to learn again using the feature quantities obtained by converting the sensor values ​​of the received sensor. The inference unit 64 uses the machine learning model relearned by the learning unit 63 to output the aforementioned predicted value. According to this structure, by receiving the sensor input or selected by the operator or user, for example, a sensor whose output is based on a highly correlated feature quantity, the prediction accuracy of the machine learning model 71 after relearning can be improved.

[0091] In this embodiment, as an example, the input data for learning the machine learning model further includes dwell time, which is the dwell time spent in each unit included in the substrate processing apparatus, counted accordingly. Based on this, the unit dwell time counting unit 61 counts the dwell time spent in each unit included in the substrate processing apparatus 1. In this embodiment, as an example, the object data input to the learned machine learning model further includes the dwell time spent in each unit, counted by the unit dwell time counting unit.

[0092] In this embodiment, as an example, the input data for learning the machine learning model further includes a second feature value, which is obtained by transforming the position of the component used for grinding or cleaning. Based on this, the conversion unit 62 converts the position of the component used for grinding or cleaning into the second feature value. In this case, the object data input to the learned machine learning model further includes the second feature value of each component after transformation by the conversion unit 62.

[0093] In this embodiment, as an example, the input data for learning the machine learning model further includes process information, which contains instruction values ​​for the units included in the substrate processing apparatus. These instruction values ​​are set values ​​for physical quantities of the object (e.g., the rotational speed of the polishing table 23), and each set value for a physical quantity of the object is a different value in each processing step. Based on this, the object data input to the learned machine learning model further includes process information, which contains instruction values ​​for the units included in the substrate processing apparatus.

[0094] Next, use Figure 3 The dwell time of the substrate in each unit is explained. Figure 3 This is a graph showing examples of dwell time in each unit of this embodiment. Figure 3 In the graph, the vertical axis represents the processing step number, and the horizontal axis represents time. Figure 3 The following describes the dwell time of each unit in the following process. Specifically, the conveyor robot 12 delivers the substrate to the elevator 41, then the conveyor arm 45A of the first linear conveyor 42 removes the substrate from the elevator 41, and the conveyor arm 45A delivers the substrate W to the grinding device 21A at the second conveyor position TP2. After grinding in the grinding device 21A, the conveyor arm 45D of the first linear conveyor 42 receives the substrate from the grinding device 21A and delivers the substrate W to the oscillating conveyor 44 at the fourth conveyor position TP4. The oscillating conveyor 44 flips the received substrate vertically and places it on the temporary placement table 47. Afterwards, the lower arm of the conveyor robot 35 receives the substrate W and conveys it to the cleaning device 31A. After cleaning in the cleaning device 31A, the upper arm of the conveyor robot 35 removes the substrate W from the cleaning device 31A and conveys it to the cleaning device 31B. After cleaning in the cleaning device 31B, the arm of the conveyor robot 36 removes the substrate W from the cleaning device 31B and conveys it to the substrate drying device 32. Then, after drying in the substrate drying apparatus 32, the transport robot 12 removes the substrate from the substrate drying apparatus 32.

[0095] exist Figure 3In this context, TRBDs represents the dwell time of the substrate at the arm of the conveyor robot 12 before the grinding process, TLFT represents the dwell time of the substrate at the elevator 41, TLTP1 represents the dwell time of the substrate at the conveyor arm 45A of the first linear conveyor 42, and TPoliA represents the dwell time of the substrate at the grinding device 21A. The dwell time TPoliA of the substrate at the grinding device 21A includes not only the grinding process time but also standby time. This standby time can also be used as input data for the learning and inference of a machine learning model. TLTP3 is the dwell time of the substrate at the conveying arm 45D of the first linear conveying device 42, TSTP is the dwell time of the substrate at the swing conveying device 44, TWS1 is the dwell time of the substrate at the temporary placement table, TRB1L is the dwell time of the substrate at the lower arm of the conveying robot 35, TCL1A is the dwell time of the substrate at the cleaning device 31A, TRB1LU is the dwell time of the substrate at the upper arm of the conveying robot 35, TCL3A is the dwell time of the substrate at the cleaning device 31B, TRB3 is the dwell time of the substrate at the arm of the conveying robot 36, TCL4A is the dwell time of the substrate at the substrate drying device 32, and TRBDe is the dwell time of the substrate at the arm of the dried conveying robot 12.

[0096] like Figure 3 As shown, the grinding process has multiple processing steps, the first cleaning process also has multiple processing steps, the second cleaning process also has multiple processing steps, and the drying process also has multiple processing steps. Each of these processing steps is assigned a processing step number to identify it. Here, we will explain these processing steps as numbered 1 to N (i.e., processing step numbers are 1 to N, where N is a natural number).

[0097] Next, use Figure 4 and Figure 5 Here, an example of a data structure for the feature quantities contained in the input data to the machine learning model will be described. In the substrate processing apparatus 1, M sensors (M is a natural number) from sensor D1 to sensor DM will be used for the description. Figure 4 This is an example of a data structure containing the features of the input data to a machine learning model. For example... Figure 4As shown, the feature values ​​of sensor D1 are represented by an array (or vector) that treats the feature values ​​from processing steps 1 to N as elements. Similarly, the feature values ​​of sensor D2 are represented by an array (or vector) that treats the feature values ​​from processing steps 1 to N as elements, and the feature values ​​of sensor DM are represented by an array (or vector) that treats the feature values ​​from processing steps 1 to N as elements. Thus, the feature values ​​of sensor Di (where i is an integer from 1 to M) contained in the input data to the machine learning model are represented by an array (or vector) that treats the feature values ​​from processing steps 1 to N as elements.

[0098] Here, the components included in the aforementioned substrate processing apparatus 1 will be referred to as components U1 to UL (L is a natural number). The processor 6, for example, stores the positions of components 1 to L of the substrate processing apparatus 1 in a time sequence in the storage unit 7. The positions of these components U1 to UL can be calculated by the unit control unit 51 using a predetermined formula based on command signals instructing the operation of components U1 to UL, or determined based on a known correspondence between command signals and positions, or determined by positions detected by sensors. Figure 4 As shown, the positional features of component U1 are represented by an array (or vector) that uses the features from processing steps 1 to N as elements. Similarly, the positional features of component U2 are represented by an array (or vector) that uses the features from processing steps 1 to N as elements. Likewise, the positional features of component UL are represented by an array (or vector) that uses the features from processing steps 1 to N as elements. Thus, the positional features of component Uj (j is an integer from 1 to L) in the input data to the machine learning model are represented by an array (or vector) that uses the features from processing steps 1 to N as elements.

[0099] Figure 5 This is an example of a data structure that represents the dwell time of each unit in the input data fed into a machine learning model. For example... Figure 5 As shown, the dwell time of each unit in the input data to the machine learning model is represented by an array (or vector) that takes the dwell time of the substrate of each unit as a feature.

[0100] Next, use Figure 6A and Figure 6B An example of the learning and inference processes of machine learning model 71 will be provided. Figure 6A This is a schematic diagram illustrating an example of the learning process in a machine learning model. Figure 6B This is a diagram illustrating an example of the inference process in a machine learning model. For example... Figure 6AAs shown, in the learning process, machine learning model 71 uses the above-mentioned... Figure 4 The training uses an array of feature quantities for each of the more than one processing steps, process information, and an array of dwell times for each unit as input data, and the number of defects on the substrate as the output data. Figure 6B As shown, in the inference process, when the machine learning model 71 is input with an array of feature quantities for each processing step of the substrate of the object, process information, and an array of dwell times for each unit, the number of defects in the substrate of the object is output. Here, the feature quantities in the inference process are the same as the feature quantities in the learning process.

[0101] In addition, an array of features for each processing step, process information, and dwell time for each unit can be used as input data for machine learning, but it can be any one of them or any combination of two.

[0102] Figure 7 This is an example of a graph comparing the measured number of defects with the predicted number of defects. Figure 7 In the diagram, the vertical axis represents the predicted number of defects, and the horizontal axis represents the measured number of defects. The grinding devices 21A, 21B, 21C, and 21D are labeled by varying the concentration of the label. The closer the label group is to the dashed line L1, the higher the accuracy of the defect number prediction. For example... Figure 7 As shown, in each grinding device, the marked groups are distributed close to the dashed line L1, indicating that the prediction accuracy of the number of defects in each grinding device is high.

[0103] The substrate processing apparatus 1 of this embodiment includes: at least one sensor that detects physical quantities of an object during polishing and / or cleaning and / or drying of the substrate; a conversion unit 62 that converts sensor values ​​detected by the sensor during polishing and / or cleaning and / or drying into feature quantities for each processing step, based on a learned machine learning model; and an inference unit 64 that outputs a predicted value of the number of defects in the substrate by inputting object data containing the aforementioned feature quantities into a learned machine learning model 71. The learned machine learning model 71 is trained using a learning data set whose input data includes feature quantities obtained by converting sensor values ​​detected by the sensor during polishing and / or cleaning in the production line of the object or a production line similar to the production line of the object into feature quantities for each processing step, and whose output data is the number of defects in the substrate.

[0104] According to this structure, at least one of the following can be predicted: the number of defects, the size of defects, and the location of defects in the substrate after substrate processing, without inspection by a defect inspection device. Therefore, it is possible to presume whether the substrate after substrate processing is a defective product without inspection by a defect inspection device.

[0105] <Example 1 of a Machine Learning Model>

[0106] Next, use Figure 8A and Figure 8B Example 1, a variation of the learning and inference processes of machine learning model 71, will be explained. Figure 8A This is a schematic diagram illustrating a variation of the learning process of a machine learning model, as shown in Example 1. Figure 8B This is a schematic diagram illustrating a variation of Example 1, used to explain the inference process of a machine learning model. For example... Figure 8A As shown, in the learning process, machine learning model 71 is used in the above-mentioned... Figure 4 The learning data set, described in the text, uses an array of feature quantities for each of the more than one processing steps, process information, and an array of dwell times for each unit as input data, and uses the number of defects in the substrate and the size of the defects in the substrate as output data. Figure 8B As shown, in the inference process, when the machine learning model 71 is input with an array of feature quantities for each processing step of the substrate of the object, process information, and an array of dwell times for each unit, the output is the number of defects in the substrate of the object and the size of the defects in the substrate of the object. The feature quantities in this inference process are the same as the feature quantities in the learning process.

[0107] <Modification of Machine Learning Model 2>

[0108] Next, use Figure 9A and Figure 9B Example 2 illustrates a variation of the learning and inference processes of machine learning model 71. Figure 9A This is a schematic diagram of Variation 2 used to illustrate the learning process of a machine learning model. Figure 9B This is a schematic diagram illustrating a variation of Example 2, used to explain the inference process of a machine learning model. For example... Figure 9A As shown, in the learning process, machine learning model 71 uses the methods described above. Figure 4 The learning data set, which includes an array of feature quantities for each of the more than one processing steps described above, process information, and an array of dwell times for each unit, is used as input data, and the number and location of defects on the substrate are used as output data. Figure 9BAs shown, in the inference process, when the machine learning model 71 is input with an array of feature quantities for each processing step of the substrate of the object, process information, and an array of dwell times for each unit, the output is the number of defects in the substrate of the object and the location of the defects in the substrate of the object. Here, the feature quantities in this inference process are the same as the feature quantities in the learning process.

[0109] <Modification of Machine Learning Model 3>

[0110] Next, use Figure 10A and Figure 10B Example 3 illustrates a variation of the learning and inference processes of machine learning model 71. Figure 10A This is a schematic diagram of variation 3 used to illustrate the learning process of a machine learning model. Figure 10B This is a schematic diagram illustrating variation 3 of the inference process in a machine learning model. For example... Figure 10A As shown, in the learning process, machine learning model 71 uses the methods described above. Figure 4 The learning data set, which includes an array of feature quantities for each of the more than one processing steps described above, process information, and an array of dwell times for each unit, is used as input data, and the number of defects in the substrate, the size of the defects in the substrate, and the location of the defects in the substrate are used as output data. For example... Figure 10B As shown, in the inference process, when the machine learning model 71 is input with an array of feature quantities for each processing step of the substrate of the object, process information, and an array of dwell times for each unit, the output is the number of defects in the substrate of the object, the size of the defects in the substrate of the object, and the location of the defects in the substrate of the object. Here, the feature quantities in the inference process are the same as the feature quantities in the learning process.

[0111] Furthermore, not limited to the combination of output data from Variations 1 to 3, the inference unit 64 can also output one or both of the number of defects, the size of defects, and the location of defects in the substrate of the object. In this way, the inference unit 64 can output a predicted value of at least one of the number of defects, the size of defects, and the location of defects in the substrate of the object by inputting object data containing the above-mentioned feature quantities into a learned machine learning model.

[0112] In this embodiment, the substrate processing apparatus 1 includes a processor 6 and a storage unit 7, but is not limited thereto.

[0113] <Modification 1 of this embodiment>

[0114] Figure 11A This is a schematic structural diagram of a variation of this embodiment, Example 1. Figure 11AAs shown, an information processing system S1 may also be included. This information processing system S1 is connected to the substrate processing device 1 in a manner that enables information exchange. The information processing system S1 has a processor 6 and a storage unit 7 that stores a machine learning model 71. The processor 6 functions as a unit dwell time counting unit 61, a conversion unit 62, a learning unit 63, an inference unit 64, a regression analysis unit 65, and a receiving unit 66 by reading and executing programs from the storage unit 7.

[0115] <Modification 2 of this embodiment>

[0116] Figure 11B This is a schematic structural diagram of a variation of this embodiment, Example 2. Figure 11B As shown, an information processing system S1 may also be included. This information processing system S1 is connected to the board processing device 1 in a manner that enables information exchange via a communication circuit network CN. The information processing system S1 has a processor 6 and a storage unit 7 that stores a machine learning model 71. The processor 6 functions as a unit dwell time counting unit 61, a conversion unit 62, a learning unit 63, an inference unit 64, a regression analysis unit 65, and a receiving unit 66 by reading from the storage unit 7 and executing a prescribed program.

[0117] <Modification 3 of this embodiment>

[0118] Figure 12B This is a schematic structural diagram of variation 4 of this embodiment. Figure 12B As shown, a server 70 may also be included, which is connected to the substrate processing apparatus 1 in a manner that enables information exchange via a communication circuit network CN. The substrate processing apparatus 1 has a processor 6 and a storage unit 7a storing a predetermined program, while the server 70 has a storage unit 7b storing a machine learning model 71. In this case, the processor 6 of the substrate processing apparatus 1 may function as a unit dwell time counting unit 61, a conversion unit 62, a learning unit 63, an inference unit 64, a regression analysis unit 65, and a processing unit 66 by reading and executing the predetermined program from the storage unit 7a. The inference unit 64 inputs object data containing the aforementioned feature quantities into the learned machine learning model 71 of the server 70, and outputs a predicted value of at least one of the following: the number of defects in the substrate of the object, the size of the defects, and the location of the defects.

[0119] <Modification 4 of this embodiment>

[0120] Figure 12A This is a schematic structural diagram of a variation of this embodiment, Example 3. Figure 12AAs shown, the system may have a structure comprising a substrate processing apparatus 1 and an information processing system S3. The information processing system S3 may include an information processing apparatus 60 connected to the substrate processing apparatus 1 in a manner capable of exchanging information, and a server 70 connected to the information processing apparatus 60 in a manner capable of exchanging information via a communication circuit network CN. In this case, the information processing apparatus 60 may have a processor 6 and a storage unit 7a storing a predetermined program, and the server 70 may have a storage unit 7b storing a machine learning model 71. Here, the processor 6 of the information processing apparatus 60 may function as a unit dwell time counting unit 61, a conversion unit 62, a learning unit 63, an inference unit 64, a regression analysis unit 65, and a receiving unit 66 by reading and executing the predetermined program from the storage unit 7a. The inference unit 64 may output a predicted value of at least one of the following: the number of defects, the size of defects, and the location of defects in the substrate of the object, by inputting object data containing the aforementioned feature quantities into the learned machine learning model 71 of the server 70.

[0121] Such information processing systems S1, S2, and S3 may also include a conversion unit 62 and an inference unit 64. The conversion unit 62, for a learned machine learning model, converts sensor values ​​detected by sensors in the substrate processing apparatus 1 during the grinding and / or cleaning and / or drying of the substrate into feature quantities according to each processing step. The inference unit 64, by inputting object data containing the aforementioned feature quantities, outputs a predicted value for at least one of the number of defects in the substrate, the size of the defects, and the location of the defects in the substrate. Here, the learned machine learning model is learned using a learning data set. The input data of this learning data set includes feature quantities obtained by converting sensor values ​​detected by sensors during grinding and / or cleaning in the production line of the object or a production line similar to the production line of the object according to each processing step, and the output data is at least one of the number of defects in the substrate, the size of the defects, and the location of the defects.

[0122] According to this structure, at least one of the following can be predicted: the number of defects, the size of defects, and the location of defects in the substrate after substrate processing, without inspection by a defect inspection device. Therefore, it is possible to presume whether the substrate after substrate processing is a defective product without inspection by a defect inspection device.

[0123] Furthermore, at least a portion of the functions of the information processing systems S1, S2, S3, or processor 6 described in the above embodiments can be implemented in hardware or software. In the case of software implementation, a program that implements at least a portion of the functions of the information processing systems S1, S2, S3, or processor 6 can be stored in a computer-readable storage medium and executed by the computer. The storage medium is not limited to removable storage media such as disks or optical discs, but can also be a fixed storage medium such as a hard disk drive or a memory.

[0124] Alternatively, programs that implement at least a portion of the functions of information processing systems S1, S2, S3, or processor 6 can be distributed via communication lines such as networks (including wireless communication). Furthermore, the same program can be distributed via wired or wireless lines such as networks, or stored on storage media, while being encrypted, modulated, or compressed.

[0125] Furthermore, information processing systems S1, S2, and S3 can also function through one or more information devices. When using multiple information devices, one of them can be used as a computer, and the computer can execute a prescribed program to perform at least one function of information processing systems S1, S2, and S3.

[0126] In summary, the present invention is not limited to the embodiments directly described above, but can be embodied by modifying the constituent elements during the implementation stage without departing from its spirit. Furthermore, various inventions can be formed by appropriately combining the multiple constituent elements disclosed in the above embodiments. For example, several constituent elements may be deleted from all the constituent elements shown in the embodiments. Moreover, constituent elements from different embodiments may be appropriately combined.

[0127] Symbol Explanation

[0128] 1. Substrate processing apparatus

[0129] 50 Control Department

[0130] Unit 51 Control Section

[0131] 6 processors

[0132] Unit 61 Dwell Time Counting Section

[0133] 62 Conversion Section

[0134] 63. Study Department

[0135] 64. Inference Section

[0136] 65 Regression Analysis Department

[0137] 66 Reception Department

[0138] 7. Storage Section

[0139] 71 Machine Learning Models

[0140] 8 Display devices

[0141] S1, S2, and S3 information processing systems

Claims

1. A substrate processing apparatus, characterized in that, have: At least one sensor, which detects physical quantities of an object during the grinding and / or cleaning and / or drying of a substrate; A conversion unit, for a learned machine learning model, converts sensor values ​​detected by the sensors during grinding and / or washing and / or drying into feature quantities according to each processing step; and The inference unit, by inputting object data containing the said feature quantities into a learned machine learning model, outputs a predicted value for at least one of the following: the number of defects, the size of the defects, and the location of the defects in the substrate of the object. The learned machine learning model is trained using a learning dataset. The input data of the learning dataset includes feature values ​​obtained by converting sensor values ​​detected by sensors during grinding and / or cleaning in the production line of the object or a production line of the same type as the object, according to each processing step. The output data of the learning dataset is at least one of the following: the number of defects in the substrate, the size of the defects, and the location of the defects. The input data for the machine learning model also includes dwell time, which is the dwell time of the substrate in each unit of the substrate processing apparatus, counted according to the dwell time of the substrate in that unit. The substrate processing apparatus further includes a unit dwell time counting unit, which counts the dwell time of the substrate in each unit included in the substrate processing apparatus. The object data input to the learned machine learning model also includes the dwell time of the substrate in the unit, which has been counted by the unit dwell time counting unit.

2. The substrate processing apparatus according to claim 1, characterized in that, The input data for the machine learning model also includes a second feature quantity, which is obtained by transforming the position of the component used for grinding or cleaning. The conversion unit converts the position of the component used for grinding or cleaning into a second feature quantity. The object data input to the learned machine learning model also includes a second feature quantity for each component after being transformed by the transformation unit.

3. The substrate processing apparatus according to claim 1 or 2, characterized in that, The input data for the machine learning model during learning also includes process information, which contains instruction values ​​for the units contained in the substrate processing apparatus. The object data input to the learned machine learning model also includes process information, which contains instruction values ​​for the units contained in the substrate processing apparatus.

4. A substrate processing apparatus, characterized in that, have: At least one sensor, which detects physical quantities of an object during the grinding and / or cleaning and / or drying of a substrate; The conversion unit, for the learned machine learning model, converts the sensor values ​​detected by the sensors during grinding and / or washing and / or drying into feature quantities according to each processing step; The inference unit outputs a predicted value for at least one of the number of defects, the size of defects, and the location of defects in the substrate of the object by inputting object data containing the said feature quantities into a learned machine learning model. The regression analysis unit, according to a predetermined regression analysis algorithm, outputs a correlation parameter for each of the multiple sensor values, representing the correlation between the sensor value and any one of the following: the number of defects, the size of the defects, or the location of the defects in the substrate. The receiving department accepts at least one sensor, which outputs sensor values ​​as the basis for the feature quantities contained in the input data of a machine learning model. as well as The learning unit enables the machine learning model to learn again using the feature quantities obtained by converting the sensor values ​​of the received sensors. The learned machine learning model is trained using a learning dataset. The input data of the learning dataset includes feature values ​​obtained by converting sensor values ​​detected by sensors during grinding and / or cleaning in the production line of the object or a production line of the same type as the object, according to each processing step. The output data of the learning dataset is at least one of the following: the number of defects in the substrate, the size of the defects, and the location of the defects. The inference unit uses a machine learning model that has been relearned by the learning unit to output the predicted value.

Images