JP2020120075A - Polishing method and polishing apparatus - Google Patents

Abstract

To provide a polishing method capable of maximizing a polishing rate of a substrate such as a wafer without requiring knowledge, experience, and labor of a worker.SOLUTION: The polishing method includes the steps of: inputting a type of polishing tool specified from sizes of abrasive grains and a value of a polishing width and a value of a polishing depth of a substrate polished by the polishing tool into a model constructed by a machine learning algorithm; and outputting from the model a polishing recipe having polishing conditions including a condition for polishing the substrate by contact pressure of the polishing tool to the substrate when polishing power of the polishing tool is maximum.SELECTED DRAWING: Figure 8

Description

The present invention relates to a polishing method and a polishing apparatus.

In recent years, attention has been paid to the management of the surface condition of a wafer from the viewpoint of improving the yield in manufacturing semiconductor devices. In the process of manufacturing a semiconductor device, various materials are deposited on a silicon wafer. Therefore, unnecessary films and surface roughness are formed on the peripheral portion of the wafer. In recent years, a method of holding a peripheral edge of a wafer by an arm and transporting the wafer has become common. Under such a background, the unnecessary film remaining on the peripheral portion is peeled off and adheres to the device formed on the wafer during various processes, thereby lowering the yield.

Therefore, in order to remove the unnecessary film formed on the peripheral portion of the wafer, the peripheral portion of the wafer is polished by using a polishing device. Further, in the recent stacking of wafers to improve the degree of integration, control of the cross-sectional shape of the peripheral edge portion when thinning the film prevents wafer cracking and the like, and is an important factor for improving the yield.

JP, 2017-124471, A JP, 2018-14465, A JP, 2011-171647, A

FIG. 13 is a diagram showing a multilayer wafer. As shown in FIG. 13, the wafer W includes a surface layer L1 (for example, a silicon nitride (SiN) film or a metal film) that forms an exposed surface, and a silicon (Si) layer L2 under the surface layer L1. have. The polishing tool (for example, a polishing tape or a grindstone) is pressed against the peripheral edge of the wafer W by the pressing member to polish the peripheral edge of the wafer W. In this way, the polishing tool forms a stepped depression in the peripheral portion of the wafer W.

The surface layer L1 and the silicon layer L2 may have different hardness. In this case, the value of the friction coefficient (μ) between the surface layer L1 and the polishing tool is different from the value of the friction coefficient (μ) between the silicon layer L2 and the polishing tool. Therefore, the optimum polishing load to be applied to the surface layer L1 and the optimum polishing load to be applied to the silicon layer L2 are different. Therefore, in order to maximize the polishing rate (polishing amount per unit time) and improve the polishing efficiency, it is necessary to apply different polishing loads to the surface layer L1 and the silicon layer L2.

There are various types of abrasive tools having different sizes of abrasive particles. Therefore, the optimum polishing load to be applied to the peripheral portion of the wafer W differs depending on the type of polishing tool used.

As shown in FIG. 13, the cross-sectional shape of the peripheral portion of the wafer W is not uniform in the thickness direction of the wafer W. That is, the width of the portion of the wafer W to be polished (polishing width Wp) changes according to the depth of the portion of the wafer W to be polished (polishing depth Dp). Therefore, the area of the wafer W to be polished (polishing area) gradually changes as the polishing of the wafer W progresses. Therefore, in order to maximize the polishing rate, it is necessary to change the polishing load to be applied from the polishing tool to the wafer W according to the change in the polishing area of the wafer W.

As described above, in order to maximize the polishing rate of the wafer W, it is necessary to construct an optimum polishing recipe in consideration of various factors. However, conventionally, the ability (knowledge and experience) of an operator is required to generate an optimum polishing recipe, which is troublesome.

Therefore, there is provided a polishing method capable of maximizing the polishing rate of a substrate such as a wafer without requiring the knowledge, experience, and labor of the operator. Further, there is provided a polishing apparatus capable of executing such a polishing method.

In one aspect, the type of the polishing tool specified from the size of the abrasive grains, the value of the polishing width and the value of the polishing depth of the substrate to be polished by the polishing tool, in the model constructed by the machine learning algorithm. Inputting, outputting a polishing recipe having polishing conditions from the model, including conditions for polishing the substrate with the contact pressure of the polishing tool with respect to the substrate when the polishing power of the polishing tool is maximum, the polishing method is Provided.

In one aspect, the polishing recipe includes at least a polishing load applied to the substrate from the polishing tool, and the polishing condition is the polishing load according to a polishing area of the substrate that changes with progress of polishing of the substrate. Including the condition to change.
In one aspect, the polishing recipe includes at least a polishing load applied to the substrate from the polishing tool, and the polishing condition changes the polishing load according to a plurality of layers having different hardness of the substrate. Including conditions.
In one aspect, in the construction of the model, the explanatory variables of the learning data are given as polishing information data including the pre-polishing acquisition data and the during-polishing acquisition data, and the objective variable is the deviation of the actual polishing rate from the predicted polishing rate. Is given as a numerical value.

In one aspect, the pre-polishing acquisition data includes image data of an edge portion of the substrate generated by an image acquiring device, and the during-polishing acquisition data includes a type of a polishing tool used and a substrate holding the substrate. The rotation speed and rotation torque of the holding unit, the vibration of the polishing head that presses the polishing tool against the substrate, the position of the polishing head in the thickness direction of the substrate, the descending speed of the polishing head, the polishing tool applied to the substrate. The polishing load applied, the moving speed of the polishing tool, and the flow rate and temperature of the liquid supplied onto the substrate.

In one aspect, a substrate holding unit that holds a substrate, a polishing head that presses a polishing tool against the substrate, and a computer having a model constructed by a machine learning algorithm are provided, and the computer stores the model. A storage device, the type of the polishing tool specified from the size of the abrasive grains, the value of the polishing width and the value of the polishing depth of the substrate to be polished by the polishing tool, is input to the model, the A processing device that executes a calculation for outputting a polishing recipe having a polishing condition from the model, including a condition for polishing the substrate with a contact pressure of the polishing tool with respect to the substrate when the polishing power of the polishing tool is maximum. There is provided a polishing device including:

In one aspect, in the input layer of a model including a neural network having an input layer, a plurality of hidden layers, and an output layer, a type of a polishing tool specified by the size of abrasive grains, and polishing by the polishing tool. The step of inputting the value of the polishing width and the value of the polishing depth of the substrate, and the step of polishing the polishing tool from the output layer by performing an operation according to the algorithm of the multilayer perceptron forming the neural network. Outputting a polishing recipe having polishing conditions, including conditions for polishing the substrate with the contact pressure of the polishing tool on the substrate when the force is maximized, and a computer-readable program recording a program for executing the computer. Recording medium is provided.

By using the model constructed by the machine learning algorithm, it is possible to automatically and accurately generate the polishing recipe which has been conventionally generated depending on the knowledge and experience of the operator. As a result, the polishing rate of the substrate can be maximized.

1A and 1B are enlarged cross-sectional views showing a peripheral portion of a wafer as an example of a substrate. It is a top view which shows one Embodiment of a polishing device typically. FIG. 3 is a side view showing the polishing apparatus shown in FIG. 2. It is a figure which shows a polishing head assembly. It is a figure which shows the polishing head at the time of polishing a wafer. FIG. 6A is a diagram showing a state of the stopper when the stopper does not receive a load from the positioning member, and FIG. 6B shows a state of the stopper when a load is applied from the positioning member to the stopper. It is a figure. It is a figure which shows an example of the database with which the load determination part is equipped. It is a figure which shows an example of the performance characteristic of a polishing tape. It is a figure which shows the polishing area of a wafer, and a notch part. 10(a) and 10(b) are diagrams showing contact with the notch portion of the polishing tape. It is a schematic diagram which shows one Embodiment of the model which produces|generates a polishing recipe. FIG. 6 is a schematic diagram showing an embodiment of a computer that constitutes at least a part of the control device shown in FIGS. 4 and 5. It is a figure which shows a multilayer wafer.

Embodiments of the present invention will be described below with reference to the drawings. A polishing method and a polishing apparatus according to embodiments described below polish a peripheral edge of a substrate by bringing a polishing tool into contact with the peripheral edge of the substrate. Here, in this specification, the peripheral portion of the substrate is defined as a region including a bevel portion located at the outermost periphery of the substrate, and a top edge portion and a bottom edge portion located radially inside the bevel portion.

1A and 1B are enlarged cross-sectional views showing a peripheral portion of a wafer as an example of a substrate. More specifically, FIG. 1A is a sectional view of a so-called straight type wafer, and FIG. 1B is a sectional view of a so-called round type wafer. In the wafer W of FIG. 1A, the bevel portion is a wafer W including an upper inclined portion (upper bevel portion) P, a lower inclined portion (lower bevel portion) Q, and a side portion (apex) R. It is the outermost peripheral surface (indicated by reference numeral B). In the wafer W of FIG. 1B, the bevel portion is a portion having a curved cross section (indicated by reference numeral B) that constitutes the outermost peripheral surface of the wafer W. The top edge portion is a flat portion E1 located inside the bevel portion B in the radial direction. The bottom edge portion is a flat portion E2 located on the opposite side of the top edge portion and on the inner side of the bevel portion B in the radial direction. The top edge portion E1 may include a region where a device is formed. In the following description, the top edge portion is simply referred to as the edge portion.

2 is a plan view schematically showing an embodiment of the polishing apparatus, and FIG. 3 is a side view showing the polishing apparatus shown in FIG. The polishing apparatus includes a wafer holding unit (substrate holding unit) 1 that holds and rotates a wafer W that is an example of a substrate. The wafer holder 1 has a wafer stage (substrate stage) 2 that can hold a wafer W and a stage motor 3 that rotates the wafer stage 2 around its axis. The wafer W to be polished is held on the upper surface of the wafer stage 2 by vacuum suction or the like, and is rotated by the stage motor 3 together with the wafer stage 2.

A stage torque sensor 4 that detects a motor torque of the stage motor 3 is electrically connected to the stage motor 3. The stage torque sensor 4 detects the motor torque of the stage motor 3 based on the motor current value of the stage motor 3.

The polishing apparatus of this embodiment includes a polishing head assembly 10 including a polishing head 11 that presses a polishing tool for polishing the wafer W against the edge portion of the wafer W. The polishing head 11 is arranged above the wafer stage 2. In this embodiment, the polishing tool is the polishing tape 7. One end of the polishing tape 7 is fixed to the take-up reel 14, and the other end of the polishing tape 7 is fixed to the take-up reel 15. Most of the polishing tape 7 is wound on both the take-up reel 14 and/or the take-up reel 15, and a part of the polishing tape 7 extends between the take-up reel 14 and the take-up reel 15. The unwinding reel 14 and the take-up reel 15 are respectively applied with torques in opposite directions by the reel motors 17 and 18, whereby tension is applied to the polishing tape 7.

A tape feeding device 20 is arranged between the take-up reel 14 and the take-up reel 15. The polishing tape 7 is sent from the unwinding reel 14 to the take-up reel 15 by the tape feeding device 20 at a constant speed. The polishing tape 7 extending between the take-up reel 14 and the take-up reel 15 is supported by two guide rollers 21 and 22. These two guide rollers 21 and 22 are arranged between the take-up reel 14 and the take-up reel 15. The lower surface of the polishing tape 7 extending between the guide rollers 21 and 22 constitutes a polishing surface for polishing the wafer W.

The polishing head 11 is located between the two guide rollers 21 and 22. At the contact point between the edge portion of the wafer W and the polishing tape 7, the guide rollers 21 and 22 are arranged so that the polishing tape 7 between the guide rollers 21 and 22 extends in the tangential direction of the wafer W.

The polishing of the wafer W is performed as follows. The wafer W is held on the wafer stage 2 so that a film (for example, a device layer) formed on the surface of the wafer W faces upward, and the wafer W is rotated around the axis of the wafer W together with the wafer stage 2. A liquid (for example, pure water) is supplied to the center of the rotating wafer W from a liquid supply nozzle 24 (see FIG. 3). In this state, the polishing head 11 of the polishing head assembly 10 presses the polishing tape 7 against the edge portion of the wafer W. The wafer W is polished by the sliding contact between the rotating wafer W and the polishing tape 7. The polishing tape 7 when polishing the wafer W extends in the tangential direction of the wafer W at the contact point between the wafer W and the polishing tape 7, as shown in FIG.

FIG. 4 is a view showing the polishing head assembly 10. As shown in FIG. 4, the polishing head assembly 10 includes the polishing head 11 that presses a polishing tool against the wafer W, and an air cylinder 25 that applies a force to the polishing head 11. The polishing head 11 is connected to the air cylinder 25. The air cylinder 25 is an actuator that pushes the polishing head 11 in a predetermined direction toward the wafer W on the wafer holder 1. In the present embodiment, the air cylinder 25 is configured to push the polishing head 11 downward toward the edge portion of the wafer W. In this embodiment, the direction of the polishing head 11 pushed by the air cylinder 25 is a direction parallel to the axis of the wafer holder 1, that is, the vertical direction.

The polishing head 11 includes a pressing member 12 that holds the polishing tape 7, a pressing member holder 27a that detachably holds the pressing member 12, and a load transmission member 27 that connects the pressing member holder 27a and the air cylinder 25. ing. The lower end of the load transfer member 27 is connected to the pressing member holder 27 a, and the upper end of the load transfer member 27 is connected to the air cylinder 25. The force generated by the air cylinder 25 is transmitted to the pressing member 12 via the load transmission member 27.

The pressing member 12 has a through hole 12a formed therein. One end of the through hole 12a is opened on the lower surface of the pressing member 12, and the other end of the through hole 12a is connected to the vacuum line 30. A valve (not shown) is provided in the vacuum line 30, and a vacuum is formed in the through hole 12a of the pressing member 12 by opening the valve. When a vacuum is formed in the through hole 12 a with the pressing member 12 in contact with the upper surface of the polishing tape 7, the upper surface of the polishing tape 7 is held by the lower surface of the pressing member 12.

In one embodiment, the polishing tool may be a grindstone instead of the polishing tape 7. In this case, the grindstone is held by a grindstone holder that detachably holds the grindstone, and the grindstone holder is connected to the lower end of the load transmission member 27. The polishing head 11 brings a grindstone into contact with the edge portion of the wafer W to polish the wafer W.

A positioning member 31 is fixed to the load transmission member 27 of the polishing head 11. The polishing head 11 and the positioning member 31 form an integral structure and are moved integrally by the air cylinder 25. The load transmission member 27 is movably connected to a linear motion guide 33 extending along the axis of the wafer holding unit 1. Therefore, the entire moving direction of the polishing head 11 and the positioning member 31 is limited to the direction parallel to the axis of the wafer holder 1. In this embodiment, the axis of the wafer holder 1 extends in the vertical direction.

The polishing head assembly 10 further includes a first distance sensor 23 that measures the moving distance of the polishing head 11. The first distance sensor 23 is fixed to the frame 39 of the polishing head assembly 10 and is configured to measure the moving distance of the polishing head 11. Examples of the first distance sensor 23 include a displacement sensor and the like.

The polishing head assembly 10 further includes a stopper 35 arranged below the positioning member 31 and a stopper moving mechanism 37 connected to the stopper 35. The stopper moving mechanism 37 is a device for moving the stopper 35 in a predetermined direction at a controlled speed. The stopper moving mechanism 37 includes a ball screw mechanism 38 connected to the stopper 35 and a servo motor 36 that drives the ball screw mechanism 38. In this embodiment, the stopper moving mechanism 37 moves the stopper 35 downward during polishing of the wafer W.

The direction in which the stopper moving mechanism 37 moves the stopper 35 is the same as the direction in which the air cylinder 25 pushes the polishing head 11 toward the edge portion of the wafer W. The servo motor 36 includes an encoder (not shown) that detects the number of rotations and the rotation angle of the servo motor 36. The movement distance that the stopper moving mechanism 37 moves the stopper 35 is determined based on the output value of the encoder indicating the number of rotations and the rotation angle of the servo motor 36. The air cylinder 25, the linear guide 33, and the stopper moving mechanism 37 are fixed to the frame 39.

The stopper 35 is located directly below the positioning member 31. Therefore, the downward movement of the polishing head 11 and the positioning member 31 forming the integral structure is restricted by the stopper 35. The stopper 35 has a protrusion 34 on its upper surface. The protrusion 34 is configured to be able to contact the lower surface of the positioning member 31. When the polishing head 11 and the positioning member 31 are lowered by the air cylinder 25, the positioning member 31 contacts the protrusion 34. In one embodiment, the protrusion 34 may be fixed to the lower surface of the positioning member 31.

The edge portion of the wafer W is polished as follows. As shown in FIG. 5, while rotating the wafer W about its axis, a liquid (not shown) such as pure water is supplied to the upper surface of the wafer W. The air cylinder 25 moves the polishing head 11 downward toward the edge portion of the wafer W, and the pressing member 12 of the polishing head 11 presses the polishing tape 7 against the edge portion of the wafer W to polish the edge portion.

During polishing of the wafer W, the air cylinder 25 generates a constant force. Further, during polishing of the wafer W, with the stopper 35 in contact with the lower surface of the positioning member 31, the stopper 35 is lowered by the stopper moving mechanism 37 while restricting the downward movement of the positioning member 31. As the stopper 35 descends, the polishing head 11 and the positioning member 31 descend together. The polishing tape 7 is pressed against the edge of the wafer W by the pressing member 12 of the descending polishing head 11 to polish the wafer W.

During the polishing of the wafer W, a part of the force generated by the air cylinder 25 is transmitted to the stopper 35 via the positioning member 31. That is, a part of the force generated by the air cylinder 25 is applied from the positioning member 31 to the stopper 35. When the force generated by the air cylinder 25 is represented by F, the polishing load applied from the polishing tape 7 to the wafer W is represented by F1, and the load applied from the positioning member 31 to the stopper 35 is represented by F2, F=F1+F2 (unit: N) is established.

One end of the stopper 35 is connected to the ball screw mechanism 38, and the other end is free. As shown in FIGS. 6A and 6B, when the stopper 35 receives a load from the positioning member 31, the stopper 35 slightly tilts downward. FIG. 6A is a diagram showing a state of the stopper 35 when the stopper 35 is not receiving a load from the positioning member 31, and FIG. 6B is a state when the load F2 is applied from the positioning member 31 to the stopper 35. It is a figure which shows the state of the stopper 35 of FIG. In the example shown in FIG. 6B, the inclination amount of the stopper 35 when the load F2 is applied from the positioning member 31 to the stopper 35 is represented by δ (unit: μm). In the present specification, the inclination amount of the stopper 35 is defined as the displacement of the contact point between the positioning member 31 and the stopper 35 when the stopper 35 is inclined downward by the load received from the positioning member 31. In this embodiment, since the protrusion 34 of the stopper 35 contacts the positioning member 31, the inclination amount of the stopper 35 can be represented by the displacement of the protrusion 34 when the stopper 35 receives a load from the positioning member 31. ..

Returning to FIG. 4, the polishing apparatus of this embodiment further includes a stopper tilt amount determination unit 41 that determines the tilt amount of the stopper 35 when a load is applied from the positioning member 31. The stopper tilt amount determining unit 41 tilts the stopper 35 based on the difference between the moving distance of the polishing head 11 measured by the first distance sensor 23 and the moving distance of the stopper moving mechanism 37 moving the stopper 35. Calculate the amount. As described above, the moving distance that the stopper moving mechanism 37 moves the stopper 35 is determined based on the output value of the encoder built in the servo motor 36.

With the stopper 35 in contact with the lower surface of the positioning member 31, the stopper 35 is lowered by the stopper moving mechanism 37 while limiting the downward movement of the positioning member 31. When the moving distance of the polishing head 11 measured by the first distance sensor 23 is represented by d1 (unit: μm), and the moving distance by which the stopper moving mechanism 37 moves the stopper 35 is represented by d2 (unit: μm), When the inclination amount of the stopper 35 is 0 μm, the relationship of d1=d2 is established. On the other hand, when the stopper 35 is inclined downward by δ μm due to the load received from the positioning member 31, the relationship of d1=d2+δ is established. Therefore, the stopper inclination amount determination unit 41 determines, based on the difference between the moving distance d1 of the polishing head 11 measured by the first distance sensor 23 and the moving distance d2 of the stopper moving mechanism 37 moving the stopper 35. The amount of inclination of the stopper 35 can be determined.

The polishing apparatus of the present embodiment includes a second distance sensor 32 that measures the size of the gap between the positioning member 31 and the stopper 35. The second distance sensor 32 is used for the purpose of measuring the size of the gap between the positioning member 31 and the protrusion 34 of the stopper 35 when the positioning member 31 and the stopper 35 are separated from each other.

The polishing apparatus of this embodiment further includes a load determining unit 42 that determines the load applied from the positioning member 31 to the stopper 35. The load determination unit 42 is electrically connected to the stopper inclination amount determination unit 41, and the inclination amount of the stopper 35 determined by the stopper inclination amount determination unit 41 is sent to the load determination unit 42. The load determining unit 42 includes a pre-built database showing the relationship between the load applied from the positioning member 31 to the stopper 35 and the amount of inclination of the stopper 35.

FIG. 7 is a diagram illustrating an example of a database included in the load determining unit 42. The database shown in FIG. 7 is obtained from an experimental result obtained by applying a plurality of different loads from the positioning member 31 to the stopper 35 and measuring the inclination amount of the stopper 35 corresponding to each load. As shown in FIG. 7, the amount of inclination of the stopper 35 changes linearly depending on the magnitude of the load applied to the stopper 35. That is, the amount of inclination of the stopper 35 is proportional to the load applied to the stopper 35. Therefore, the load determination unit 42 can determine the load applied to the stopper 35 from the positioning member 31 based on the above-mentioned database from the amount of inclination of the stopper 35.

When the descending speed of the stopper 35 is constant, the polishing load applied from the polishing tape 7 to the wafer W may change depending on the surface state of the wafer W. For example, when the surface layer of the wafer W is hard, the lowering speed (polishing speed) of the polishing head 11 and the positioning member 31 is lower than the lowering speed of the stopper 35, and the polishing load is large. On the other hand, when the surface layer of the wafer W is soft, the descending speed (polishing speed) of the polishing head 11 and the positioning member 31 is higher than the descending speed of the stopper 35, and the polishing load is small. Such fluctuations in the polishing load make the polishing efficiency unstable. For example, if the polishing load is too large, an excessive pressure is applied to the polishing tape 7, the polishing tape 7 deteriorates, and the polishing efficiency decreases as a result. On the other hand, if the polishing load is too small, the polishing rate will decrease and the polishing efficiency will decrease. That is, there is a polishing load that can optimize the polishing efficiency.

Therefore, as shown in FIG. 4, the polishing apparatus according to the present embodiment includes a stopper speed determination unit 43 that determines the moving speed of the stopper 35 that allows the polishing load to fall within the target range. The stopper speed determination unit 43 is electrically connected to the load determination unit 42, and the load determined by the load determination unit 42 is sent to the stopper speed determination unit 43. The stopper inclination amount determination unit 41, the load determination unit 42, and the stopper speed determination unit 43 basically constitute the control device 50.

During polishing of the wafer W, the air cylinder 25 generates a constant force. The stopper speed determination unit 43 subtracts the load applied to the stopper 35 from the positioning member 31 determined by the load determination unit 42 from the value of the force generated in the air cylinder 25, so that the polishing tape 7 is transferred to the wafer W. The polishing load to be applied is determined, and the moving speed of the stopper 35 that can keep this polishing load within the target range is determined. As described above, since the relationship of F=F1+F2 is established, the stopper speed determination unit 43 controls the polishing load applied from the polishing tape 7 to the wafer W by controlling the load applied from the positioning member 31 to the stopper 35. can do. In one embodiment, the moving speed of the stopper 35 that allows the load applied from the positioning member 31 to the stopper 35 to fall within the target range may be determined. The force generated by the air cylinder 25 can be calculated from the pressure of the gas supplied to the air cylinder 25 and the pressure receiving area of the piston of the air cylinder 25.

Hereinafter, a specific configuration for maximizing the polishing rate of the wafer W will be described. The controller 50 is configured to learn a polishing recipe capable of maximizing the polishing rate of the wafer W, including various elements, by a machine learning algorithm such as deep learning, and generate an optimum polishing recipe. ing.

The polishing method inputs the type of polishing tool specified from the size of the abrasive grains, the value of the polishing width and the value of the polishing depth of the substrate to be polished by the polishing tool, into the model constructed by the machine learning algorithm. And a step of outputting from the model a polishing recipe having polishing conditions including conditions for polishing the wafer with the contact pressure of the polishing tool on the wafer when the polishing power of the polishing tool is maximum. The polishing apparatus polishes the wafer W according to this polishing recipe.

The polishing width corresponds to the polishing width Wp shown in FIG. 13, and the polishing depth corresponds to the polishing depth Dp shown in FIG. The type of polishing tool used is determined based on the size (that is, particle size) of the abrasive grains of the polishing tool. The polishing tool may be a polishing tape or a whetstone.

A deep learning method (deep learning method) is suitable as the machine learning algorithm. The deep learning method is a learning method based on a neural network in which hidden layers are multilayered. In this specification, machine learning using a neural network composed of an input layer, two or more hidden layers, and an output layer is referred to as deep learning. By using the deep learning method, the polishing recipe generated based on the ability (knowledge and experience) of the operator so far can be compared with the computer based on the polishing depth, the polishing width, and the type of polishing tool used. Can be generated by.

Generally, the polishing amount Q is represented by Q=c×P×v×t as an empirical rule. Here, "c" is a polishing constant, "P" is a contact pressure of the polishing tool with respect to the wafer, "v" is a relative speed between the polishing tool and the wafer, and "t" is time. Therefore, in order to obtain the maximum polishing rate, the polishing load applied to the wafer from the polishing tool is determined based on the value of the contact pressure at which the polishing force of the polishing tool is maximum, and the rotation speed of the wafer can be determined. Determine the maximum speed.

FIG. 8 is a diagram showing an example of performance characteristics of the polishing tape. In FIG. 8, the horizontal axis represents the contact pressure [MPa] of the polishing tape with respect to the wafer, and the vertical axis represents the polishing force of the polishing tape in percentage (%). In FIG. 8, a first polishing tape (see a solid line), a second polishing tape (see a dotted line), and a third polishing tape (see a dashed line) are shown. The grain size of the abrasive grains held by the first polishing tape is smaller than the grain size of the abrasive grains held by the second polishing tape. The grain size of the abrasive grains held by the second polishing tape is smaller than the grain size of the abrasive grains held by the third polishing tape.

As shown in FIG. 8, when the polishing power is 100%, the polishing power of the polishing tape becomes maximum. The performance characteristics of the polishing tape differ depending on the type of polishing tape (more specifically, the size of abrasive grains). Therefore, the control device 50 determines the contact pressure of the polishing tape with respect to the wafer when the polishing force of the polishing tape is maximum, based on the type of polishing tape used (more specifically, the grain size of the abrasive grains). , And the correlation between the polishing force and the contact pressure (that is, the database shown in FIG. 8). The controller 50 determines the polishing load applied to the wafer from the polishing tape based on this contact pressure. The above correlation is stored in the control device 50 as a database.

As described above, the cross-sectional shape of the peripheral portion of the wafer W in the thickness direction of the wafer W is not uniform. Therefore, the polishing area of the wafer W to be polished changes as the polishing of the wafer W progresses. Therefore, in this embodiment, it is necessary to change the polishing load to be applied from the polishing tape 7 to the wafer W based on the change in the polishing area of the wafer W so that the polishing force of the polishing tool is maximized.

The polishing recipe generated by the control device 50 includes at least the polishing load applied to the wafer W from the polishing tool (the polishing tape 7 in this embodiment) and the rotation speed of the wafer W. The polishing conditions include conditions for changing the polishing load according to the polishing area of the wafer W that changes with the progress of polishing of the wafer W. The polishing conditions may include conditions for changing the rotation speed of the wafer W according to the polishing area of the wafer W that changes with the progress of polishing of the wafer W.

As shown in FIG. 13, the wafer W may be formed with a plurality of layers having different hardnesses. In this case, it is necessary to change the polishing load applied from the polishing tape 7 to the wafer W according to the hardness of each of the plurality of layers. Therefore, the polishing conditions include conditions for changing the polishing load according to the plurality of layers having different hardness of the wafer W. The polishing conditions may include conditions that change the rotation speed of the wafer W according to the hardness of each of the plurality of layers.

The model for generating the optimum polishing recipe is constructed based on the polishing information data including the pre-polishing acquisition data and the during-polishing acquisition data. The control device 50 takes in sensor data measured by a sensor provided in the polishing device as an element when building a model.

The control device 50 is electrically connected to the stage torque sensor 4 (see FIG. 3) and acquires the motor torque (that is, the rotation torque) detected by the stage torque sensor 4. The controller 50 is electrically connected to the stage motor 3 and acquires the rotation speed of the wafer stage 2 based on the voltage applied to the stage motor 3. The control device 50 takes in the rotation speed and the rotation torque of the wafer stage 2 as elements when constructing the model.

The control device 50 acquires the position information (that is, the polishing depth) of the pressing member 12 in the thickness direction of the wafer W based on the moving distance of the polishing head 11 measured by the first distance sensor 23. Further, the control device 50 acquires the descending speed (polishing speed) of the polishing head 11 based on the measurement value of the first distance sensor 23. The controller 50 acquires the polishing load applied to the wafer W from the polishing tape 7 based on the result calculated by the load determining unit 42. The control device 50 takes in the position information of the pressing member 12, the lowering speed of the polishing head 11, and the polishing load applied to the wafer W from the polishing tape 7 as elements for constructing the model.

FIG. 9 is a diagram showing a polishing area Ap and a notch portion N of the wafer W. 10A and 10B are views showing contact with the notch portion N of the polishing tape 7. As shown in FIGS. 9 and 10, the bevel portion of the wafer W may be provided with a notch portion N in order to facilitate identification and alignment of the crystal direction of the wafer W. The pressing member 12 of the polishing head 11 presses the polishing tape 7 against the edge portion of the wafer W. Therefore, when the wafer W rotates in the direction of the arrow in FIG. 9, the pressing member 12 periodically contacts the notch portion N via the polishing tape 7. The polishing head 11 vibrates in the vertical direction due to the contact with the notch portion N. As a result, the polishing load applied to the wafer W from the polishing tape 7 changes according to the vibration of the polishing head 11. Such a change in the polishing load affects the polishing rate.

FIG. 10A shows an angle (contact angle) θ _S formed by the tangent line in the vertical direction passing through the contact point with the notch portion N of the polishing tape 7 and the polishing tape 7. FIG. 10B shows an angle (contact angle) θ _D formed by the tangent line in the vertical direction passing through the contact point with the notch portion N of the polishing tape 7 and the polishing tape 7. As the polishing of the wafer W by the polishing tape 7 progresses, the contact angle between the polishing tape 7 and the notch portion N increases (θ _S <θ _D ). Under a condition that the polishing load applied from the polishing tape 7 to the wafer W and the rotation speed of the wafer W are constant, when the contact angle with the notch portion N of the polishing tape 7 increases, the vibration of the polishing head 11 increases. That is, the vibration of the polishing head 11 increases as the polishing of the wafer W progresses.

When the polishing head 11 vibrates, the first distance sensor 23 measures the change in the vertical movement distance of the pressing member 12. Therefore, the control device 50 acquires the vibration of the polishing head 11 based on the change in the moving distance of the polishing head 11 measured by the first distance sensor 23. In one embodiment, the polishing apparatus may include a vibration sensor (not shown) attached to the frame 39 and electrically connected to the controller 50. The control device 50 takes in the vibration of the polishing head 11 as an element when constructing the model.

The controller 50 is electrically connected to the reel motors 17 and 18, and acquires the moving speed of the polishing tape 7 based on the voltages applied to the reel motors 17 and 18, respectively. In one embodiment, the polishing apparatus may include a speed sensor (not shown) that measures the moving speed of the polishing tape 7 and is electrically connected to the controller 50. When a grindstone is used as the polishing tool, the control device 50 may acquire the movement speed (that is, the rotation speed) of the grindstone based on the rotation speed of a motor (not shown) to which the grindstone is fixed. The control device 50 acquires the moving speed of the polishing tool as an element when building the model.

As shown in FIG. 3, a liquid supply line 26 is connected to the liquid supply nozzle 24. A temperature sensor 28 for measuring the temperature of the liquid flowing through the liquid supply line 26 and a liquid supply line 26 are connected to the liquid supply line 26. A flow rate sensor 29 that measures the flow rate of the liquid flowing through the supply line 26 is attached. The control device 50 is electrically connected to the temperature sensor 28 and the flow rate sensor 29, and acquires the flow rate and temperature of the liquid flowing through the liquid supply line 26. The controller 50 acquires the flow rate and temperature of the liquid as an element when building the model.

Polishing information data including data acquired before polishing the wafer W (acquired data before polishing) and data acquired during polishing of the wafer W (acquired data during polishing) is input to the control device 50. The control device 50 builds a model based on the polishing information data.

The data acquired during polishing includes the type of polishing tool used, the rotation speed and rotation torque of the wafer stage 2, the vibration of the polishing head 11, the polishing head 11 in the thickness direction of the wafer W (more specifically, the pressing member 12). ) Position, the lowering speed of the polishing head 11, the polishing load applied from the polishing tape 7 to the wafer W, the moving speed of the polishing tape 7, and the flow rate and temperature of the liquid supplied from the liquid supply nozzle 24 onto the wafer W. ..

The pre-polishing acquisition data includes image data of the bevel portion of the wafer W including the notch portion N. This image data is acquired by the image acquisition device 55 (see FIG. 2) electrically connected to the control device 50. An example of the image acquisition device 55 is an optical profiler configured to acquire the three-dimensional shape of the wafer W based on the reflection of the light with which the wafer W is irradiated. The image acquisition device 55 in FIG. 2 is schematically drawn. The image data of the bevel portion of the wafer W may be acquired before the wafer W is placed on the wafer stage 2. By acquiring the pre-polishing acquisition data, the control device 50 can calculate the polishing area of the wafer W corresponding to the polishing depth of the wafer W to be polished.

In the model construction, the explanatory variables of the learning data are polishing information data (including the pre-polishing acquisition data and the during-polishing acquisition data), and the objective variable of the learning data is the current polishing when the wafer W is actually being polished. The numerical value indicating the deviation of the rate (actual polishing rate) from the theoretically calculated predicted polishing rate, for example, 1: very poor, 2: bad, 3: good, 4: good, 5: very good 5 It can be given as a graded rating. In one embodiment, the numerical value indicating the deviation is not limited to the 5-level evaluation, and may be, for example, the 3-level evaluation or the 10-level evaluation. The deviation of the actual polishing rate from the predicted polishing rate may be expressed as a numerical value indicating the polishing efficiency.

An optimum polishing recipe can be automatically generated from the result of the polishing efficiency determination. For example, if the determination result is 1: "very bad", the controller 50 causes the contact pressure of the current polishing tape on the wafer to increase (or decrease) by a predetermined first numerical value (or ratio). , The load applied to the stopper 35 is calculated based on the amount of inclination of the stopper 35 (see FIG. 7), and the operation of the stopper moving mechanism 37 is controlled to control the movement of the stopper 35. The controller 50 may control the rotation speed of the wafer stage 2.

If the determination result is 2: "bad", the control device 50 controls the stopper 35 to increase (or decrease) the contact pressure of the current polishing tape with respect to the wafer by a predetermined second numerical value (or ratio). The load applied to the stopper 35 is calculated based on the amount of inclination (see FIG. 7), the operation of the stopper moving mechanism 37 is controlled, and the movement of the stopper 35 is controlled. The controller 50 may control the rotation speed of the wafer stage 2.

If the determination result is 5: "very good", the controller 50 does not change the current contact pressure of the polishing tape with respect to the wafer. This step is repeated until the optimum determination result is output. The first numerical value (or ratio) is larger (or smaller) than the second numerical value (or ratio).

As described above, the control device 50 analyzes the cause of the deviation of the actual polishing rate from the predicted polishing rate, and changes the actual polishing rate so that the actual polishing rate matches the predicted polishing rate (at least, The polishing load and/or the rotation speed of the wafer W) are corrected. The control device 50 reflects the content of the above-mentioned elements that are the basis of the correction of the polishing rate in the content of creating the polishing recipe to generate the optimum polishing recipe.

FIG. 11 is a schematic diagram showing an embodiment of a model for generating a polishing recipe. The model is a neural network having an input layer 301, a plurality of hidden layers (also referred to as intermediate layers) 302, and an output layer 303. The model shown in FIG. 11 has four hidden layers 302, but the configuration of the model is not limited to the embodiment shown in FIG. In the input layer 301 of the model, the polishing width, the polishing depth, and the type of polishing tool used are input.

When the polishing depth and the polishing width and the type of the polishing tool used (more specifically, the size of the abrasive grain) are input to the model constructed as described above, the computer configures a neural network. The calculation is executed according to the algorithm of the multilayer perceptron, and the optimum polishing recipe is output. The polishing apparatus polishes the wafer W based on the output polishing recipe.

The model using the machine learning algorithm is continuously updated on the basis of newly obtained data in the polishing apparatus that is operating using the model. As a result, even if the polishing conditions have changed significantly since the model was first built, the model that is updated according to the actual situation can continuously output the optimum polishing recipe. Become.

The control device 50 includes a program for building and updating a model, a storage device 110 in which the model is stored, and a processing device 120 (such as a GPU or a CPU) that executes an operation according to the program.

The control device 50 is composed of at least one computer. The at least one computer may be one server or multiple servers. The model for determining the polishing efficiency based on at least the polishing information data is stored in the storage device 110 of the control device 50.

FIG. 12 is a schematic diagram showing an embodiment of a computer forming at least a part of the control device 50 shown in FIGS. 4 and 5. The computer includes a storage device 110 in which programs and data are stored, a processing device 120 such as a CPU (central processing unit) or GPU (graphic processing unit) that performs an operation according to a program stored in the storage device 110, and data. , A program, and an input device 130 for inputting various information to the storage device 110, an output device 140 for outputting a processing result or processed data, and a connection to a communication network such as the Internet or a local area network. The communication device 150 of FIG.

The storage device 110 includes a main storage device 111 accessible by the processing device 120 and an auxiliary storage device 112 that stores data and programs. The main storage device 111 is, for example, a random access memory (RAM), and the auxiliary storage device 112 is a storage device such as a hard disk drive (HDD) or a solid state drive (SSD).

The input device 130 includes a keyboard and a mouse, and further includes a recording medium reading device 132 for reading data from the recording medium, and a recording medium port 134 to which the recording medium is connected. The recording medium is a computer-readable recording medium that is a non-transitory tangible material, and is, for example, an optical disk (eg, CD-ROM, DVD-ROM) or a semiconductor memory (eg, USB flash drive, memory card). is there. Examples of the recording medium reading device 132 include an optical drive such as a CD-ROM drive and a DVD-ROM drive, and a memory reader. An example of the recording medium port 134 is a USB port. The program and/or data stored in the recording medium is introduced into the computer via the input device 130 and stored in the auxiliary storage device 112 of the storage device 110. The output device 140 includes a display device 141 and a printing device 142.

The trained model constructed by the machine learning algorithm is stored in the storage device 110. This learned model is a neural network having an input layer 301, a plurality of hidden layers (also referred to as intermediate layers) 302, and an output layer 303, as shown in FIG. The computer operates according to a program electrically stored in the storage device 110. That is, the computer models the type of the polishing tool specified from the size of the abrasive grains, the value of the polishing width and the value of the polishing depth of the wafer W polished by the polishing tool, by a machine learning algorithm. And a step of outputting the polishing recipe from the output layer 303 by executing an operation according to the algorithm of the multilayer perceptron that constitutes the neural network. In addition, the computer performs the steps of building the model and updating the model. A program for causing a computer to execute these steps is recorded in a computer-readable recording medium that is a non-transitory tangible material, and is provided to the computer via the recording medium. Alternatively, the program and the model may be input to the computer from the communication device 150 via a communication network such as the Internet.

The above-described embodiments are described for the purpose of enabling a person having ordinary knowledge in the technical field to which the present invention belongs to implement the present invention. Various modifications of the above-described embodiment can be naturally made by those skilled in the art, and the technical idea of the present invention can be applied to other embodiments. Therefore, the present invention is not limited to the described embodiments, but is to be construed in the broadest scope according to the technical idea defined by the claims.

1 Wafer holder (Substrate holder)
2 Wafer stage (substrate stage)
3 Stage Motor 4 Stage Torque Sensor 7 Polishing Tape 10 Polishing Head Assembly 11 Polishing Head 12 Pressing Member 12a Through Hole 14 Unwinding Reel 15 Winding Reel 17, 18 Reel Motor 20 Tape Feeder 21, 22 Guide Roller 23 First Distance sensor 24 Liquid supply nozzle 25 Air cylinder 26 Liquid supply line 27 Load transfer member 27a Pressing member holder 28 Temperature sensor 29 Flow rate sensor 30 Vacuum line 31 Positioning member 32 Second distance sensor 33 Linear guide 34 Projection 35 Stopper 36 Servo Motor 37 Stopper moving mechanism 38 Ball screw mechanism 39 Frame 41 Stopper inclination amount determining unit 42 Load determining unit 43 Stopper speed determining unit 50 Control device 55 Image acquisition device 110 Storage device 120 Processing device 130 Input device 140 Output device 150 Communication device 301 Input Layer 302 Hidden layer 303 Output layer

Claims (7)

Type of the polishing tool specified from the size of the abrasive grains, and the value of the polishing width and the polishing depth of the substrate to be polished by the polishing tool, input to the model constructed by the machine learning algorithm,
A polishing method, wherein a polishing recipe having polishing conditions including conditions for polishing the substrate with a contact pressure of the polishing tool with respect to the substrate when the polishing power of the polishing tool is maximized is output from the model. The polishing method according to claim 1, wherein
The polishing recipe includes at least a polishing load applied to the substrate from the polishing tool,
The polishing condition includes a condition for changing the polishing load according to a polishing area of the substrate which changes with the progress of polishing of the substrate. The polishing method according to claim 1 or 2, wherein
The polishing recipe includes at least a polishing load applied to the substrate from the polishing tool,
The polishing condition includes a condition for changing the polishing load according to a plurality of layers having different hardness of the substrate. The polishing method according to any one of claims 1 to 3,
In the construction of the model, the explanatory variables of the learning data are given as polishing information data including the pre-polishing acquisition data and the polishing-in-process acquired data, and the objective variable is a numerical value indicating the deviation of the actual polishing rate from the predicted polishing rate. Given polishing method. The polishing method according to claim 4, wherein
The pre-polishing acquisition data includes image data of an edge portion of the substrate generated by an image acquisition device,
The data acquired during polishing includes the type of polishing tool used, the rotation speed and rotation torque of the substrate holding unit that holds the substrate, the vibration of the polishing head that presses the polishing tool against the substrate, and the thickness direction of the substrate. A polishing method including the position of the polishing head, the descending speed of the polishing head, the polishing load applied to the substrate from the polishing tool, the moving speed of the polishing tool, and the flow rate and temperature of the liquid supplied onto the substrate. .. A substrate holding unit that holds the substrate;
A polishing head for pressing a polishing tool onto the substrate,
A computer having a model constructed by a machine learning algorithm,
The computer is
A storage device in which the model is stored,
The type of the polishing tool specified from the size of the abrasive grains, the value of the polishing width and the value of the polishing depth of the substrate to be polished by the polishing tool are input to the model to polish the polishing tool. A processing device that executes a calculation for outputting a polishing recipe having a polishing condition from the model, including a condition for polishing the substrate with a contact pressure of the polishing tool with respect to the substrate when the force is maximized. The polishing device. An input layer, a plurality of hidden layers, an input layer of a model consisting of a neural network having an output layer, the type of polishing tool specified from the size of the abrasive grains, and the substrate to be polished by the polishing tool A step of inputting a polishing width value and a polishing depth value,
The substrate is polished from the output layer by a contact pressure of the polishing tool with respect to the substrate when the polishing force of the polishing tool is maximized, by performing an operation according to an algorithm of a multilayer perceptron forming the neural network. A computer-readable recording medium recording a program for causing a computer to execute a step of outputting a polishing recipe having polishing conditions including the conditions.

Images