Dressing condition determination device and dressing condition determination method

Abstract

The present invention relates to a technique for determining a dressing condition for dressing a polishing pad used for polishing a workpiece such as a wafer, a substrate, or a panel. A dressing condition determination device (60) is provided with: a storage device (60a) for storing an estimation model (62) constructed by machine learning; and a processor (60b) that executes calculations in accordance with the algorithm of the estimation model (62). The estimation model (62) is configured so as to output a dressing condition for achieving a target trajectory when a target trajectory image representing the target trajectory of a plurality of abrasive grains constituting the dressing surface (51a) on the polishing pad (2) is input.

Description

Technical Field

[0001] This invention relates to a technique for determining the dressing conditions for polishing pads used to dress workpieces such as wafers, substrates, or panels. Background Technology

[0002] Chemical mechanical polishing (CMP) is a process in which an polishing slurry is supplied to a polishing pad, and a workpiece (such as a wafer, substrate, panel, etc.) is slidably brought into contact with the polishing pad to polish the workpiece. The polishing slurry is typically a slurry containing abrasive particles. A polishing apparatus for performing CMP includes: a polishing table supporting a polishing pad with polishing surfaces; and a polishing head for pressing the workpiece onto the polishing pad.

[0003] The grinding apparatus grinds a workpiece as described below. While the grinding table and grinding pad rotate as a single unit, a grinding fluid (typically a slurry) is supplied to the grinding surface of the grinding pad. The grinding head rotates the workpiece while pressing its surface against the grinding surface of the grinding pad. The workpiece slides into contact with the grinding pad in the presence of the grinding fluid. The surface of the workpiece is ground by the chemical action of the grinding fluid, the abrasive particles contained in the grinding fluid, and / or the mechanical action of the grinding pad.

[0004] The grinding efficiency (also known as the removal rate) of a workpiece varies depending on various factors. The condition of the grinding surface of the grinding pad is one of these factors. Specifically, the workpiece slides in contact with the grinding surface while the grinding fluid is held on it. The grinding pad has numerous tiny protrusions on its grinding surface. These protrusions facilitate the mechanical grinding of the workpiece. Therefore, the condition of the grinding surface of the grinding pad affects the grinding efficiency of the workpiece.

[0005] Therefore, in order to improve and stabilize the grinding rate of the workpiece, the grinding surface of the grinding pad is dressed (hereinafter referred to as pad dressing) before grinding the workpiece. Pad dressing is performed using a dressing tool with a dressing surface composed of abrasive particles such as diamond particles. More specifically, by rotating the grinding pad and the dressing tool separately, the dressing surface of the dressing tool is pressed against the grinding surface of the grinding pad, thereby roughening the grinding surface (i.e., forming many tiny protrusions on the grinding surface).

[0006] Existing technical documents

[0007] Patent documents

[0008] Patent Document 1: Japanese Patent Application Publication No. 2010-76049

[0009] The problem the invention aims to solve

[0010] Pad dressing is performed under predetermined dressing conditions. The dressing conditions should ensure a uniform asperity across the entire grinding surface of the pad. Therefore, the dressing conditions should be determined in a way that achieves this grinding surface. However, in the past, users determined dressing conditions through trial and error, sometimes without obtaining the desired dressing results. Summary of the Invention

[0011] Therefore, the present invention provides a technique that can automatically determine the dressing conditions that can achieve the desired state of the polished surface.

[0012] (Solutions)

[0013] One approach provides a dressing condition determining apparatus for determining dressing conditions for dressing a polishing pad by pressing the dressing surface of a rotating dresser against a rotating polishing pad, comprising: a storage device storing a estimation model constructed by machine learning; and a processor performing calculations according to an algorithm of the estimation model, the estimation model being configured to output dressing conditions for achieving the target trajectory upon input of a target trajectory image, the target trajectory image representing the target trajectory of a plurality of abrasive grains constituting the dressing surface on the polishing pad.

[0014] One approach is that the dressing conditions include at least one of the following: the rotational speed of the dresser, the rotational speed of the abrasive pad, the oscillation speed of the dresser on the abrasive pad, and the dressing time.

[0015] One approach is that the target trajectory image is a simulated image of the trajectories of the plurality of abrasive grains generated by performing a simulation, which is a simulation of the action of pressing the dressing surface of the rotating dresser against the rotating abrasive pad to dress the abrasive pad.

[0016] One approach is that the inference model is a learned model constructed using multiple trimming conditions and training data and through machine learning, the training data comprising multiple trajectory images obtained through trimming simulations performed under the multiple trimming conditions.

[0017] One approach is that the estimation model comprises: an encoder that extracts features from the target trajectory image; a decoder that reconstructs the target trajectory image from the features; and a regression analyzer that performs regression analysis on the extracted features and outputs the trimming conditions.

[0018] One approach is that the target trajectory is the trajectory of the abrasive grains evenly distributed on the abrasive pad, and the intersection points of the trajectories of the abrasive grains are evenly distributed on the trajectory of the abrasive pad.

[0019] One approach provides a method for determining trimming conditions, which are used to trim a grinding pad by pressing the trimming face of a rotating trimmer against a rotating grinding pad. A target trajectory image is input into an estimation model constructed by machine learning, the target trajectory image representing the target trajectory of a plurality of abrasive grains constituting the trimming face on the grinding pad. Trimming conditions for achieving the target trajectory are output from the estimation model.

[0020] One approach is that the dressing conditions include at least one of the following: the rotational speed of the dresser, the rotational speed of the abrasive pad, the oscillation speed of the dresser on the abrasive pad, and the dressing time.

[0021] One approach is that the target trajectory image is a simulated image of the trajectories of the plurality of abrasive grains generated by performing a simulation, which is a simulation of the action of pressing the dressing surface of the rotating dresser against the rotating abrasive pad to dress the abrasive pad.

[0022] One approach is that the inference model is a learned model constructed using multiple trimming conditions and training data and through machine learning, the training data comprising multiple trajectory images obtained through trimming simulations performed under the multiple trimming conditions.

[0023] One approach is that the estimation model comprises: an encoder that extracts features from the target trajectory image; a decoder that reconstructs the target trajectory image from the features; and a regression analyzer that performs regression analysis on the extracted features and outputs the trimming conditions.

[0024] One approach is that the target trajectory is the trajectory of the abrasive grains evenly distributed on the abrasive pad, and the intersection points of the trajectories of the abrasive grains are evenly distributed on the trajectory of the abrasive pad.

[0025] (Invention effect)

[0026] The aforementioned dressing condition determining device and method can use a presupposition model to automatically determine the dressing conditions that can achieve the desired state of the ground surface. Attached Figure Description

[0027] Figure 1 This is a schematic diagram illustrating one embodiment of the grinding apparatus.

[0028] Figure 2 This is a schematic diagram illustrating an example of the trajectory of abrasive grains forming the trimmed surface on the abrasive surface of the abrasive pad.

[0029] Figure 3 This is a diagram representing an example of a target trajectory image.

[0030] Figure 4 This is a schematic diagram illustrating one implementation of the presumption model.

[0031] Figure 5 This is a schematic diagram representing an example of training data containing multiple trimming conditions and corresponding multiple trajectory images.

[0032] Figure 6 This is a diagram illustrating the training phase of building an inference model using machine learning with training data.

[0033] Figure 7 This is a schematic diagram illustrating a practical stage in using a presuppositional model constructed through machine learning to determine the trimming conditions for achieving the trajectory of abrasive particles on a target trajectory image. Detailed Implementation

[0034] Hereinafter, embodiments of the present invention will be described with reference to the figures. Figure 1 This is a schematic diagram illustrating one embodiment of a grinding apparatus. The grinding apparatus is an example of a device used for chemical mechanical grinding of a wafer W, a workpiece used in the manufacture of semiconductor devices. Figure 1 As shown, the polishing apparatus includes: a polishing table 5 supporting a polishing pad 2 having a polishing surface 2a; a polishing head 7 pressing a wafer W against the polishing surface 2a; a polishing slurry supply nozzle 8 supplying polishing slurry (e.g., a slurry containing polishing particles) to the polishing surface 2a; and an operation control unit 30 controlling the operation of the polishing apparatus.

[0035] The polishing head 7 is configured to hold the wafer W on its lower surface. The wafer W has a polishing film. The following embodiments use a wafer as an example of a workpiece, but the workpiece is not limited to a wafer and can also be used for circular substrates, rectangular substrates, panels, etc., used in the manufacture of semiconductor devices.

[0036] The grinding apparatus further includes: a support shaft 14; a grinding head swing arm 16 connected to the upper end of the support shaft 14; and a grinding head shaft 18 rotatably supported on the free end of the grinding head swing arm 16. The grinding head 7 is fixed to the lower end of the grinding head shaft 18. A grinding head rotation mechanism (not shown) equipped with a motor or the like is disposed within the grinding head swing arm 16. This grinding head rotation mechanism is connected to the grinding head shaft 18 and configured to rotate the grinding head shaft 18 and the grinding head 7 in the direction indicated by the arrow.

[0037] The grinding head shaft 18 is connected to a grinding head lifting mechanism (including a ball screw mechanism, etc.) not shown in the figure. This grinding head lifting mechanism is configured to allow the grinding head shaft 18 to move up and down relative to the grinding head swing arm 16. By moving the grinding head shaft 18 up and down, as shown by the arrow, the grinding head 7 can move up and down relative to the grinding head swing arm 16 and the grinding table 5.

[0038] The polishing apparatus further includes a table rotation motor 21 that rotates the polishing pad 2 and the polishing table 5 about their axes. The table rotation motor 21 is positioned below the polishing table 5, which is connected to it via a table shaft 5a. The polishing table 5 and the polishing pad 2 can rotate about the table shaft 5a via the table rotation motor 21 in the direction indicated by the arrow. The polishing pad 2 is attached to the upper surface of the polishing table 5. The exposed surface of the polishing pad 2 forms the polishing surface 2a of the polishing wafer W.

[0039] The polishing of wafer W is performed as follows. Wafer W is held in polishing head 7 with its polished surface facing downwards. While polishing head 7 and polishing table 5 are rotated, polishing slurry (e.g., a slurry containing abrasive particles) is supplied to the polishing surface 2a of polishing pad 2 from polishing slurry supply nozzle 8 located above polishing table 5. Polishing pad 2 rotates integrally with polishing table 5 around its central axis. Polishing head 7 is moved to a predetermined height by polishing head lifting mechanism (not shown). Furthermore, while maintaining the predetermined height, polishing head 7 presses wafer W against polishing surface 2a of polishing pad 2. Wafer W rotates integrally with polishing head 7. With polishing slurry present on polishing surface 2a of polishing pad 2, wafer W slides in contact with polishing surface 2a. The surface of wafer W is polished by the combination of the chemical action of polishing slurry, the abrasive particles contained in polishing slurry, and / or the mechanical action of polishing pad 2.

[0040] The motion control unit 30 comprises at least one computer. The motion control unit 30 includes: a storage device 30a storing a program; and a processing unit 30b executing operations according to the commands contained in the program. The storage device 30a includes: a main storage device such as random access memory (RAM), and an auxiliary storage device such as a hard disk drive (HDD) or a solid-state drive (SSD). Examples of the processing unit 30b include: a CPU (central processing unit) and a GPU (graphics processing unit). However, the specific configuration of the motion control unit 30 is not limited to these examples.

[0041] The grinding apparatus includes a dressing mechanism 50 for dressing the grinding surface 2a of the grinding pad 2. The dressing mechanism 50 includes: a dresser 51 that slides in contact with the grinding surface 2a of the grinding pad 2; a dresser shaft 52 connected to the dresser 51; and a dresser swing arm 55 that supports the dresser shaft 52 for free rotation. The lower surface of the dresser 51 forms the dressing surface 51a, which is composed of abrasive grains (e.g., diamond particles).

[0042] The dresser shaft 52 is connected to a dresser pressing mechanism (e.g., including an air cylinder) disposed within the dresser swing arm 55. This dresser pressing mechanism is configured to press the dressing surface 51a of the dresser 51 relative to the grinding surface 2a of the grinding pad 2 via the dresser shaft 52. Furthermore, the dresser shaft 52 is connected to a dresser rotating mechanism (e.g., including an electric motor) disposed within the dresser swing arm 55. This dresser rotating mechanism is configured to rotate the dresser 51 in the direction indicated by the arrow via the dresser shaft 52.

[0043] The dressing of the grinding surface 2a of the grinding pad 2 is performed as follows. The grinding pad 2 and the grinding table 5 rotate together via the table rotation motor 21, while pure water is supplied to the grinding surface 2a from a pure water supply nozzle (not shown). The dresser 51 rotates around the dresser shaft 52 via a dresser rotation mechanism (not shown), and the dressing surface 51a of the dresser 51 is pressed against the grinding surface 2a by a dresser pressing mechanism (not shown). With pure water present on the grinding surface 2a, the dresser 51 slides in contact with the grinding surface 2a. As the dresser 51 rotates, the dresser swing arm 55 rotates around the support shaft 58, causing the dresser 51 to swing in the radial direction of the grinding surface 2a. Thus, the grinding pad 2 is slightly cut by the dressing surface 51a of the dresser 51, thereby dressing (regenerating) the grinding surface 2a.

[0044] The polishing pad 2 is trimmed before it is first used to polish a wafer, during the polishing of a wafer, or after polishing a wafer and before polishing the next wafer.

[0045] The dressing of the grinding pad 2 is carried out under predetermined dressing conditions. The dressing conditions include at least one of the following: the rotational speed of the dresser 51, the rotational speed of the grinding pad 2 (grinding table 5), the dressing time, and the oscillation speed of the dresser 51 on the grinding surface 2a.

[0046] During the dressing of the abrasive pad 2, the abrasive grains constituting the dressing surface 51a move on the abrasive surface 2a of the abrasive pad 2 while scratching the abrasive surface 2a, thus roughening the abrasive surface 2a. Figure 2 This is a schematic diagram showing an example of the trajectory of the abrasive grains constituting the dressing surface 51a on the abrasive surface 2a of the abrasive pad 2. Figure 2 The symbol T shown represents the trajectory of the abrasive grains on the abrasive surface 2a. For example... Figure 2 As shown, the grinding surface 2a is scratched by the abrasive grains, and a portion of the grinding surface 2a bulges to form a protrusion 40. Similarly, at the intersection of multiple trajectories T of the abrasive grains, a portion of the grinding surface 2a bulges to form a protrusion 40.

[0047] At the intersection of trajectory T, multiple trajectories intersect at an angle greater than 0 degrees. When the intersection angle is 0 degrees, the protrusion 40 formed by one abrasive grain will be removed by another abrasive grain. Therefore, the intersection angle of the abrasive grain trajectory T is greater than 0 degrees, for example, in the range of 10 to 170 degrees.

[0048] The numerous protrusions 40 formed by the abrasive grains of the dressing device 51 facilitate the mechanical polishing of the wafer. Therefore, the condition of the polishing surface 2a of the polishing pad 2 affects the wafer polishing rate. To improve and stabilize the wafer polishing rate, the abrasive grain trajectory should be uniformly distributed on the polishing pad 2, and the intersections of the abrasive grain trajectories should be uniformly distributed on the polishing pad 2. The trajectory of the abrasive grains within the polishing surface 2a is altered by the dressing conditions.

[0049] Therefore, in order to achieve the desired state of the grinding surface 2a of the grinding pad 2, such as... Figure 1 As shown, the grinding apparatus further includes a dressing condition determining device 60 for determining dressing conditions. The dressing condition determining device 60 is connected to the motion control unit 30.

[0050] The adjustment condition determination device 60 comprises at least one computer. The adjustment condition determination device 60 includes: a storage device 60a storing a program and a estimation model 62; and a processor 60b performing calculations according to the commands contained in the program and the algorithm of the estimation model 62. The storage device 60a includes: a main storage device such as random access memory (RAM), and an auxiliary storage device such as a hard disk drive (HDD) or a solid-state drive (SSD). Examples of the processor 60b include: a CPU (central processing unit) and a GPU (graphics processing unit). However, the specific configuration of the adjustment condition determination device 60 is not limited to these examples.

[0051] In one embodiment, the motion control unit 30 and the trim condition determination device 60 may also be integrally formed. For example, the motion control unit 30 and the trim condition determination device 60 may also be composed of a single computer.

[0052] The presupposition model 62 is configured to output trimming conditions for achieving the target trajectory when given an image representing the target trajectory of multiple abrasive grains constituting the trimming surface 51a on the abrasive pad 2. The presupposition model 62 is a learning-completed model constructed through machine learning. An example of the presupposition model 62 is a neural network.

[0053] The target trajectory image is a simulated image generated by simulating the action of pressing the trimming surface 51a of the rotating trimmer 51 against the rotating grinding pad 2 to trim the grinding pad 2. The above simulation can also be performed by the trimming condition determining device 60, or by other simulation devices.

[0054] Figure 3This is an example diagram showing a target trajectory image. In this example, the trajectories of the target abrasive grains appearing in the target trajectory image are evenly distributed on the abrasive pad 2, and the intersections of the abrasive grain trajectories are also evenly distributed on the abrasive pad 2. In other words, in the abrasive surface 2a of the dressing pad 2, multiple abrasive grains constituting the dressing surface 51a of the dressing device 51 move evenly on the abrasive surface 2a, forming a trajectory with a geometric pattern. Multiple abrasive grains move on the abrasive surface 2a with evenly distributed travel distances.

[0055] The target trajectory image is input into the trimming condition determination device 60. The trimming condition determination device 60 inputs the target trajectory image into the estimation model 62 and outputs trimming conditions from the estimation model 62 for achieving the target trajectory of the abrasive grains appearing in the target trajectory image. The trimming conditions include at least one of the following: rotational speed of the trimmer 51, rotational speed of the abrasive pad 2, oscillation speed of the trimmer 51 on the abrasive pad 2, and trimming time.

[0056] Figure 4 This is a schematic diagram illustrating one embodiment of the estimation model 62. The estimation model 62 includes: an encoder 65 that extracts feature quantities from a target trajectory image; a decoder 67 that reconstructs the target trajectory image from the feature quantities; and a regression analyzer 69 that performs regression analysis on the extracted feature quantities and outputs trimming conditions. The encoder 65 has an input layer 65a that receives the target trajectory image as input. The decoder 67 has an output layer 67a that outputs an image substantially identical to the target trajectory image input to the encoder 65. The encoder 65 is configured to compress and extract feature quantities from the input target trajectory image. The decoder 67 is configured to reconstruct the target trajectory image from the feature quantities extracted by the encoder 65. This combination of the encoder 65 and decoder 67 constitutes an autoencoder.

[0057] The regression analyzer 69 is configured to perform regression analysis on the (compressed) features extracted by the encoder 65 and output trimming conditions that achieve the target trajectory, which is the trajectory of the abrasive grains on the target trajectory image input to the encoder 65. The trimming conditions output from the regression analyzer 69 include at least one of the following: the rotational speed of the trimmer 51, the rotational speed of the abrasive pad 2, the oscillation speed of the trimmer 51 on the abrasive pad 2, and the trimming time.

[0058] The presumption model 62 is a learning-complete model constructed through machine learning. More specifically, the presumption model 62 is a learning-complete model constructed through machine learning using training data that includes multiple trimming conditions and multiple trajectory images obtained through trimming simulations performed under multiple trimming conditions. The trimming simulation is a simulation of the trimming action performed by the trimming surface 51a of the trimmer 51 on the grinding pad 2.

[0059] The trajectory images included in the training data represent the trajectories of the abrasive grains constituting the trimming surface 51a on the trimming simulation. The aforementioned trimming conditions included in the training data are objective variables (i.e., positive solution labels), while the aforementioned trajectory images corresponding to the aforementioned trimming conditions are explanatory variables. The trimming simulation can also be performed by the trimming condition determination device 60, or by other simulation devices.

[0060] Figure 5 This is a schematic diagram illustrating an example of training data containing multiple trimming conditions and corresponding trajectory images. Trimming simulations are performed under each trimming condition to obtain trajectory images of abrasive particles corresponding to each trimming condition. Figure 5 The multiple trimming conditions A to F shown are all different, and the multiple trajectory images A to F obtained through trimming simulation are also all different. The multiple trimming conditions A to F contained in the training data correspond to the multiple trajectory images A to F contained in the training data in a one-to-one manner.

[0061] The condition determination device 60 constructs a prediction model 62 by performing machine learning using the aforementioned training data. Examples of machine learning include deep learning, and examples of the prediction model 62 include neural networks. However, the specific details of the machine learning and the prediction model 62 are not particularly limited as long as the desired purpose is achieved.

[0062] Figure 6 This is a schematic diagram illustrating the training phase of constructing an inference model 62 using machine learning with training data. When one of multiple trajectory images contained in the training data is input to the input layer 65a of the encoder 65, the trajectory image is output from the output layer 67a of the decoder 67, and the trimming conditions are output from the regression analyzer 69. The parameters (weighting values ​​or thresholds, etc.) of each node (neuron) of the encoder 65, decoder 67, and regression analyzer 69 are adjusted by machine learning in such a way that the trajectory image output from the decoder 67 matches the trajectory image input to the encoder 65, and the trimming conditions output from the regression analyzer 69 match the trimming conditions corresponding to the trajectory image input to the encoder 65. When the inference model 62 constructed by machine learning has a trajectory image that is not present in the input training data (e.g., the target trajectory image mentioned above), it can output trimming conditions for achieving the trajectory of the abrasive particles on the input trajectory image.

[0063] Figure 7This is a schematic diagram illustrating the practical stage of determining the trimming conditions for achieving the trajectory of abrasive particles on a target trajectory image using a prediction model 62 constructed through machine learning. A simulated pre-made target trajectory image is input to the input layer 65a of the encoder 65 of the prediction model 62. The regression analyzer 69 of the prediction model 62 outputs the trimming conditions for achieving the trajectory of abrasive particles on the target trajectory image (i.e., the target trajectory).

[0064] The trimming conditions output from the presumed model 62 are sent to Figure 1 The motion control unit 30 is shown. The motion control unit 30 controls the operation of the grinding apparatus, including the dressing mechanism 50, according to the dressing conditions, and the dressing mechanism 50 dresses the grinding surface 2a of the grinding pad 2. The dressing operation performed according to the dressing conditions determined by the dressing condition determination device 60 can be displayed on the screen. Figure 7 The trajectory of the abrasive particles in the target trajectory image shown is formed on the abrasive surface 2a of the abrasive pad 2.

[0065] The aforementioned dressing condition determining device 60 and dressing condition determining method can automatically determine the dressing conditions that can achieve the desired state of the grinding surface 2a using the estimation model 62.

[0066] The dressing of the polishing pad 2 is performed at several times. For example, before using a new polishing pad 2 for polishing, the dressing device 51 dresses the new polishing pad 2. This dressing action is also called break-in dressing. In other examples, the polishing pad 2 is dressed by the dressing device 51 after polishing a wafer and before polishing the next wafer. In yet another example, the polishing pad 2 is dressed by the dressing device 51 during wafer polishing.

[0067] The above embodiments are described with the aim of enabling those skilled in the art to practice the present invention. Those skilled in the art will naturally be able to devise various variations of the above embodiments. The technical concept of the present invention can also be applied to other embodiments. Therefore, the present invention is not limited to the described embodiments, but is interpreted within the broadest scope derived from the technical concept defined by the claims.

[0068] Industrial utilization

[0069] This invention enables the use of techniques for determining the dressing conditions of polishing pads used in polishing workpieces such as wafers, substrates, or panels.

[0070] Symbol Explanation

[0071] W: Chip

[0072] T: Trajectory of abrasive particles

[0073] 2: Grinding pad

[0074] 2a: Grinding surface

[0075] 5: Grinding table

[0076] 5a: Table spindle

[0077] 7: Grinding head

[0078] 8: Grinding fluid supply nozzle

[0079] 14: Support shaft

[0080] 16: Grinding head swing arm

[0081] 18: Grinding head shaft

[0082] 21: Rotary motors

[0083] 30: Motion Control Department

[0084] 30a: Storage device

[0085] 30b: Computing device

[0086] 40: Protrusion

[0087] 50: Repair and maintenance organization

[0088] 51: Dresser

[0089] 51a: Trimmed surface

[0090] 52: Dresser shaft

[0091] 55: Dresser swing arm

[0092] 60: Repair conditions determine the device

[0093] 60a: Storage device

[0094] 60b: Processor

[0095] 62: Presumed Model

[0096] 65: Encoder

[0097] 67: Decoder

[0098] 69: Regression Analyzer.

Claims

1. A dressing condition determining device, comprising determining dressing conditions for dressing a grinding pad by pressing the dressing surface of a rotating dresser against a rotating grinding pad, characterized in that, have: Storage device, which stores an inference model constructed through machine learning; and A processor that performs calculations according to the algorithm of the presumed model. The estimation model is configured such that, when a target trajectory image is input, it outputs trimming conditions for achieving the target trajectory, wherein the target trajectory image represents the target trajectory of multiple abrasive particles constituting the trimming surface on the abrasive pad.

2. The trimming condition determining device as described in claim 1, characterized in that, The dressing conditions include at least one of the following: the rotational speed of the dresser, the rotational speed of the abrasive pad, the oscillation speed of the dresser on the abrasive pad, and the dressing time.

3. The trimming condition determining device as described in claim 1, characterized in that, The target trajectory image is a simulated image of the trajectory of the plurality of abrasive grains generated by performing a simulation, which is a simulation of the action of pressing the dressing surface of the rotating dresser against the rotating abrasive pad to dress the abrasive pad.

4. The trimming condition determining device as described in claim 1, characterized in that, The inference model is a learned model constructed using multiple trimming conditions and training data through machine learning. The training data includes multiple trajectory images obtained through trimming simulations performed under the multiple trimming conditions.

5. The trimming condition determining device as described in claim 1, characterized in that, The presumption model has the following characteristics: An encoder that extracts feature quantities from the target trajectory image; A decoder that reconstructs the target trajectory image from the said feature quantities; and A regression analyzer performs regression analysis on the extracted features and outputs the trimming conditions.

6. The trimming condition determining device as described in claim 1, characterized in that, The target trajectory is the trajectory of the abrasive particles evenly distributed on the abrasive pad, and the intersection points of the trajectories of the abrasive particles are evenly distributed on the trajectory of the abrasive pad.

7. A method for determining dressing conditions, wherein the dressing conditions are used to dress the polishing pad by pressing the dressing surface of a rotating dresser against a rotating polishing pad, characterized in that, The target trajectory image is input into the estimation model constructed through machine learning. The target trajectory image represents the target trajectory of the multiple abrasive particles constituting the trimmed surface on the abrasive pad. The model outputs adjustment conditions for achieving the target trajectory.

8. The method for determining trimming conditions as described in claim 7, characterized in that, The dressing conditions include at least one of the following: the rotational speed of the dresser, the rotational speed of the abrasive pad, the oscillation speed of the dresser on the abrasive pad, and the dressing time.

9. The method for determining trimming conditions as described in claim 7, characterized in that, The target trajectory image is a simulated image of the trajectory of the plurality of abrasive grains generated by performing a simulation, which is a simulation of the action of pressing the dressing surface of the rotating dresser against the rotating abrasive pad to dress the abrasive pad.

10. The method for determining trimming conditions as described in claim 7, characterized in that, The inference model is a learned model constructed using multiple trimming conditions and training data through machine learning. The training data includes multiple trajectory images obtained through trimming simulations performed under the multiple trimming conditions.

11. The method for determining trimming conditions as described in claim 7, characterized in that, The presumption model has the following characteristics: An encoder that extracts feature quantities from the target trajectory image; A decoder that reconstructs the target trajectory image from the said feature quantities; and A regression analyzer performs regression analysis on the extracted features and outputs the trimming conditions.

12. The method for determining trimming conditions as described in claim 7, characterized in that, The target trajectory is the trajectory of the abrasive particles evenly distributed on the abrasive pad, and the intersection points of the trajectories of the abrasive particles are evenly distributed on the trajectory of the abrasive pad.

Images