Machine learning for classifying retaining rings

Abstract

This application discloses machine learning for classifying retaining rings. A method for optimizing polishing includes, for each of a plurality of retaining rings mounted on a particular bearing head, performing measurements on the bottom surface of the corresponding retaining ring mounted on the particular bearing head using a coordinate measuring machine and collecting a corresponding removal profile of a substrate polished using the corresponding retaining ring. A machine learning model is trained based on the measurements of the bottom surface of the retaining ring and the corresponding removal profile.

Description

Technical Field

[0001] This disclosure generally relates to chemical mechanical polishing, and more specifically, to classifying retaining rings using machine learning. Background Technology

[0002] Integrated circuits are typically formed on a substrate (e.g., a semiconductor wafer) on a silicon wafer by sequentially depositing conductive layers, semiconductive layers, or insulating layers and by subsequent processing of these layers.

[0003] One manufacturing step involves depositing a filler layer on a non-planar surface and planarizing that filler layer. For some applications, the filler layer is planarized until the top surface of the patterned layer is exposed or the desired thickness is retained above the underlying layer. Additionally, planarization can be used to planarize, for example, the surface of a substrate with a dielectric layer for photolithography.

[0004] Chemical mechanical polishing (CMP) is a recognized planarization method. This method typically requires a substrate to be mounted on a carrier head. The exposed surface of the substrate is placed against a rotating polishing pad. The carrier head provides a controlled load on the substrate, thereby pushing it against the polishing pad. In some polishing machines, the carrier head includes a membrane forming multiple independently pressurized, radially concentric chambers, where the pressure in each chamber controls the polishing rate in each corresponding region of the substrate. A polishing fluid, such as a slurry containing abrasive particles, is supplied to the surface of the polishing pad.

[0005] As a separate problem, machine learning is widely used to assist automation through experience (e.g., historical data), such as classifying projects into different categories. Machine learning can be divided into three main categories: supervised learning using input data with known class labels, unsupervised learning using input data with unknown class labels, and reinforcement learning aimed at achieving specific goals through navigation and interaction with the environment. As some examples, unsupervised learning has been used in marketing (e.g., customer segmentation), biology (e.g., clustering DNA patterns), and finance (e.g., anomaly detection or fraud detection).

[0006] One branch of machine learning is deep learning, which typically employs neural networks. A neural network is a machine learning model that uses one or more non-linear units to predict its output from a given input. Some neural networks include one or more hidden layers in addition to the output layer. The output of each hidden layer is used as the input to the next layer in the network (i.e., the next hidden layer or output layer). Each layer of the network generates its output from the given input based on the current values ​​of its respective set of parameters. Summary of the Invention

[0007] A method for evaluating polishing includes: for each of a plurality of retaining rings mounted on a particular carrier head, measuring the bottom surface of the corresponding retaining ring on the particular carrier head using a coordinate measuring machine. The measurements represent features of the bottom surface. An unsupervised learning algorithm is executed to classify each of the plurality of retaining rings into a corresponding category based on the measurements of the plurality of retaining rings; the classifications generated by the unsupervised learning algorithm are stored; and the classifications are evaluated based on polishing profile measurements to determine parameters driving profile differences.

[0008] On the other hand, a method for optimizing polishing includes, for each of a plurality of retaining rings mounted on a particular carrier head, performing measurements on the bottom surface of the corresponding retaining ring mounted on the particular carrier head using a coordinate measuring machine and collecting a corresponding removal profile of the substrate polished using the corresponding retaining ring. A machine learning model is trained based on the measurements of the bottom surface of the retaining ring and the corresponding removal profile.

[0009] Some implementations may include, but are not limited to, one or more of the following possible advantages.

[0010] Measurements can be performed at high resolution across the bottom surface of the retaining ring, for example, by performing a mesh scan across the entire bottom surface of the retaining ring. This generates sufficient geometric information to enable the analysis of retaining ring features. The features analyzed for a specific retaining ring can be correlated with one or more polished profiles of a substrate polished using the retaining ring, thereby allowing the determination of one or more retaining ring-related parameters that drive differences in the polished profiles.

[0011] Furthermore, polishing profiles are predicted based on measurements collected by mesh scanning. Specifically, unsupervised machine learning algorithms can be used to classify one or more retaining rings into corresponding categories among multiple classes. Each classified retaining ring can be associated with one or more polishing profiles of a corresponding substrate polished using that classified retaining ring.

[0012] A neural network model can be trained by receiving geometric information of a classified retaining ring with associated polished profiles as input. The retaining ring then uses the neural network model to predict its polished profile by performing inference operations within the trained neural network, which receives the measured bottom surface profile of the retaining ring as input.

[0013] Production time can be improved. Using the described technology, the system can more effectively detect whether a retaining ring can be used to polish the substrate according to different polishing requirements before or after the break-in process. The system can also effectively determine whether the retaining ring has been adequately broken in and provide user information or guidance for further break-in processes based on the classification results.

[0014] Quality control during the polishing process can be improved, and throughput can be increased. Specifically, defective retaining rings can be identified and replaced. The polishing rate in one or more areas of the substrate being polished can be adjusted based on a predicted polishing profile associated with the retaining rings, thereby enhancing uniformity within the wafer during polishing and ultimately achieving higher throughput.

[0015] Furthermore, the described technique can be easily scaled with low computational cost. The system can be further adjusted or modified based on subsequently measured geometric data and polishing profiles, thereby eliminating unnecessary redundant calculations and reducing computational costs. The stored classification data and neural network model can be accessed and used simultaneously by one or more polishing devices located at different locations without recalculation, allowing for easy scaling of the system to multiple polishing devices.

[0016] The following figures and description illustrate details of one or more embodiments of the present invention. Other features, objects, and advantages will become apparent from the specification, figures, and claims. Attached Figure Description

[0017] Figure 1 The illustration shows a schematic cross-sectional view of an example of a polishing apparatus including a retaining ring.

[0018] Figure 2A The illustration shows a schematic perspective view and a cross-sectional view of an example retaining ring.

[0019] Figure 2B The illustration shows a schematic cross-sectional view of the upper part of the retaining ring.

[0020] Figure 2C The illustration shows a schematic bottom view of an example retaining ring.

[0021] Figure 3 The illustration shows a schematic top view of an example retaining ring measured using a coordinate measuring machine.

[0022] Figure 4A and Figure 4B A schematic top view is shown of different types of grid scans used for measurement.

[0023] Figure 5 This is a flowchart illustrating an example classification process for a hold-on loop using a machine learning algorithm.

[0024] Figure 6 This is a flowchart illustrating an example process of training a neural network for predicting polished profiles based on input data.

[0025] The same reference numerals and names in the various figures indicate the same elements. Detailed Implementation

[0026] Ideally, the polishing rate of a polished substrate is substantially uniform across its entire surface. However, in practice, different radial and / or angular regions of the substrate may exhibit different polishing rates. Additionally, the substrate to be polished may have initial radial and / or angular non-uniformity.

[0027] One source of polishing inhomogeneity is variation in the profile of the bottom surface of the retaining rings. That is, even minute differences in the bottom surface profiles of the two retaining rings can lead to different polishing rates in the edge regions. Polishing performance can be better understood by correlating the bottom surface profiles of the retaining rings with the polishing profile of the substrate. Furthermore, the polishing process can be tuned, for example, by varying the pressure applied to the substrate, to improve the uniformity of the polishing rate in the edge regions.

[0028] Conventionally, when inspecting the bottom surface profile, only a few measurement points are made across the entire bottom surface of the retaining ring. Therefore, existing techniques may not provide sufficient information to analyze the characteristics of the retaining ring's bottom surface. While techniques such as "ring break-in" were applied to retaining rings before their use in integrated circuit manufacturing, increasingly stringent precision requirements and the added downtime associated with "ring break-in" have increased production costs.

[0029] However, the techniques described herein can increase throughput, reduce production costs, and improve polishing quality (e.g., reduce non-uniformity near substrate edges).

[0030] A coordinate measuring machine (CMM) can be used to measure the bottom surface of the retaining rings with high definition, generating a high-quality bottom surface profile for further analysis. Unsupervised machine learning algorithms can be used to classify each of the multiple retaining rings into its corresponding category, and the classification information can be used to obtain parameters related to differences in the polished profile.

[0031] A neural network model can be trained to take the input as a classification of each retaining ring and use the corresponding classified retaining rings to polish the contour of the substrate. After training, the neural network can infer the trained neural network from the geometric information of the input retaining rings, thereby generating a prediction of the polished contour (e.g., edge polishing contour) of the substrate polished using retaining rings.

[0032] Therefore, the system employing this technology can detect defects in the retaining ring based on the measured bottom surface profile of the retaining ring, adjust the polishing process when using a retaining ring based on the obtained parameters, and predict the polishing profile of the substrate being polished using the retaining ring. This system can improve the uniformity of the polished wafer and increase polishing throughput.

[0033] Figure 1An example of a polishing apparatus 100 is illustrated. The polishing apparatus 100 includes a rotatable disc-shaped worktable 120 on which a polishing pad 110 is located.

[0034] The worktable 120 is operable to rotate about axis 125. For example, motor 121 can rotate drive shaft 124 to rotate worktable 120. Polishing pad 110 can be detachably attached to worktable 120, for example, by an adhesive layer. Polishing pad 110 can be a double-layer polishing pad having an outer polishing layer 112 and a softer backing layer 114.

[0035] The polishing apparatus 100 may include a dispensing port 130 for delivering polishing fluid 132 (such as abrasive slurry) onto the polishing pad 110.

[0036] The polishing apparatus may also include a polishing pad adjuster for grinding the polishing pad 110 to maintain the polishing pad 110 in a consistent grinding state.

[0037] The polishing apparatus 110 may include a support head 140 operable to hold the substrate 10 against the polishing pad 110. The support head 140 may be configured to independently control polishing parameters, such as pressure, for each of a plurality of regions on the substrate 10.

[0038] The bearing head 140 is suspended from a support structure 150 (e.g., a turntable) and connected to a bearing head rotary motor 154 via a drive shaft 152, enabling the bearing head to rotate about axis 155. Optionally, the bearing head 140 may oscillate laterally, for example, on a slider on the turntable 150 or by the rotational oscillation of the turntable itself. In operation, the worktable rotates about its central axis 125. Each bearing head rotates about its central axis 155 and translates laterally on the top surface of the polishing pad.

[0039] The carrier head 140 may include: a housing 144 that can be connected to a drive shaft 152; a support plate 184 that extends over a flexible central diaphragm 182; an annular pressure control assembly 195 that surrounds the flexible central diaphragm 182; and a retaining ring 142 that surrounds the annular pressure control assembly 195 to hold the substrate 10 below the flexible central diaphragm 182.

[0040] The lower surface of the flexible central diaphragm 182 provides a mounting surface for the substrate 10. The flexible central diaphragm 182 may include one or more flaps fixed to a support plate 184 for forming one or more pressurizable chambers. These chambers are connected to one or more pressure supplies 181 via corresponding pressure supply lines 183 to apply different pressures to the inner regions of the substrate (e.g., regions at least 6 mm away from the substrate edge) during polishing, allowing the system to adjust the corresponding polishing rate in the corresponding regions of the substrate.

[0041] Figure 2A The illustration shows a schematic perspective cross-sectional view of the example retaining ring 142.

[0042] As described above, the retaining ring 142 is a generally annular ring that can be fixed to the bearing head 140 of the polishing device 100.

[0043] like Figure 2A As shown, the upper portion 205 of the retaining ring 100 has a cylindrical inner surface 265, a cylindrical outer surface 250, and a top surface 215 substantially perpendicular to both the inner and outer surfaces. The top surface includes holes 220 to receive mechanical fasteners, such as bolts, screws, or other hardware (such as screw sleeves or inserts), for securing the retaining ring 142 and the bearing head 140 together (not shown). Additionally, one or more alignment holes 225 may be located in the top surface 215 of the upper portion 205 to properly align the retaining ring 142 and the bearing head 140.

[0044] The upper part 205 can be formed of a rigid or high tensile modulus material, such as metal, ceramic, or hard plastic. Suitable metals for forming the upper part include stainless steel, molybdenum, titanium, or aluminum. Alternatively, composite materials, such as composite ceramics, can be used.

[0045] The second portion of retaining ring 142, namely the lower portion 210, can be formed of a material chemically inert to the CMP process and can be softer than the material of the upper portion 205. The material of the lower portion 210 should have sufficient compressibility or elasticity to ensure that contact between the substrate edge and retaining ring 142 does not cause chipping or cracking of the substrate. The lower portion 210 should also be durable and highly abrasion-resistant, although wear on the lower portion 210 is acceptable. For example, the lower portion 130 can be made of plastics such as polyphenylene sulfide (PPS), polyethylene terephthalate (PET), polyetheretherketone (PEEK), carbon-filled PEEK, polyetheretherketone (PEKK), polybutylene terephthalate (PBT), polytetrafluoroethylene (PTFE), polybenzimidazole (PBI), polyetherimide (PEI), or composite materials.

[0046] The lower portion may also have a cylindrical inner surface 235, a cylindrical outer surface 230, and a bottom surface 255. Although the bottom surface 255 of the retaining ring 142 may begin as a plane, after break-in or use, the bottom surface 255 of the lower portion typically has a non-planar profile. In some implementations, the radial profile of the bottom surface 255 may include a curved section, a truncated cone section, or a planar section. It would be advantageous for the radial profile of the bottom surface 255 of the retaining ring 100 to substantially match a reference profile to achieve uniformity between wafers, but this may not be possible due to variations during manufacturing or break-in processes, or different wear modes during polishing.

[0047] Figure 2BThe illustration shows a schematic cross-sectional view of the retaining ring 142, with the deformation of the bottom surface 255 magnified for easy observation. In practice, each retaining ring may have a bottom surface profile that differs slightly from the reference profile, which could result in unwanted inhomogeneities or inter-wafer variations in the edge regions of the substrate being polished.

[0048] Figure 2C The illustration shows a schematic bottom view of an example retaining ring 142. As described above, the characteristics of the bottom surface profile of the retaining ring may include the height distribution of the bottom surface 255 (or equivalently, the thickness distribution of the lower portion 210) and the flatness of the bottom surface 255 (i.e., the height or thickness variation of the bottom surface 255), as well as the roundness of the retaining ring.

[0049] like Figure 2C As shown, the retaining ring has an inner diameter of 260 and an outer diameter of 280. If the ring were perfectly circular, the width of the retaining ring could be determined by measuring the difference between the inner and outer diameters. However, in practice, the retaining ring 142 may have different curvatures at different angular locations on its inner and outer boundaries, resulting in imperfect roundness. For example, the curvatures of the outer boundaries at locations 290A, 290B, 290C, and 290D may differ from each other. As another example, the curvatures of the inner boundaries at locations 291A and 291B may be substantially the same.

[0050] Figure 3 The illustration shows a schematic top view of an example retaining ring 142 measured using a coordinate measuring machine 300. The retaining ring 142 is positioned with its bottom surface facing upwards.

[0051] To perform measurements on the bottom surface of the retaining ring, the retaining ring is mounted on a bearing head, and a coordinate measuring machine (CMM) 300 measures the bottom surface of the retaining ring. The system can measure the bottom of the retaining ring.

[0052] The CMM 300 may include a sensor 310 configured to measure the vertical position of each of a plurality of points on the bottom surface of the retaining ring. The sensor 310 can be any suitable sensor, such as a laser sensor or a contact probe. In some implementations, the sensor 310 may perform a measurement of the height or thickness coordinate (Z-axis coordinate) at each measurement point on the bottom surface of the retaining ring.

[0053] CMM 300 may include actuators 350 and 355, and a sensor 310 may be mounted on and movable around actuators 350 and 355 to detect height information of the bottom surface of the retaining ring. For example, sensor 310 is mounted on actuator 355. Each of the two actuators may move in unrelated (e.g., non-parallel) directions, such that sensor 310 can move to cover the entire XY plane. For example, actuator 355 may move in the X direction along track 360, and actuator 350 may move in the Y direction perpendicular to the X direction along another track 370.

[0054] The CMM 300 may also include a controller 390, which has control lines 380a and 380b to control actuators 350 and 355 respectively, and a data or control line 380c to control the operation of the sensor and receive the captured height information (Z coordinate) of the measurement area.

[0055] Figure 4A and Figure 4B This is a schematic diagram of different types of meshes. Measurements can be distributed in the radial mesh 410 (e.g., as shown in the diagram). Figure 4A As shown), or distributed in a rectangular grid 420 (e.g., as shown). Figure 4B (As shown). Based on the corresponding measurement density, each grid can be of any suitable size. The CMM 300 can measure areas from millimeters to centimeters. For example, Figure 4A Each radial grid in the CMM 300 can have an edge of approximately 1 mm. The CMM 300 can generate 1000 to 10000 measurements on the bottom surface of the retaining ring. For example, Figure 4B The total number of grids is approximately 3000.

[0056] Figure 5 This is a flowchart illustrating an example classification process 500 for maintaining loops, for example, using a machine learning algorithm.

[0057] The CMM 300 measures the bottom surface of each retaining ring. Each measurement represents the vertical height or thickness (502) at the measurement location. As described above, each retaining ring is mounted to its corresponding bearing head before measurement. In some implementations, the system measures one or more previously run-in retaining rings.

[0058] The controller 390 can receive measurement values ​​and convert them into bottom surface characteristics at the measurement location, such as the overall flatness, ring taper, roundness, and average inner or outer diameter of each white surface.

[0059] In some implementations, controller 390 can determine roundness by generating data representing the degree of symmetry of the bottom surface of the retaining ring relative to the axis of rotation. For example, the degree of symmetry may include data representing the curvature distribution on the boundaries of the retaining ring. Therefore, the system can measure and determine the roundness of the retaining ring by determining an asymmetrical curvature distribution based on the measured curvature data.

[0060] To analyze the flatness, roundness, and taper of the retaining rings, the system first generalizes the spatial mapping of the retaining rings based on measurements in a Cartesian coordinate system. Then, the system employs appropriate numerical techniques to process the spatial mapping to generate a consistent mapping across different measurements of different retaining rings. More specifically, each measurement includes its own reference plane and its own reference center point, and each measurement represents the height information of each measurement point relative to its respective reference plane and reference center point. The system can use different techniques to normalize each measurement. For example, the system can use numerical techniques such as "best-fit plane" or "minimum closed circle" to generate a shared center for each measurement to normalize the spatially mapped measurement data. As another example, the system can use one or more data filters to screen peripheral data to improve the integrity of each measurement.

[0061] Following the above data processing, the system then stores processed data representing one or more features of the bottom surface profile of the retaining ring for analysis.

[0062] The system then executes an unsupervised learning algorithm to classify each of the multiple retaining rings into a corresponding category based on measurements of the multiple retaining rings (504). To classify each of the multiple retaining rings, the system first generates one or more features from the stored data in step 502. The one or more features can be of any suitable type to represent one or more features of the measured data representing the bottom surface profile. For example, features representing the measured data of the bottom surface could be the overall average thickness; thickness variation in angular direction, radial direction, or both (i.e., smoothness); roundness of the inner boundary, outer boundary, or both; or degree of symmetry. Each of the feature types has its own feature map for analysis.

[0063] The system can use any suitable unsupervised learning algorithm for classification. In one preferred embodiment, the system employs the K-means method. More specifically, the system can assume several classes (e.g., a scalar value K in K-means) for the measured hold loops, assume a cluster center for each class in a specific feature map, and assign each hold loop to the nearest class (i.e., the hold loop is assigned to a class whose cluster center is closest to the feature type of the hold loop in a specific feature map). The system can generalize the global error (e.g., squared Euclidean distance) by assigning each hold loop to the corresponding class and minimizing the global error by adjusting the assignments. The system can also update the corresponding cluster centers of the classes based on the corresponding features of the hold loops assigned to the corresponding classes at each assignment adjustment.

[0064] The system can determine the number of classes. To do this, the system can first select multiple candidate numbers of classes for the K-means algorithm and obtain the corresponding candidate errors (i.e., the corresponding minimum global errors) for assigning the hold-ring to one of the multiple candidate numbers of classes (i.e., different K values). For example, the system can perform classification on 2 classes (i.e., K=2), 3 classes (i.e., K=3), and 10 classes (i.e., K=10) and obtain the corresponding minimum global errors. The system can select one number of classes from the multiple candidate numbers of classes as the number of classes for the K-means algorithm to classify the hold-ring. For example, the system can set K to the number of classes with the minimum candidate error. As another example, the system can set K to the number of classes with the second minimum candidate error at the cost of minimal computation time.

[0065] The system stores the classifications generated by the unsupervised learning algorithm (506). For example, the system stores the number of categories K determined, the cluster centers of each category in the corresponding feature map, and the classification label of each hold-up loop. In some implementations, the system may store the category with the most hold-up loops as the baseline category.

[0066] Then, the classification can be evaluated based on the polished profile measurements to determine the parameters driving the profile differences (508). For example, the user can enable the system to plot various performance metrics based on various ring features. For example, edge uniformity can be plotted as a function of the inner edge roundness shape to determine whether and how the inner edge shape affects the polished profile.

[0067] The system can be used to adjust retaining rings that are not classified into one or more preset categories. For example, the system can store data indicating that several categories provide acceptable results in polishing. Sample rings can then be measured using the CMM system and followed by a classification algorithm. If a sample ring does not belong to a specified category, corrective measures can be taken. For example, the retaining ring can be further "run-in" for a period of time, and the "run-in" bottom surface can be measured. This process can be repeated until the retaining ring is classified into an acceptable category.

[0068] This system can predict the post-polished profile of a substrate using classified retaining rings. For prediction, the system can collect multiple polishing profiles using retaining rings from a specific category and generate an average polishing profile as the predicted post-polished profile for polishing the substrate using retaining rings from that category.

[0069] Figure 6 This is a flowchart illustrating an example process 600 for training a neural network to predict polishing profiles based on input data. Process 600 may be executed by one or more computers located at one or more locations. Alternatively, process 600 may be stored as instructions in one or more computers. Once executed, the instructions may cause one or more components of the polishing apparatus, one or more components of the CMM, or one or more computers to perform the process. For example, at least some steps of the process are performed by, for example... Figure 1 The controller 190 shown is executed.

[0070] Similarly, such as Figure 5 As described in step 502, a coordinate measuring machine is used to measure the bottom surface of the corresponding retaining ring. The measured value represents a characteristic of the bottom surface (602). The characteristic of the bottom surface can be a surface height or a surface thickness. More specifically, the system can generate a specific mapping for each retaining ring and store the measured data in memory. In some implementations, the system can perform measurements on retaining rings that have been previously "broken in".

[0071] The system then collects the corresponding removal profiles (604) of the substrate polished with the corresponding retaining ring, and trains a machine learning model (606) based on the measurements of the bottom surface of the retaining ring and the corresponding removal profiles.

[0072] The machine learning model includes a convolutional neural network model that can be trained using training examples. The training examples include training inputs (such as the bottom surface profile of each retaining ring), polishing profiles of the substrate using the corresponding retaining rings, and labels for training each polishing profile.

[0073] In some implementations, the system can label each polishing profile. For example, the system can assign a first label (e.g., "fast edge") to multiple polishing profiles, each having a fast edge removal rate compared to a reference profile, and assign the label "slow edge" to multiple polishing profiles with a slow edge removal rate.

[0074] This system trains a neural network model by minimizing the global misclassification error based on training examples. During training, the system updates the weights of each layer of the neural network through backpropagation to minimize the global error.

[0075] After training the neural network, the system can use the trained neural network to predict the removal profile of the substrate. More specifically, the system can measure the bottom surface profile of the retaining ring, or feed stored data representing the measured bottom surface profile to the trained neural network, and perform inference operations using the trained neural network and trained weights to generate a prediction of the polishing profile using the retaining ring. Alternatively or additionally, the system can use the trained neural network to predict the label of the predicted polishing profile.

[0076] This system can continue training the neural network using input measurements, allowing the network's weights to be updated based on newer measurement data. The trained neural network with updated weights can be stored in the memory of one or more computers located in one or more locations. The trained neural network can be accessed by one or more computers or computing units to accelerate inference operations.

[0077] As used in this specification, the term substrate may include, for example, product substrates (e.g., comprising multiple memory or processor dies), test substrates, bare substrates, and gate substrates. Substrates can be at various stages of integrated circuit manufacturing; for example, a substrate may be a bare wafer, or a substrate may comprise one or more deposited and / or patterned layers. The term substrate may include disks and rectangular wafers.

[0078] The polishing apparatus and methods described above can be applied in various polishing systems. The polishing pad, or the support head, or both, can be movable to provide relative movement between the polishing surface and the substrate. For example, the stage can run around a track rather than rotate. The polishing pad can be a circular (or some other shaped) pad fixed to the stage. Some aspects of the endpoint detection system can be applied to linear polishing systems, for example, where the polishing pad is a linearly moving, continuous, or roll-to-roll strip. The polishing layer can be a standard (e.g., polyurethane with or without filler) polishing material, a soft material, or a fixed abrasive. Using the term "relative positioning," it should be understood that the polishing surface and the substrate can be held in a vertical orientation or some other orientation.

[0079] Control of the various systems and processes, or portions thereof, described in this specification can be implemented as a computer program product comprising instructions stored in one or more non-transitory computer-readable storage media, and such instructions being executable on one or more processing devices. The systems or portions thereof described in this specification can be implemented as apparatus, methods, or electronic systems, which may include one or more processing devices and a memory for storing executable instructions for performing the operations described in this specification.

[0080] The embodiments of classification and machine learning model training described in this specification can be implemented in digital electronic circuits, tangibly implemented computer software or firmware, computer hardware, including the structures disclosed in this specification and their structural equivalents, or combinations thereof. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more computer program instructions encoded on a tangible, non-transient storage medium for execution by a data processing device or for controlling the operation of a data processing device. The computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access storage device, or a combination thereof. Optionally or additionally, the program instructions can be encoded on artificially generated propagation signals (e.g., machine-generated electrical, optical, or electromagnetic signals) that are generated to encode information for transmission to a suitable receiver device for execution by the data processing device.

[0081] A computer program (also referred to or described as a program, software, software application, application, module, software module, script, or code) can be written in any form of programming language, including compiled or interpreted languages ​​or declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A program may, but is not required to, correspond to a file in a file system. A program may be stored as a portion of a file containing other programs or data, for example, as one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinating files (e.g., a file storing portions of one or more modules, subroutines, or code). A computer program may be deployed to execute on a single computer located at one location, or on multiple computers distributed across multiple locations and interconnected by data communication networks.

[0082] The processes and logic flows described in this specification can be executed by one or more programmable computers, which execute one or more computer programs to perform functions by manipulating input data and generating outputs. The processes and logic flows can also be executed by special-purpose logic circuitry (e.g., FPGA or ASIC), or by a combination of special-purpose logic circuitry and one or more programmable computers.

[0083] A computer suitable for executing computer programs can be based on a general-purpose or special-purpose microprocessor, or both, or any other type of central processing unit. Generally, the central processing unit receives instructions and data from read-only memory, random access memory, or both. The basic components of a computer are the central processing unit for running or executing instructions, and one or more memory devices for storing instructions and data. The central processing unit and memory may be supplemented by or incorporated into a dedicated logic circuit system. Generally, a computer will also include one or more mass storage devices (e.g., magnetic disks, magneto-optical disks, or optical disks) for storing data, or operatively coupled to receive data from or transfer data to one or more mass storage devices, or both. However, a computer need not have such devices.

[0084] Computer-readable media suitable for storing computer program instructions and data include various forms of non-volatile memory, media, and memory devices, such as: semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices); magnetic disks (e.g., internal hard disks or removable disks); magneto-optical disks; and CD ROMs and DVD-ROMs.

[0085] The data processing apparatus used to implement machine learning models may also include, for example, a dedicated hardware accelerator unit for processing the common and computationally intensive parts of machine learning training or production, i.e., inference, workloads.

[0086] Machine learning models can be implemented and deployed using machine learning frameworks, such as the TensorFlow framework, the Microsoft Cognitive Toolkit framework, the Apache Singa framework, or the Apache MXNet framework.

[0087] Embodiments of the subject matter described in this specification can be implemented in a computing system that includes back-end components (e.g., as a data server), middleware components (e.g., an application server), or front-end components (e.g., a client computer with a graphical user interface, a web browser, or an application through which a user can interact with an implementation of the subject matter described in this specification), or any combination of one or more such back-end, middleware, or front-end components. The components of the system can be interconnected by digital data communication (e.g., a communication network) of any form or medium. Examples of communication networks include local area networks (LANs) and wide area networks (WANs), such as the Internet.

[0088] The computing system may include clients and servers. Clients and servers are typically geographically separated and usually interact via a communication network. The client-server relationship is determined by computer programs running on their respective computers that have a client-server relationship with each other. In some embodiments, the server sends data (e.g., HTML pages) to a user device acting as a client, for example, to display data to a user interacting with that device and to receive user input from that user. Data generated on the user device (e.g., the result of user interaction) can be received from that device on the server.

[0089] Although this specification contains numerous specific implementation details, these details should not be construed as limiting the scope of any invention or the scope of what may be claimed, but rather as descriptions of features specific to particular embodiments of a particular invention. Certain features described herein, in the context of individual embodiments, may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may be implemented individually in multiple embodiments or in any suitable sub-combination. Furthermore, while features may be described above as functioning in certain combinations, and even initially claimed in this way, in some cases one or more features from a claimed combination may be removed from that combination, and the claimed combination may involve sub-combinations or variations thereof.

[0090] Specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims.

[0091] Other embodiments are within the scope of the appended claims.

Claims

1. A method for evaluating polishing, the method comprising: For each of a plurality of retaining rings mounted on a particular carrier head, a coordinate measuring machine is used to measure the vertical position of each of a plurality of points on the bottom surface of the respective retaining ring mounted on the particular carrier head to generate a surface profile of the bottom surface on the retaining ring, wherein the measured values ​​represent the height of the bottom surface of the retaining ring at the plurality of points; An unsupervised learning algorithm is executed to classify each of the plurality of holding loops into a corresponding category based on measurements of the plurality of holding loops; Store the classifications generated by the unsupervised learning algorithm; as well as The classification is evaluated based on polishing profile measurements to determine the parameters driving profile differences.

2. The method of claim 1, wherein performing a measurement on the bottom surface of the respective retaining ring comprises: The roundness of the corresponding retaining ring is measured by determining the asymmetric curvature distribution.

3. The method of claim 1, wherein performing a measurement on the bottom surface of the corresponding retaining ring comprises: Multiple measurements were performed on the bottom surface at different widths and angles.

4. The method of claim 3, wherein performing multiple measurements on the bottom surface comprises: The bottom surface of the retaining ring is divided into multiple regions of the retaining ring; as well as For each of the plurality of regions, the in-plane location of that region and the average thickness of that region are measured.

5. The method of claim 1, wherein the bottom surface is characterized by a surface height or annular layer thickness.

6. The method of claim 1, wherein the unsupervised learning algorithm includes the K-means algorithm.

7. The method of claim 6, further comprising using the K-means algorithm to determine the number of categories in the classification, including: For each of the multiple possible number of categories: Select the number of candidates to represent the total number of categories; as well as Determine the corresponding candidate error for classifying the plurality of holding loops into a number of candidate categories using the K-means algorithm; and One of the candidate numbers is selected as the category number based on the minimum candidate error.

8. The method of claim 1, further comprising: Determine the category of retaining rings, which has the most retaining rings among all categories; Set the determined category as the baseline category; as well as Adjust the holding rings that are not classified into the baseline category according to the classification.

9. The method of claim 1, further comprising: The post-polished profile of a substrate polished using the classified retaining rings is predicted.

10. The method of claim 1, wherein the plurality of retaining rings comprises one or more retaining rings that have been run-in.

11. A method for optimizing polishing, the method comprising: For each of the multiple retaining rings mounted on a specific bearing head A coordinate measuring machine is used to measure the vertical position of each of a plurality of points on the bottom surface of a corresponding retaining ring mounted on the particular bearing head to generate a surface profile of the bottom surface on the retaining ring, wherein the measured values ​​represent the height of the bottom surface of the retaining ring at the plurality of points; Collect the corresponding removal profile of the substrate polished using the corresponding retaining ring; and The machine learning model is trained based on measurements of the bottom surface of the retaining ring and the corresponding removed profile.

12. The method of claim 11, comprising: Measure the bottom surface of a specific retaining ring; and The measurements are fed into a trained machine learning model to generate predicted removed contours.

13. The method of claim 11, wherein the bottom surface is characterized by a surface height or annular layer thickness.

14. The method of claim 11, wherein collecting the respective removed profile of the substrate polished using the respective retaining ring comprises: Determine the corresponding removed contour labels for training.

15. The method of claim 14, wherein the corresponding removal contour label includes a first label indicating a fast removal rate in the edge region of the substrate, and a second label indicating a slow removal rate in the edge region of the substrate.

16. The method of claim 11, wherein the machine learning model comprises a convolutional neural network.

17. A non-transient computer-readable medium encoded using a computer program, said computer program comprising instructions to cause one or more computers to perform the following operations: Before polishing the substrate, multiple measurements of the vertical position of each of a plurality of points on the bottom surface of a corresponding retaining ring mounted on a specific carrier head are received using a coordinate measuring machine to generate a surface profile of the bottom surface on the retaining ring, wherein the measurements represent the height of the bottom surface of the retaining ring at the plurality of points. The plurality of measurements are fed to a trained machine learning model to generate a predicted removed profile of the substrate. Adjusted polishing parameters are generated in response to the predicted removal profile to improve the polishing uniformity of the substrate; as well as This enables the polishing system to polish the substrate using the adjusted polishing parameters.

18. The computer-readable medium of claim 17, wherein the polishing parameters include pressure on the substrate.

19. The computer-readable medium of claim 17, wherein the machine learning model comprises a neural network.

20. The computer-readable medium of claim 17, wherein the profile is a surface height profile or a layer thickness distribution.

Images