Real-time in-SITU substrate orientation detection

Abstract

A method of determining an orientation of a substrate during polishing includes generating a virtual model of an interface between a polishing pad and a substrate based on rotating the polishing pad about a surface of the substrate, wherein the polishing pad comprises a sensor configured to scan the surface of the substrate. The method further includes using the sensor to scan the surface of the substrate and produce a signal. The method further includes comparing the signal to one or more signals associated with the virtual model to determine a rotational orientation of the substrate.

Description

PATENTAttorney Docket No.: 44024954WO01REAL-TIME IN-SITU SUBSTRATE ORIENTATION DETECTIONBACKGROUNDField

[0001] The present disclosure relates to chemical mechanical polishing (CMP), and more specifically to detecting an orientation of a substrate during the CMP process.Description of the Related Art

[0002] An integrated circuit is typically formed on a substrate by the sequential deposition of conductive, semiconductive, and / or insulative layers on a semiconductor substrate. A variety of fabrication processes require planarization of a layer on the substrate. For example, one fabrication step involves depositing a filler layer over a non-planar surface and planarizing the filler layer. For certain applications, the filler layer is planarized until the top surface of a patterned layer is exposed. For example, a metal layer can be deposited on a patterned insulative layer to fill trenches and holes in the insulative layer. After planarization, the remaining portions of the metal in the trenches and holes of the patterned layer form vias, plugs, and lines to provide conductive paths between integrated circuits (ICs) on the substrate. As another example, a dielectric layer can be deposited over a patterned conductive layer, and then planarized to enable subsequent photolithographic steps.

[0003] Chemical mechanical polishing (CMP) is one accepted method of planarization. This planarization method typically requires that the substrate be mounted on a carrier head. The exposed surface of the substrate, the surface with the layer deposition, is typically placed against a rotating polishing pad. The carrier head provides a controllable load on the substrate to urge it against the polishing pad. A polishing slurry with abrasive particles is typically supplied to the surface of the polishing pad and spreads in between the substrate and the polishing pad. The polishing pad and the carrier head each rotate at a constant rotational speed and the abrasive slurry removes material from one or more of the layers.PATENTAttorney Docket No.: 44024954WO01

[0004] Conventional CMP operations, however, cannot determine the rotational orientation of the substrate relative to the carrier head while the substrate is being polished. Thus, substrate precession may go undetected during polishing. As a result, it may also be impossible with conventional techniques to determine the true speed of substrate rotation during polishing. There is a need in the art to determine the rotational orientation and rotational speed of a substrate in situ to improve the analysis of a CMP process.SUMMARY

[0005] In one embodiment, a method of determining an orientation of a substrate during polishing includes generating a virtual model of an interface between a polishing pad and a substrate based on rotating the polishing pad about a surface of the substrate, wherein the polishing pad comprises a sensor configured to scan the surface of the substrate. The method further includes using the sensor to scan the surface of the substrate and produce a signal. The method further includes comparing the signal to one or more signals associated with the virtual model to determine a rotational orientation of the substrate.

[0006] In one embodiment, a polishing system includes a polishing station, including: a platen including a polishing pad and a sensor configured to scan the surface of a substrate; a carrier head configured to rotate a substrate; and a computing device capable of: generating a virtual model of an interface between the polishing pad and the substrate based on rotating the polishing pad about a surface of the substrate; using the sensor to scan the surface of the substrate and produce a signal; and comparing the signal to one or more signals associated with the virtual model to determine a rotational orientation of the substrate.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] So that the manner in which the above recited features of the present disclosure can be understood in detail, a more particular description of the disclosure, briefly summarized above, may be had by reference to embodiments, some of which are illustrated in the appended drawings. It is toPATENTAttorney Docket No.: 44024954WO01 be noted, however, that the appended drawings illustrate only exemplary embodiments of the disclosure and are therefore not to be considered limiting of its scope, as the disclosure may admit to other equally effective embodiments.

[0008] Figure 1 depicts a schematic top view of an exemplary chemical mechanical polishing (CMP) system.

[0009] Figure 2 depicts a schematic sectional view of an exemplary polishing station of the CMP system from Figure 1 according to embodiments described herein.

[0010] Figure 3 depicts a top view of a front surface of a substrate showing the scan paths of sensors of the polishing station of Figure 2 according to embodiments described herein.

[0011] Figure 4 depicts a schematic top view of the carrier head on the polishing pad from the polishing station of Figure 2 according to embodiments described herein.

[0012] Figure 5 depicts a schematic diagram of a front surface of a substrate according to embodiments described herein.

[0013] Figure 6 depicts an examples of sensor signal graphs, virtual model graphs, and a comparison of a sensor signal graph and a virtual model graph according to embodiments described herein.

[0014] Figure 7 is a flowchart of a method of determining an orientation of a substrate during polishing according to embodiments described herein.

[0015] To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures. It is contemplated that elements and features of one embodiment may be beneficially incorporated in other embodiments without further recitation.PATENTAttorney Docket No.: 44024954WO01DETAILED DESCRIPTION

[0016] An apparatus and methods for determining the rotational orientation of a substrate in real time during polishing are disclosed herein. This rotational orientation is used to improve the evaluation of chemical mechanical polishing (CMP) processes, such as by allowing for a determination of the rotational speed of a substrate during polishing. Additionally, by determining the rotational orientation of the substrate in real time during polishing, the substrate does not have to be removed from a polishing module to ascertain the rotational orientation of the substrate prior to polishing the substrate on a second or third polishing station.

[0017] Figure 1 is a top plan view illustrating one embodiment of a CMP system 100. The CMP system 100 includes a factory interface module 102, a cleaner 104, a polishing module 106, and a controller 190. A substrate 115, such as a silicon wafer with one or more layers deposited thereon, is processed within the CMP system 100 to polish the surface of the substrate 115.

[0018] A wet robot 108 is provided to transfer the substrates 115 between the factory interface module 102 and the polishing module 106. The wet robot 108 may also be configured to transfer the substrates 115 between the polishing module 106 and the cleaner 104. The factory interface module 102 includes a dry robot 110 which is configured to transfer the substrates 115 between one or more cassettes 114, one or more transfer platforms 116, one or more metrology stations 117, and one or more pre-aligner stations 118 of the factory interface 102. Substrates 115 are loaded into the CMP system 100 via the cassettes 114. In one embodiment depicted in Figure 1 , four substrate storage cassettes 114 are shown. The dry robot 110 within the factory interface 102 has sufficient range of motion to facilitate transfer between the four cassettes 114 and the one or more transfer platforms 116. Optionally, the dry robot 110 may be mounted on a rail or track 112 to position the robot 110 laterally within the factory interface module 102. The dry robot 110 additionally is configured to receive the substrates 115 from the cleaner 104 and return the clean polished substrates to the substrate storage cassettes 114.PATENTAttorney Docket No.: 44024954WO01

[0019] Figure 1 shows an exemplary polishing module 106 that includes a plurality of polishing stations 124 on which the substrates 115 are polished while being retained in a carrier head 210 (e.g., polishing head). Each polishing station 124 includes a conditioning assembly 132 and a polishing fluid delivery module 135. While the polishing module 106 is shown having three polishing stations 124, the polishing module 106 may have more than three polishing stations 124. For example, the polishing module 106 may have a two pairs of polishing stations 124, each pair of stations 124 processing a substrate 115 independently of the other pair. The polishing stations 124 are sized to interface with one or more carrier heads 210 to facilitate polishing the substrate 115. The carrier heads 210 are coupled to a carriage (not shown) that is mounted to an overhead track 128 that is shown in phantom in Figure 1. The overhead track 128 allows the carriage to be selectively positioned around the polishing module 106 which facilitates positioning of the carrier heads 210 selectively over the polishing stations 124 and the load cup 122. In the embodiment depicted in Figure 1 , the overhead track 128 has a circular configuration which allows the carriages retaining the carrier heads 210 to be selectively and independently rotated over and / or clear of the load cups 122 and the polishing stations 124. Additionally, the overhead tracks 128 facilitate the carriage sweeping the rotating carrier heads 210 relative to a polishing station 124 during polishing. The polishing stations 124 will be described in greater detail in relation to Figure 2.

[0020] Each polishing station 124 includes a polishing pad 204 having a polishing surface (e.g., a polishing surface 204A in Figure 2) capable of polishing a substrate 115. Each polishing station 124 includes a conditioning assembly 132 and a polishing fluid delivery module 135. In one embodiment, the conditioning assembly 132 may comprise a pad conditioning assembly 140 which dresses the polishing surface of the polishing pad 204 by removing polishing debris and opening the pores of the polishing pad 204 by use of a pad condition disk 133. In another embodiment, the polishing fluid delivery module 135 may comprise a fluid delivery arm 134 to deliver a slurry. In one embodiment, each polishing station 124 comprises a pad conditioningPATENTAttorney Docket No.: 44024954WO01 assembly 132. In one embodiment, the fluid delivery arm 134 is configured to deliver a fluid stream (e.g., a slurry 222 in Figure 2) to a polishing station 124. The polishing pad 204 is supported on a platen (e.g., a platen 202 in Figure 2) which rotates the polishing pad 204 during processing. Each polishing station 124 includes a polishing pad 204 secured to a rotatable platen 202. Different polishing pads 204 may be used at different polishing stations 124 to control the material removal of the substrate 115.

[0021] At least one load cup 122, such as the two load cups 122 shown in Figure 1 , is near the lower right corner of the polishing module 106 between the polishing stations 124 closest to the wet robot 108. The load cups 122 may serve multiple functions, including washing the carrier head 210, receiving the substrate 115 from the wet robot 108, washing the substrate 115, and loading the substrate 115 into the carrier heads (e.g., a carrier head 210 in Figure 2).

[0022] The substrate 115 will typically have a reference mark, such as a notch, flat edge, or other type of feature that can be used to identify crystalline orientations of the substrate 115 and note a rotational orientation of a front surface of the substrate 115 relative to a central axis. In certain embodiments, the factory interface module 102 can also include a pre-aligner 118 to position the substrate 115 in a known and desirable rotational orientation. The prealignment of the substrate 115 to a desired rotational orientation allows the substrate 115 to be transferred to the load cup 122 having a known rotational orientation. Thus, the carrier head 210 is able to retrieve the substrate 115 at a known rotational orientation relative to the carrier head 210. For example, the pre-aligner 118 may include a reference mark detection system, such as an optical interrupter sensor (not shown), to sense when the reference mark is at a specific angular position.

[0023] In certain embodiments, the substrate 115 is placed in the metrology station 117 by the dry robot 110 prior to placing the substrate 115 on the transfer platform 116. For example, the dry robot 110 may transfer the substrate 115 from the pre-aligner 118 to the metrology station 117. The metrology station 117 is used to measures various aspects of the substrate 115. The metrologyPATENTAttorney Docket No.: 44024954WO01 station 117 may use an optical, eddy current, resistive, or other sensors to measure the substrate 115. For example, the metrology station 117 may measure a thickness of the upper layer on the patterned surface of the substrate 115. The controller 190 receives the measurements which may be used to facilitate processing the substrate 115 within the CMP system 100. The dry robot 110 may transfer the substrate 115 to the transfer platform 116 after the substrate 115 is measured in the metrology station 117.

[0024] The wet robot 108 is configured to transfer the substrate 115 from the transfer platform 116 to one of the load cups 122. A rinsed-clean carrier head 210 is moved above the load cup 122 with the unpolished substrate 115. The unpolished substrate 115 is thereafter chucked to the carrier head 210, which then moves to a position above the pad 204 of a polishing station 124 to begin the CMP process.

[0025] The controller 190 controls aspects of the CMP system 100 during a CMP process (e.g., polishing process, polishing operation, polishing). In certain embodiments, the controller 190 is one or more programmable digital computers executing digital control software. The controller 190 can include a CPU (e.g., processor) situated near the polishing apparatus, e.g., a programmable computer, such as a personal computer. The controller can include a memory and support circuits 193. The controller 190 can, for example, coordinate rotation of the polishing pad 204 and the carrier head 210 to perform the desired CMP process and to facilitate monitoring for the endpoint of the CMP process. The CMP process system 100 is powered by power source 180, such as an electric power source configured to supply electric power to the components of the CMP process system 100.

[0026] The platen 202 and the carrier head 210 each have a rotation sensor such as an encoder, to determine their rotational position during the CMP operation. As shown in Figure 1 , a platen encoder 195, a first head encoder 196, and a second head encoder 197 are integrated into the controller 190. The platen encoder 195 is configured to determine the rotational (e.g., angular) orientation of the platen 202 and the pad 204. The first head encoder 196 isPATENTAttorney Docket No.: 44024954WO01 configured to determine the rotational orientation of each carrier head 210. The second head encoder 197 is configured to determine the location of each carrier head 210 above the polishing pad 204 (e.g., along the sweep path 302 of the carrier head 210 in Figure 3A). Thus, the controller 190 is able to determine and track the rotational orientation of the carrier head 210 with respect to the platen 202 during the CMP process. In some embodiments, each carrier head 210 has its own dedicated first and second head encoders 196, 197. In further embodiments, the controller 190 may calculate a rotation rate of the carrier head 210 and / or platen 202 and polishing pad 204 using the encoder and an internal timing element.

[0027] Encoder may further comprise a substrate orientation module 198. The substrate orientation module 198 may include a virtual model of a substrate. As discussed in further detail below with respect to Figure 6, the virtual model may allow the substrate orientation module 198 to determine the orientation of the substrate in real time during a CMP process based on a signal received from a sensor 227 that is embedded within the platen 202. Substrate orientation module 198 may further comprise a machine learning model such as a neural network that is configured to generate a virtual model for a given substrate design.

[0028] The substrate 115 may be polished in one or more of the polishing stations 124. For example, a carrier head 210 may retrieve an unpolished substrate 115 from a load cup 122. The carrier head 210 and substrate 115 chucked thereto are then moved to a first polishing station 124, such as the polishing station 124 in the upper right corner of the polishing module 106 closest to the cleaner 104. The substrate 115 is then subjected to a CMP polishing operation on the first polishing station 124, such as removing a first layer formed on the substrate 115. Once the substrate 115 is done polishing in the first polishing station 124, then the carrier head 210 moves the substrate 115 to a second polishing station 124 (e.g., the polishing station 124 in the upper left corner of the polishing module 106) for additional CMP polishing. For example, the second polishing station 124 may polish the surface of thePATENTAttorney Docket No.: 44024954WO01 substrate 115 to form trench lines of a desired height. In some embodiments, the carrier head 210 and substrate 115 may optionally be transferred from the second polishing station 124 to a third polishing station 124 (e.g., the polishing station 124 in the lower left corner of the polishing module 106) to subject the substrate 115 to additional polishing.

[0029] After polishing, the carrier head 210 moves the polished substrate 115 chucked thereto above a load cup 122 where the polished substrate 115 is thereafter placed into the load cup 122. The wet robot 108 transports the polished substrate 115 from the load cup 122 to a cleaning chamber in the cleaner 104, where slurry residues and other contaminants that have accumulated on the substrate’s 115 surface during polishing are removed. In the embodiment depicted in Figure 1 , the cleaner 104 includes two pre-clean modules 144, two megasonic cleaner modules 146, two brush box modules 148, two spray jet modules 150, and two dryers 152. The dry robot 110 then removes the substrate 115 from the cleaner 104. In some embodiments, the dry robot 110 transfers the substrate 115 to the metrology station 117 to be measured again. In certain embodiments, the post-polish measurements can be used to adjust the polishing process parameters for a subsequent substrate. Finally, the dry robot 110 returns the substrate 115 to one of the cassettes 114.

[0030] Figure 2 illustrates a schematic cross-sectional view of a polishing station 124 of the example CMP system 100 of Figure 1. As shown, the polishing station 124 further includes a plurality of sensors 227. A substrate 115 disposed in the carrier head 210 is shown engaged with the polishing surface 204A of the pad 204 that is coupled to the platen 202.

[0031] The polishing pad 204 may be secured to the platen 202, such as being secured using an adhesive, such as a pressure sensitive adhesive (PSA) layer (not shown), disposed between the polishing pad 204 and the platen 202. The carrier head 210, facing the platen 202 and the polishing pad 204 mounted thereon, includes a flexible diaphragm 212 configured to impose different pressures against a backside surface of a substrate 115 that is disposed between the carrier head 210 and the polishing pad 204. This flexiblePATENTAttorney Docket No.: 44024954WO01 diaphragm 212 is also configured to chuck the substrate 115 to the carrier head 210 to allow the carrier head 210 to move the substrate 115 around the polishing module 106. The carrier head 210 includes a carrier ring 218 surrounding the substrate 115 which holds the substrate 115 within the head 210 during polishing. The carrier head 210 rotates about a carrier head axis 216 while the flexible diaphragm 212 urges the front surface of the substrate 115 against the polishing surface 204A of the polishing pad 204. During polishing, a downforce on the carrier ring 218 urges the carrier ring 218 against the polishing pad 204 to improve the polishing process uniformity and prevent the substrate 115 from slipping out from under the carrier head 210. In certain embodiments, the carrier head 210 includes a shaft 211 which has an axis that is colinear with carrier head axis 216. In further embodiments, the platen 202 and the carrier head 210 each have a mechanism or motor (not shown) driving their rotation.

[0032] In some embodiments, the platen 202 and polishing pad 204 both rotate about a common platen axis 205. In some embodiments, the polishing pad 204 rotates in the same rotational direction as the rotation direction of the carrier head 210. For example, the polishing pad 204 and carrier head 210 both rotate in a counter-clockwise direction. The polishing pad 204 and carrier head 210 may be rotated at the same or different speed during a polishing operation. As shown in Figure 2, the polishing pad 204 has a surface area that is greater than the front surface of the substrate 115. However, in further embodiments, the polishing pad 204 has a surface area that is less than the surface area of the front surface of the substrate 115.

[0033] Figure 2 also shows an exemplary embodiment of one of the sensors 227. Each sensor 227 is positioned radially from the platen axis 205. The sensor 227 may be an eddy current sensor or an inductive current sensor. The eddy current sensor and inductive current sensor may be embedded in the platen 202 and / or pad 204. The sensor 227 may also be an optical sensor disposed in a platen opening formed in the platen 202 and beneath an optically transparent feature (e.g., window) of the polishing pad 204. An optical sensorPATENTAttorney Docket No.: 44024954WO01 may direct light through a platen opening and window at the front surface of the substrate 115. The sensor 227 may detect properties of the front surface as the sensor 227 passes beneath the substrate 115 during polishing. The controller 190 uses the data collected by the sensors 227 to determine the orientation of the substrate 115 in real time. For example, the controller may compared the data collected by the sensor 227 to data associated with a virtual model of the substrate 115, as discussed in further detail below with respect to Figure 6. While the polishing station 124 is shown having three sensors 227 disposed around the platen axis 205, the polishing station 127 may include less than or more than three endpoint sensors 227.

[0034] Each sensor 227 may be positioned at a fixed distance from the rotational center (e.g., platen axis 205) of the platen 202. The platen encoder 195 tracks the rotational position of the platen 202 and pad 204. The controller 190 is able to determine the location of the sensor 227 as the platen 202 rotates based on the fixed location of the sensor 227 and the rotational information obtained from the platen encoder 195.

[0035] Because the sensors 227 enable the controller 190 to determine the rotational orientation of the substrate 115 in real time during polishing, the substrate 115 does not have to be removed from the polishing module 106, passed through the cleaner 104, and placed into the pre-aligner 118 or metrology station 117 to ascertain the rotational orientation of the substrate 115 prior to polishing substrate 115 on a second or third polishing station 124. Also, since the controller 190 also knows the rotational orientation of the platen 202 and pad 204, the controller 190 can determine the precession of the substrate 115. As a result, the true rotational speed of the substrate 115 can be calculated, since the rotational speed of the platen 202 and pad 204 are also known.

[0036] Figure 3 is an example top view of a substrate 115 to illustrate the front surface 230 of the substrate 115 that is engaged with the polishing pad 204 during polishing. The front surface 230 includes a patterned portion 232 and a non-patterned portion 234. The patterned portion 232 (e.g., patternedPATENTAttorney Docket No.: 44024954WO01 surface) is the portion of the substrate 115 where a plurality semiconductor devices are formed during one or more processes. As shown, the patterned portion 232 is divided into a plurality of full dies 233 arranged in a grid-pattern. Each die 233 is a particular semiconductor device being formed on the substrate 115. For example, the semiconductor devices include one or more layers formed by from one or more processes, such as through physical vapor deposition (PVD) or atomic layer deposition (ALD).

[0037] The non-patterned portion 234 is the portion of the front surface 230 around the patterned portion 232. Semiconductor devices are not formed on the non-patterned surface 234. The non-patterned portion 234 may be exposed to the same processing environments that form the patterned portion 232. Materials, such as barrier metals, may be deposited on the non-patterned surface 234 while the semiconductor devices are formed on the patterned portion 232. In some embodiments, the non-patterned portion 234 may be a partially patterned portion that includes only partial, and not full, dies. The surface area of the non-patterned portion 234 may not be uniform around the patterned portion 232. As shown in Figure 3, the surface area of the nonpatterned surface 234 fluctuates around the patterned portion 232 depending on the shape of the patterned portion 232. Thus, there are portions of the nonpatterned surface 234 with a surface area greater than other portions.

[0038] The substrate 115 includes a reference mark 236 at the edge of the substrate 115 and thus at the edge of the non-patterned portion 234. The reference mark 236 is a fixed feature formed on the substrate 115 depending on the doping type and crystalline orientation of the substrate 115. While the reference mark 236 is shown as a v-shaped notch formed on the edge of the substrate 115 in Figure 3, the reference mark 236 may be another feature. For example, the reference mark 236 may be one or more flat edges of the substrate 115.

[0039] The substrate 115 has a first line of symmetry 235 that passes through the center of the reference mark 236. The non-patterned surface 234 is generally symmetrical about this first line of symmetry 235. Thus, there is aPATENTAttorney Docket No.: 44024954WO01 first region 237 on either side of the line of symmetry 235 adjacent to the reference mark 236 that has substantially the same surface area. Additionally, there is a similar pair of second regions 238 on the opposite edge of the substrate 115 as the reference mark 236 that has a similar surface area as the first region 237. While the non-patterned surface 234 may be generally symmetric about the first line of symmetry 235, the circuits formed in the individual dies 233 of the patterned surface 232 may or may not be symmetric about this first line of symmetry 235.

[0040] In some embodiments, a layer partially or fully covers the front surface 230, with both the patterned portion 232 and non-patterned portion 234 being fully or partially covered by the layer. This layer may be deposited to form another layer or feature on the dies 233 of the patterned portion 232 that will be polished down in the CMP system. The sensor 227 is still able to scan the edge of the substrate 115 to locate the reference mark 236 even if a layer is deposited over both the patterned portion 232 and the non-patterned portion 234. For example, the sensor 227 may obtain data that shows the part of the layer scanned by the sensor 227 was over the underlying patterned portion 232 or non-patterned portion 234. In other words, the controller 190 can differentiate between the patterned portion 232 and non-patterned portion 234 even if both are at least partially obscured by the same layer. Additionally, the data obtained by the sensor 227 can show variations in the material being scanned, such as variations in the area of the non-patterned portion 234, even if a layer is fully or partially covering the front surface 230.

[0041] Figure 3 also illustrates exemplary first scan path 501 , an exemplary second scan path 502, and an exemplary third scan path 503 across the front surface 230. Each scan path 501 , 502, 503 corresponds to the path a respective sensor 227 of the polishing station 124 shown in below in Figure 4 takes as it passes below the front surface 230. As shown, each scan path is in the shape of an arc due to the motion of the pad 204, the carrier head 210, and the platen 202 during polishing. Each sensor 227 scans the outer surface along the respective endpoint scan path 501 , 502, 503 multiple times during a CMPPATENTAttorney Docket No.: 44024954WO01 process, allowing for a real-time determ ination of the orientation of the substrate 115. The orientation of scan paths relative to one another may vary depending on the location of the sensors 227 relative to one another. Additionally, the angle and curvature of the scan paths may vary based on the rotational velocity of the pad 204 and substrate 115 relative to each other, as well as the position of the carrier head 210 and platen 202 relative to the pad 204.

[0042] Figure 4 illustrates a top schematic plan view of the polishing station 124. The conditioning assembly 132 and the polishing fluid delivery module 135 are omitted. The head 210 is moveable relative to the pad 204 along a sweep path 302 to sweep the substrate 115 along the polishing surface 204A during the polishing process. The sensors 227 cross the sweep path 302 as the platen 202 rotates. The sensors 227 pass beneath the substrate 115 when the carrier head 210 places the substrate 115 at one or more positions along the sweep path 302 that is in the travel path of the sensors 227 as the platen 202 rotates.

[0043] Figure 5 illustrates an example schematic diagram 310 of the front surface 230 of the substrate 115. The schematic diagram 310 shows the features of the front surface 230, such as the patterned portion 232 and the non-patterned portion 234. As discussed in further detail below with respect to Figure 6, the schematic diagram 310 may be provided to a machine learning model in order to generate a virtual model of the interface between the polishing pad 204 and the front surface 230 of a substrate represented by the schematic diagram 310. The virtual model may comprise a set of potential sensor 227 scans along possible scan paths.

[0044] Figure 6 illustrates a graph 600 of an exemplary signal of the data collected by the sensor 227 as it scanned the front side of the rotating substrate 115. This signal shows the signal intensity as the sensor 227 passed about the surface 230 of the substrate 115. The signal may be dependent on the pattern on the surface 230 of the substrate as well as the path that the sensor 227 takes across the surface 230 of the substrate. For example, for substrates 115 with different patterns, the signal may be different for the same sensor 227 path; forPATENTAttorney Docket No.: 44024954WO01 substrates 115 with the same pattern, the signal may be different for the same sensor 227 path. However, for substrates 115 that have similar patterns on the surface 230, the signals may be similar when the same (or similar) sensor 227 path is taken. The signal graph 600 shown in Figure 6 corresponds to a signal taken by a sensor 227 of a substrate 115 that is represented by the schematic diagram 310. The signal shown in the graph 600 was taken when the substrate 115 was in a particular rotational orientation at a particular location relative to the pad 204.

[0045] A virtual model of the interface between the pad 204 and the front surface 230 of the substrate 115 represented by the schematic diagram 310 may be created. The virtual model may comprise a set of potential sensor scans 610. The potential sensor scans 610 shown in Figure 6 represent possible scans that a sensor 227 would make at various different paths along the surface 230. For example, every possible path that the sensor 227 may take across the surface 230 of the substrate 115 may be represented by the virtual model.

[0046] An empirical model of a given substrate may be used to create the virtual model. For example, the virtual model may be created based on using the empirical model to train a machine learning model to generate the potential sensor scans 610. The empirical model may be created by obtaining direct measurements of the features of the given substrate. The empirical model may comprise data relating to the features of a substrate, such as surface features. For example, the empirical model may comprise a three dimensional model of the features of the given substrate (e.g., the patterns on the surface of the given substrate), a topographical map of the given substrate, and / or any other representation of the features of the given substrate as known in the art. The empirical model may also comprise sensor scans for different paths along the surface of the given substrate. Thus, the empirical model may enable determining the orientation of the given substrate during a CMP process. For example, a particular signal associated with the empirical model may correspond to a particular path across the surface of the given substrate. If aPATENTAttorney Docket No.: 44024954WO01 scan taken in real-time during a CMP process matches the particular signal, it may be determined that the sensor traveled across the surface of the substrate along the particular path in order to create the scan.

[0047] In some embodiments, the model creation component comprises a machine learning model that is trained to generate virtual models based on a signal from a sensor 227 and the empirical model.

[0048] The machine learning model may be trained through a supervised learning process. Supervised learning techniques generally involve providing training inputs to a machine learning model, such as a neural network. The machine learning model processes the training inputs and outputs predictions based on the training inputs. The predictions are compared to the known labels associated with the training inputs to determine the accuracy of the machine learning model, and parameters of the machine learning model are iteratively adjusted until one or more conditions are met. For instance, the one or more conditions may relate to an objective function (e.g., a cost function or loss function) for optimizing one or more variables (e.g., model accuracy). In some embodiments, the conditions may relate to whether the outputs produced by the machine learning model based on the training inputs match the known labels associated with the training inputs or whether a measure of error between training iterations is not decreasing or not decreasing more than a threshold amount. The conditions may also include whether a training iteration limit has been reached. Parameters adjusted during training may include, for example, hyperparameters, values related to numbers of iterations, weights, functions used by nodes to calculate scores, and / or the like. In some embodiments, validation and testing are also performed for a machine learning model, such as based on validation data and test data, as is known in the art.

[0049] The supervised learning process for the machine learning model used to generate the virtual model may comprise providing one or more features of the empirical model and a scan signal taken in real-time during a CMP process to the machine learning model as input. Based on these inputs, the machine learning model may be used to predict scan signals for variousPATENTAttorney Docket No.: 44024954WO01 possible sensor paths across the surface of the substrate. For example, the machine learning model may be provided with a scan path from the empirical model and a corresponding sensor signal from the empirical model. The machine learning model may then be used to predict a signal for another path, and the predicted signal may be compared to a signal that corresponds to that path according to the empirical model. Parameters of the machine learning model (e.g., weights) may be iteratively adjusted based on a variance between the predicted signal and the empirical model signal associated with that path. The process may be repeated for different paths across the surface of the substrate. In some embodiments, the supervised learning process may comprise using the machine learning model to predict an orientation of a substrate based on a signal, and one or more parameters of the machine learning model may be adjusted based on variances between the predicted orientation and the orientation that the empirical model indicates is associated with the signal. As an example, the predicted orientation may comprise a predicted three dimensional map that has a particular orientation. Thus, in some embodiments, the machine learning model may be used to predict an orientation based on a real-time sensor scan.

[0050] Once trained, the machine learning model may be used to generate a virtual model comprising potential signals 610 for different paths across the surface of the substrate 115. The machine learning model may be provided with the schematic diagram 310 as input. The schematic diagram 310 may comprise data relating to features of the surface of the substrate 115 (e.g., the schematic diagram 310 may comprise a file that contains the pattern on the front surface 230 of the substrate 115) as input. For example, the schematic diagram 310 may comprise a three dimensional map or topographic map (e.g., the map may be similar to the map of the empirical model but customized for the substrate 115 that is to be polished). Since the machine learning model has been trained to predict scan signals for different paths across a particular substrate based on feature data corresponding to the particular substrate, the machine learning model may be used to predict the potential signals 610 for the substrate 115 based on the schematic diagram 310. The potential signals 610 may comprisePATENTAttorney Docket No.: 44024954WO01 signals for the various paths a sensor may take across the surface of the substrate 115.

[0051] In other embodiments, the virtual model may comprise an empirical model generated by scanning the substrate 115 along the various possible paths. As discussed above, an empirical model may be created by obtaining direct measurements of the features of the substrate 115. The empirical model may comprise a three dimensional model of the features of the substrate 115, a topographical map of the substrate 115, and / or the like. The empirical model may comprise sensor signals for each path along the surface of the substrate 115. Thus, the path that the sensor 227 takes across the surface of the substrate 115 may be determined by comparing a real-time sensor signals to the virtual model sensor signals to find a closest match. The virtual model signal that is the closest match may correspond to the path taken by the sensor 227. Creating an empirical model for the layout of each substrate to be polished (e.g., the layout shown in schematic diagram) may require more time and resources than using a machine learning model to create the virtual model, as discussed above.

[0052] Once the virtual model is generated for the substrate 115, the virtual model may be used to determine the rotational orientation of the substrate 115 in real time during CMP processing. To determine the rotational orientation of the substrate, a signal obtained in real time from a sensor 227 may be compared to one or more signals of the virtual model. A signal of the virtual model that is most similar to the real-time signal may be identified. For example, this identification may be performed based on correlating the peaks of the signals. The virtual model signal that most closely matches the real-time signal (or most closely matches the peaks of the real-time signal) may be identified as a match. The real-time rotational orientation of the substrate 115 may then be determined to be the rotational orientation associated with the identified virtual model signal. For example, the identified virtual model signal was generated based on a particular path that a sensor takes across the surface; if the real-time signal matches the virtual model signal, then this mayPATENTAttorney Docket No.: 44024954WO01 mean that the real-time path matches the particular path used to generate the virtual model signal. When the path a sensor takes across the surface of a substrate is known, the rotational orientation of the substrate may also be known because the location of the sensor is known (e.g, because the location of the sensor is determined by an encoder such as platen encoder 195). The graph 620 in Figure 6 illustrates a real signal obtained from a sensor 227 and a signal from a virtual model that was identified as most similar to the real signal.

[0053] The precession of the substrate 115 and the rotational velocity of the substrate 115 may be determined based on the determined rotational orientation. For example, an encoder such as platen encoder 195 may track the rotational orientation of the platen 202. The substrate 115 precession may be calculated based on comparing the platen 202 orientation to the substrate 115 orientation. For example, the difference between the orientations may be identified as the precession. The rotational velocity of the substrate 115 may then be calculated based on the precession. For example, an encoder such as platen encoder 195 may track the rotational velocity of the platen 202. The rotational velocity of the substrate 115 may be calculated to be slightly higher or slightly lower than the rotational velocity of the platen 202, depending on the direction of the precession.

[0054] Based on the determined rotational orientation and rotational velocity, one or more actions may be performed. For example, the rotational orientation may be adjusted to a target orientation. The target orientation may be an orientation required for additional substrate processing such as a continuation of the CMP process at a different polishing station. As discussed above, because the rotational orientation may be determined in real time during a CMP process, the substrate does not have to be removed from a polishing module to ascertain the rotational orientation of the substrate prior to polishing the substrate on a second or third polishing station. Also, determining the true rotational velocity of the substrate during CMP processing allows for improved monitoring of the CMP process.PATENTAttorney Docket No.: 44024954WO01

[0055] Figure 7 illustrates a flow chart of an exemplary method 700 of determining an orientation of a substrate during polishing. The controller 190 may control each operation of the method 700.

[0056] Operations 700 begin at step 710 with generating a virtual model of an interface between a polishing pad and a substrate based on rotating the polishing pad about a surface of the substrate, wherein the polishing pad comprises a sensor configured to scan the surface of the substrate. In some embodiments, the virtual model is created based on providing a schematic diagram of the surface of the substrate as input to a machine learning model, wherein the machine learning model has been trained to generate predicted signals for different rotational orientations of substrates. Certain embodiments provide that the machine learning model has been trained through a supervised learning process comprising comparing predicted signals for a given rotational orientation to signals associated with an empirical model. According to some embodiments, the supervised learning process further comprises iteratively adjusting parameters of the machine learning model based on detected variances between the predicted signals and the signals associated with the empirical model. Some embodiments provide that the virtual model comprises an empirical model. In certain embodiments, the sensor comprises an eddy current sensor embedded in a platen that is connected to the polishing pad. Some embodiments provide that the sensor is positioned between a rotational center of the polishing pad and an edge of the polishing pad.

[0057] Operations 700 continue at step 720 with using the sensor to scan the surface of the substrate and produce a signal.

[0058] Operations 700 continue at step 730 with comparing the signal to one or more signals associated with the virtual model to determine a rotational orientation of the substrate. Certain embodiments provide that the comparing comprises identifying a given signal associated with the virtual model that most closely matches the signal. In some embodiments, a measure of substrate precession is calculated based on a variance between the determined rotational orientation and a known rotational orientation of a carrier head used to hold thePATENTAttorney Docket No.: 44024954WO01 substrate. Some embodiments provide that a rotational speed of the substrate is calculated based on a speed of the carrier head and the measure of substrate precession.

[0059] According to certain embodiments, one or more actions may be performed based on the determined rotational orientation, such as rotating the substrate to achieve a target rotational orientation; or transporting the substrate to another location for additional processing.

[0060] While the foregoing is directed to embodiments of the present disclosure, other and further embodiments of the disclosure may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.

Claims

PATENTAttorney Docket No.: 44024954WO01What is claimed is:1 . A method of determining an orientation of a substrate during polishing, comprising: generating a virtual model of an interface between a polishing pad and a substrate based on rotating the polishing pad about a surface of the substrate, wherein the polishing pad comprises a sensor configured to scan the surface of the substrate; using the sensor to scan the surface of the substrate and produce a signal; and comparing the signal to one or more signals associated with the virtual model to determine a rotational orientation of the substrate.

2. The method of Claim 1 , wherein the virtual model is created based on providing a schematic diagram of the surface of the substrate as input to a machine learning model, wherein the machine learning model has been trained to generate predicted signals for different rotational orientations of substrates.

3. The method of Claim 2, wherein the machine learning model has been trained through a supervised learning process comprising comparing predicted signals for a given rotational orientation to signals associated with an empirical model.

4. The method of Claim 3, wherein the supervised learning process further comprises iteratively adjusting parameters of the machine learning model based on detected variances between the predicted signals and the signals associated with the empirical model.

5. The method of Claim 1 , wherein the virtual model comprises an empirical model.PATENTAttorney Docket No.: 44024954WO016. The method of Claim 1 , wherein the sensor comprises an eddy current sensor embedded in a platen that is connected to the polishing pad.

7. The method of Claim 1 , further comprising performing, based on the determined rotational orientation, one or more of: rotating the substrate to achieve a target rotational orientation; or transporting the substrate to another location for additional processing.

8. The method of Claim 1 , wherein the comparing comprises identifying a given signal associated with the virtual model that most closely matches the signal.

9. The method of Claim 1 , wherein the sensor is positioned between a rotational center of the polishing pad and an edge of the polishing pad.

10. The method of Claim 1 , wherein a measure of substrate precession is calculated based on a variance between the determined rotational orientation and a known rotational orientation of a carrier head used to hold the substrate.11 . The method of Claim 10, wherein a rotational speed of the substrate is calculated based on a speed of the carrier head and the measure of substrate precession.

12. A polishing system, comprising: a polishing station, including: a platen including a polishing pad and a sensor configured to scan a surface of a substrate; a carrier head configured to rotate a substrate; and a computing device capable of: generating a virtual model of an interface between the polishing pad and the substrate based on rotating the polishing pad about the surface of the substrate;PATENTAttorney Docket No.: 44024954WO01 using the sensor to scan the surface of the substrate and produce a signal; and comparing the signal to one or more signals associated with the virtual model to determine a rotational orientation of the substrate.

13. The polishing system of Claim 12, wherein the virtual model is created based on providing a schematic diagram of the surface of the substrate as input to a machine learning model, wherein the machine learning model has been trained to generate predicted signals for different rotational orientations of substrates.

14. The polishing system of Claim 13, wherein the machine learning model has been trained through a supervised learning process comprising comparing predicted signals for a given rotational orientation to signals associated with an empirical model.

15. The polishing system of Claim 14, wherein the supervised learning process further comprises iteratively adjusting parameters of the machine learning model based on detected variances between the predicted signals and the signals associated with the empirical model.

16. The polishing system of Claim 12, wherein the virtual model comprises an empirical model.

17. The polishing system of Claim 12, wherein the comparing comprises identifying a given signal associated with the virtual model that most closely matches the signal.

18. The polishing system of Claim 12, wherein the sensor is positioned between a rotational center of the polishing pad and an edge of the polishing pad.

19. The polishing system of Claim 12, wherein the computing device calculates a measure of substrate precession based on a variance between the determinedPATENTAttorney Docket No.: 44024954WO01 rotational orientation and a known rotational orientation of a carrier head used to hold the substrate.

20. The polishing system of Claim 19, wherein the computing device calculates a rotational speed of the substrate based on a speed of the carrier head and the measure of substrate precession.

Images