Polishing apparatus using neural networks for monitoring

Abstract

A method of polishing a layer on a substrate at a polishing station includes the acts of monitoring the layer during polishing at the polishing station with an in-situ monitoring system to generate a plurality of measured signals for a plurality of different locations on the layer, generating, for each of the plurality of different locations, an estimated measurement of thickness for the location, the generating including processing the plurality of measured signals through a neural network, and at least one of detecting a polishing endpoint or modifying a polishing parameter based on each of the estimated measurements of thickness.

Description

[0001] This application is a divisional application of Chinese patent application (PCT application number PCT / US2018 / 027562) filed on April 13, 2018, with application number 201880026059.4 and entitled "Polishing Apparatus for Monitoring Using Neural Networks". Technical Field

[0002] This disclosure relates to in-situ monitoring during substrate polishing. Background Technology

[0003] Integrated circuits are typically formed on a substrate (e.g., a semiconductor wafer) by sequentially depositing conductive, semiconductor, or insulating layers onto a silicon wafer and then processing the layers.

[0004] One manufacturing step involves depositing a filler layer onto a non-planar surface and planarizing the filler layer until the non-planar surface is exposed. For example, a conductive filler layer can be deposited on a patterned insulating layer to fill trenches or holes within the insulating layer. The filler layer is then polished until the protruding patterns of the insulating layer are exposed. After planarization, the remaining conductive layer portions between the protruding patterns of the insulating layer form vias, inserts, and lines that provide conductive paths between thin-film circuits on the substrate. Alternatively, planarization can be used to planarize the substrate surface for photolithography.

[0005] Chemical mechanical polishing (CMP) is an accepted planarization method. This planarization method typically requires mounting a substrate 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 to push it toward the polishing pad. A polishing liquid (such as a slurry containing abrasive particles) is supplied to the surface of the polishing pad.

[0006] During semiconductor processing, determining one or more characteristics of a substrate (or layers on a substrate) can be important. For example, knowing the thickness of a conductive layer during a CMP process can be important to terminate the process at the correct time. Several methods can be used to determine substrate characteristics. For example, optical sensors can be used for in-situ monitoring of the substrate during chemical mechanical polishing. Alternatively or additionally, eddy current sensing systems can be used to induce eddy currents in conductive regions on the substrate to determine parameters such as the local thickness of the conductive regions. Summary of the Invention

[0007] In one aspect, a method for polishing a layer on a substrate at a polishing station includes: monitoring the layer using an in-situ monitoring system during polishing at the polishing station to generate a plurality of measured signals for a plurality of different locations on the layer; generating an estimated thickness measurement for each of the plurality of different locations; the generation step including processing the plurality of measured signals by a neural network; and at least one of the following steps: detecting a polishing endpoint or modifying polishing parameters based on each estimated thickness measurement.

[0008] In another aspect, the corresponding computer system, apparatus, and computer program stored on one or more computer storage devices are configured to perform the method. A system of one or more computers may be configured to perform a specific operation or action by means of software, firmware, hardware, or any combination thereof installed on the system that enables the system to perform actions during operation. A system of one or more computer programs may be configured to perform a specific operation or action by including instructions that, when executed by a data processing device, cause the device to perform actions.

[0009] In another aspect, the polishing system includes: a carrier for holding a substrate; a support for a polished surface; an in-situ monitoring system with sensors; a motor for generating relative movement between the sensors and the substrate; and a controller. The in-situ monitoring system is configured to generate measured signals for multiple different locations on the layer. The controller is configured to: receive multiple measured signals from the in-situ monitoring system; for each of the multiple different locations, generate an estimated thickness measurement for said location, the generation including: processing the multiple measured signals through a neural network; and based on each estimated thickness measurement, detecting a polishing endpoint, modifying polishing parameters, or both.

[0010] Any embodiment of the foregoing aspects may include one or more of the following features.

[0011] A second plurality of measured signals can be acquired for a second plurality of different locations on the layer. For each of the second plurality of different locations, an estimated thickness measurement result for said location can be generated based on the measured signal for that location. Generating the estimated thickness measurement result may include using a static formula that correlates a plurality of values ​​of the measured signals with a plurality of values ​​of the estimated thickness measurement result. A third plurality of measured signals can be acquired for the layer on the second substrate. Each of the third plurality of measured signals may correspond to a location among a third plurality of locations on the layer on the second substrate. For each of the third plurality of different locations, an estimated thickness measurement result for said location can be generated based on the measured signal for that location. Generating the estimated thickness measurement result may include using a static formula that correlates a plurality of values ​​of the measured signals with a plurality of values ​​of the estimated thickness measurement result.

[0012] The in-situ monitoring system may include an eddy current sensor. The neural network may include one or more neural network layers, each containing an input layer, an output layer, and one or more hidden layers; each neural network layer may contain one or more neural network nodes. Each neural network node may be configured to process inputs according to a set of parameters to produce an output. The input to the neural network node in the input layer may include a measurement of the wear degree of the pad at the polishing station. One or more distinct locations may include anchor locations, and determining each first thickness measurement may include: standardizing each measured signal based on the measured signal at the anchor location to update the measured signal. The anchor locations may be spaced apart from the edge of the substrate. Each estimated thickness measurement may be a standardized value, and the method may further include: converting each estimated thickness measurement into a non-standardized value using the measured signal at the anchor location to update the estimated thickness measurement.

[0013] For each of one or more different locations within a layer, a reference true measurement of the thickness can be obtained. The measurement error between the estimated thickness measurement at each location and the corresponding reference true measurement for that location can be calculated. The parameters of the neural network system can be updated based on this measurement error. The reference true measurement of the thickness can be determined using the four-point probes method. Updating the parameters of the neural network system based on the measurement error can include the gradient of the measurement error propagated back through multiple layers of the neural network.

[0014] Some embodiments may include one or more of the following advantages. In-situ monitoring systems (e.g., eddy current monitoring systems) can generate signals as sensors scan across the substrate. The system can compensate for distortion in signal portions corresponding to substrate edges. The signals can be used for closed-loop control of endpoint control and / or polishing parameters (e.g., bearing head pressure), thus improving intra-wafer non-uniformity (WIWNU) and inter-wafer non-uniformity (WTWNU).

[0015] Details of one or more embodiments are set forth in the accompanying drawings and the description below. Other aspects, features, and advantages will be apparent from the specification, the drawings, and the claims. Attached Figure Description

[0016] Figure 1A This is a partial cross-sectional schematic side view of a chemical mechanical polishing station that includes an eddy current monitoring system.

[0017] Figure 1B This is a schematic top view of a chemical mechanical polishing station.

[0018] Figure 2This is a schematic top view of a substrate being scanned by the sensor head of a polishing device.

[0019] Figure 3 This is a schematic diagram of a static formula used to determine the substrate thickness based on the measured signal.

[0020] Figure 4 This is a schematic diagram of the measured signals acquired while monitoring the position on the substrate.

[0021] Figure 5 This is an example neural network.

[0022] Figure 6 This is a flowchart of an example process for polishing a substrate.

[0023] Figure 7 This is a flowchart of an example process for using a neural network to generate an estimated measurement result for thickness.

[0024] Figure 8 This is a flowchart of an example process for training a neural network to generate a modified signal from a measured signal group.

[0025] In the various accompanying drawings, similar component symbols indicate similar components. Detailed Implementation

[0026] Polishing equipment can use in-situ monitoring systems (such as eddy current monitoring systems) to detect the thickness of the outer layer being polished on a substrate. During the polishing of the outer layer, the in-situ monitoring system can determine the thickness of the layer at different locations on the substrate. The thickness measurements can be used to trigger the polishing endpoint and / or adjust the processing parameters of the polishing process in real time. For example, the substrate carrier head can adjust the pressure on the back side of the substrate to increase or decrease the polishing rate at multiple locations on the outer layer. The polishing rate can be adjusted so that the thickness at multiple locations of the layer is substantially the same after polishing. CMP systems can adjust the polishing rate so that the polishing of multiple locations of the layer is completed approximately simultaneously. This profile control can be referred to as real-time profile control (RTPC).

[0027] In-situ monitoring systems can withstand signal distortion in measurements taken near the substrate edge. For example, eddy current monitoring systems can generate magnetic fields. Near the substrate edge, the signal may be artificially low because the magnetic field only partially covers the conductive layer of the substrate. However, if the polishing equipment uses a neural network to generate a modified signal based on the measured signal from the in-situ monitoring system, the equipment can compensate for the distortion (e.g., reduced signal strength) at the substrate edge.

[0028] Figure 1A and Figure 1BThe illustration shows an example of a polishing apparatus 100. The polishing apparatus 100 includes a rotatable disc platform 120 on which a polishing pad 110 is located. The platform can be operated to rotate about an axis 125. For example, a motor 121 can rotate a drive shaft 124 to rotate the platform 120. The polishing pad 110 may be a double-layer polishing pad having an outer polishing layer 112 and a softer backing layer 114.

[0029] The polishing apparatus 100 may include a port 130 for dispensing polishing liquid 132 (such as slurry) onto the polishing pad 110. The polishing apparatus may also include a polishing pad adjuster for abrading the polishing pad 110 to keep the polishing pad 110 in a consistent abrasive state.

[0030] The polishing apparatus 100 includes at least one support head 140. The support head 140 is operable to hold a substrate 10 against a polishing pad 110. The support head 140 may have independent control over polishing parameters (e.g., pressure) associated with each respective substrate.

[0031] Specifically, the carrier head 140 may include a retaining ring 142 to hold the substrate 10 under the flexible membrane 144. The carrier head 140 also includes a plurality of independently controllable pressure chambers defined by the membrane, such as three chambers 146a-146c, which can apply independently controllable pressure to associated sections on the flexible membrane 144 (and therefore on the substrate 10). Although only three chambers are illustrated in FIG1 for ease of illustration, one or two chambers, or four or more chambers, such as five chambers, may be present.

[0032] The bearing head 140 is suspended from a support structure 150 (e.g., a turntable or track) and connected to a bearing head rotation motor 154 via a drive shaft 152, allowing the bearing head to rotate about axis 155. Optionally, the bearing head 140 can swing laterally, for example, on a slider on the turntable 150 or track; or by the rotational swing of the turntable itself. In operation, the platform rotates about a central axis 125, and the bearing head rotates about the central axis 155 and moves laterally across the top surface of the polishing pad.

[0033] Although only one carrier head 140 is shown, more carrier heads can be provided to hold additional substrates, so that the surface area of ​​the polishing pad 110 can be used efficiently.

[0034] The polishing equipment 100 also includes an in-situ monitoring system 160. The in-situ monitoring system 160 generates a time-varying sequence of values ​​based on the thickness of the layers on the substrate. The in-situ monitoring system 160 includes a sensor head that generates measurement results; due to the relative movement between the substrate and the sensor head, measurements are taken at different locations on the substrate.

[0035] The in-situ monitoring system 160 may be an eddy current monitoring system. The eddy current monitoring system 160 includes a driving system to induce eddy currents in a conductive layer on a substrate, and a sensing system to detect the eddy currents induced in the conductive layer by the driving system. The monitoring system 160 includes a core 162, at least one coil 164, and a driving sensing circuitry 166. The core 162 is positioned in a recess 128 to rotate with the platform, and at least one coil 164 is wound around a portion of the core 162. The driving sensing circuitry 166 is connected to the coil 164 by wiring 168. The combination of the core 162 and the coil 164 provides a sensor head. In some embodiments, the core 162 protrudes above the top surface of the platform 120, for example, into a recess 118 in the bottom of the polishing pad 110.

[0036] The drive sensing circuitry 166 is configured to apply an oscillating electrical signal to the coil 164 and measure the resulting eddy current. The drive sensing circuitry can have various possible configurations, and the arrangement and location of one or more coils can have various possible configurations, such as those described in U.S. Patent Nos. 6,924,641, 7,112,960, and 8,284,560, and U.S. Patent Publications 2011-0189925 and 2012-0276661. The drive sensing circuitry 166 can be located within the same recess 128, within different portions of the platform 120, or can be located outside the platform 120 and coupled to components within the platform via a rotary connector 129.

[0037] In operation, the drive sensing circuit system 166 drives the coil 164 to generate an oscillating magnetic field. At least a portion of the magnetic field extends through the polishing pad 110 and into the substrate 10. If a conductive layer is present on the substrate 10, the oscillating magnetic field generates eddy currents in the conductive layer. The eddy currents cause the conductive layer to act as an impedance source, which is coupled to the drive sensing circuit system 166. As the thickness of the conductive layer changes, the impedance also changes, which can be detected by the drive sensing circuit system 166.

[0038] Alternatively or additionally, an optical monitoring system can be fixed to the platform 120 in the recess 128. The optical monitoring system can function as a reflectometer or an interferometer. If both systems are used simultaneously, the optical monitoring system and the eddy current monitoring system can monitor the same portion of the substrate.

[0039] CMP equipment 100 may also include a position sensor 180 (such as an optical interrupter) to sense when the core 162 is located under the substrate 10. For example, the optical interrupter may be mounted at a fixed point opposite the carrier head 140. A flag 182 is attached to the periphery of the platform. The attachment point and length of the flag 182 are selected such that the flag 182 interrupts the optical signal of the sensor 180 as the core 162 sweeps under the substrate 10. Alternatively or additionally, the CMP equipment may include an encoder to determine the angular position of the platform.

[0040] The controller 190 (such as a general-purpose programmable digital computer) receives intensity signals from the eddy current monitoring system 160. The controller 190 may include a processor, memory, and input / output (I / O) devices, as well as output devices 192 (e.g., a monitor) and input devices 194 (e.g., a keyboard).

[0041] The signal can be transmitted from the eddy current monitoring system 160 to the controller 190 via the rotary connector 129. Alternatively, the circuit system 166 can communicate with the controller 190 via a wireless signal.

[0042] Because the core 162 sweeps across the substrate with each platform rotation, information about the conductive layer thickness is accumulated in situ and continuously in real time (once per platform rotation). The controller 190 can be programmed to sample measurements from the monitoring system when the substrate substantially covers the core 162 (determined by a position sensor). As polishing progresses, the thickness of the conductive layer changes, and the sampled signal varies over time. This time-varying sampled signal can be referred to as traces. Measurements from the monitoring system can be displayed on the output device 192 during polishing, allowing the operator to visually monitor the progress of the polishing operation.

[0043] During operation, CMP equipment 100 may use eddy current monitoring system 160 to determine when bulk material of the packing layer has been removed, and / or when the underlying stop layer has been substantially exposed. Possible process control and endpoint criteria for the detector logic include local minimum and maximum values, slope changes, thresholds for amplitude or slope, or combinations thereof.

[0044] The controller 190 can also be connected to a pressure mechanism to control the pressure applied by the bearing head 140, to the bearing head rotary motor 154 to control the bearing head rotation rate, to the platform rotary motor 121 to control the platform rotation rate, or to the slurry distribution system 130 to control the slurry composition supplied to the polishing pad. Furthermore, the computer 190 can be programmed to divide the measurements from the eddy current monitoring system 160 based on each sweep under the substrate into multiple sampling zones, calculate the radial position of each sampling zone, and arrange the amplitude measurements into a radial range, as discussed in U.S. Patent No. 6,399,501. After arranging the measurements into a radial range, information about the film thickness can be fed into the closed-loop controller in real time to periodically or continuously modify the polishing pressure profile applied by the bearing head to provide improved polishing uniformity.

[0045] The controller 190 can use a correlation curve (correlated with the signal measured by the in-situ monitoring system 160 to the thickness of the layer being polished on the substrate 10) to generate an estimated measurement of the thickness of the layer being polished. Figure 3 The diagram illustrates an example of the correlation curve 303. Figure 3 In the plotted coordinate system, the horizontal axis represents the signal values ​​received from the in-situ monitoring system 160, while the vertical axis represents the thickness values ​​of the layers on the substrate 10. For a given signal value, the controller 190 can use the correlation curve 303 to generate the corresponding thickness value. The correlation curve 303 can be considered a "static" formula because it predicts the thickness value for each signal value regardless of when or where the sensor head acquires the signal. The correlation curve can be represented by various functions, such as polynomial functions or lookup tables (LUTs) combined with linear interpolation.

[0046] Reference Figure 1B and Figure 2 A change in the position of the sensor head relative to the substrate 10 can cause a change in the signal from the in-situ monitoring system 160. In other words, as the sensor head scans across the substrate 10, the in-situ monitoring system 160 will measure multiple regions 94 (e.g., measurement points) at different locations on the substrate 10. Regions 94 may be partially overlapping (see...). Figure 2 ).

[0047] Figure 4The illustration shows graph 420, which illustrates a signal 401 from the in-situ monitoring system 160 during a single pass of the sensor head under the substrate 10. Signal 401 consists of a series of individual measurements from the sensor head as it sweeps under the substrate. Graph 420 can be a function of measurement time or measurement position on the substrate (e.g., radial position). In either case, different portions of signal 401 correspond to measurement points 94 at different locations on the substrate 10 scanned by the sensor head. Therefore, graph 420 depicts the corresponding measured signal value from signal 401 for a given location on the substrate scanned by the sensor head.

[0048] Reference Figure 2 and Figure 4 Signal 401 includes a first portion 422, a second portion 424, and a third portion 426. The first portion 422 corresponds to a position in the edge region 203 of the substrate 10 when the sensor head crosses the leading edge of the substrate 10; the second portion 424 corresponds to a position in the central region 201 of the substrate 10; and the third portion 426 corresponds to a position in the edge region 203 when the sensor head crosses the trailing edge of the substrate 10. The signal may also include a portion 428, which corresponds to a measurement result outside the substrate, i.e., when the sensor head scans beyond... Figure 2 The signal generated when the area outside the edge 204 of the middle substrate 10.

[0049] Edge region 203 may correspond to the substrate portion where the sensor head's measurement point 94 covers the substrate edge 204. Central region 201 may include an annular anchor region 202 adjacent to edge region 203, and an inner region 205 surrounded by anchor region 202. The sensor head may scan these regions on sensor head path 210 and generate a sequence of measurement results corresponding to a position sequence along path 210.

[0050] In the first part 422, the signal strength ramps up from an initial strength (typically the signal generated when the substrate and carrier head are absent) to a higher strength. This is because the monitoring position shifts from an initial monitoring position that only slightly covers the substrate edge 204 (producing an initial lower value) to a monitoring position that almost completely covers the substrate (producing a higher value). Similarly, in the third part 426, the signal strength ramps down as the monitoring position shifts to the substrate edge 204.

[0051] Although the second part 424 is illustrated as flat, this is for ease of illustration, and the actual signal in the second part 424 may contain disturbances caused by both noise and layer thickness variations. The second part 424 corresponds to the monitoring position of the central scan region 201. The second part 424 includes sub-parts 421 and 423 and sub-part 427, sub-parts 421 and 423 being generated by the monitoring position of the anchor region 202 of the central scan region 201, and sub-part 427 being generated by the monitoring position of the inner region 205 of the central scan region 201.

[0052] As mentioned earlier, the signal strength variations in regions 422 and 426 are partly caused by the measurement area where the sensor covers the edge of the substrate, rather than by intrinsic variations in the thickness or conductivity of the layer being monitored. Therefore, this distortion in signal 401 can introduce errors when calculating substrate characteristic values ​​(e.g., layer thickness) near the substrate edge. To address this issue, controller 190 may include a neural network (e.g., Figure 5 The neural network 500 generates modified signals corresponding to one or more locations on the substrate 10 based on measured signals corresponding to those locations.

[0053] Now refer to Figure 5 The neural network 500 is configured, upon proper training, to generate a modified signal to reduce and / or remove distortion of the calculated signal values ​​near the edges of the substrate. The neural network 500 receives a set of inputs 504 and processes the inputs 504 through one or more neural network layers to produce a set of outputs 550. The layers of the neural network 500 include an input layer 510, an output layer 530, and one or more hidden layers 520.

[0054] Each layer of the neural network 500 contains one or more neural network nodes. Each neural network node in the neural network layer receives one or more node input values ​​(from the input 504 given to the neural network 500, or from the outputs of one or more nodes in the previous neural network layer), processes the node input values ​​according to one or more parameter values ​​to produce a starting value, and optionally applies a nonlinear transformation function (e.g., a sigmoid or tanh function) to the starting value to produce the output of the neural network node.

[0055] Each node in the input layer 510 receives one of the inputs 504 to the neural network 500 as a node input value.

[0056] The input 504 to the neural network includes measured signal values ​​from the in-situ monitoring system 160 at multiple different locations on the substrate 10, such as a first measured signal value 501, a second measured signal value 502, and so on up to an nth measured signal value 503. The measured signal values ​​can be individual values ​​in the value sequence of signal 401.

[0057] Generally, multiple different locations are contained within the edge region 203 and anchor region 202 of the substrate 10. In some embodiments, the multiple different locations are located only within the edge region 203 and anchor region 202. In other embodiments, the multiple different locations extend across all regions of the substrate.

[0058] These measured signal values ​​are received at signal input node 544. Optionally, the input node 504 of the neural network 500 may also include one or more state input nodes 516, which receive one or more process state signals 504 (e.g., for the measurement of the wear degree of the pad 110 of the polishing equipment 100).

[0059] The nodes of hidden layer 520 and output layer 530 are illustrated as receiving input from each node of the previous layer. This is the case of a fully connected feedforward neural network. However, neural network 500 can be a non-fully connected feedforward neural network or a non-feedforward neural network. Furthermore, neural network 500 may contain at least one of the following: one or more fully connected feedforward layers; one or more non-fully connected feedforward layers; and one or more non-feedforward layers.

[0060] The neural network generates a set of modified signal values ​​550 at the nodes (i.e., "output nodes") of the output layer 530. In some embodiments, there is an output node 550 for each measured signal fed to the neural network 500 from the in-situ monitoring system. In this case, the number of output nodes 550 may correspond to the number of signal input nodes 504 of the input layer 510.

[0061] For example, the number of signal input nodes 544 may be equal to the number of measurements in the edge region 203 and the anchor region 202, and there may be an equal number of output nodes 550. Therefore, each output node 550 generates a modified signal corresponding to the corresponding measured signal supplied as input to the signal input node 544, such as a first modified signal 551 for the first measured signal 501, a second modified signal 552 for the second measured signal 502, and an nth modified signal 553 for the nth measured signal 503.

[0062] In some embodiments, the number of output nodes 550 is less than the number of input nodes 504. In some embodiments, the number of output nodes 550 is less than the number of signal input nodes 544. For example, the number of signal input nodes 544 may be equal to the number of measurements in edge region 203, or may be equal to the number of measurements in edge region 203 and anchor region 202. To reiterate, each output node 550 of output layer 530 generates a modified signal corresponding to the corresponding measured signal supplied as a signal input node 504, such as a first modified signal 551 for a first measured signal 501, but only for signal input nodes 554 that receive signals from edge region 203.

[0063] The polishing apparatus 100 may use a neural network 500 to generate modified signals. These modified signals can then be used to determine the thickness of each location in a first group of locations on the substrate, such as locations in edge regions (and possibly anchor regions). For example, referencing... Figure 4 The modified signal value for the edge region can provide a modified portion 430 of signal 401.

[0064] The modified signal value 430 can be converted into a thickness measurement result using a static formula (e.g., a correlation curve). For example, the controller 190 can use a neural network 500 to determine the thickness at the edge location of the substrate and one or more anchor locations. Alternatively, the controller 190 can directly use a static formula to generate thickness measurements for other regions (e.g., the inner region 205). In other words, signal values ​​from other regions (e.g., the inner region 205) can be converted into thickness values ​​without modification by the neural network.

[0065] In some embodiments, for a modified signal value corresponding to a given measurement location, the neural network 500 may be configured to use only input signal values ​​from measurement locations within a predetermined distance from this given location during the determination of the modified signal value. For example, if signal values ​​S1, S2, ..., S3 corresponding to measurement results at N sequential locations on path 210 are received... M ...S N Then for the Mth position (indicated by R) M The modified signal value S' M Only the signal value S can be used. M-L(min 1) ...S M ...S M+L(max N) To calculate the modified signal value S' M The value of L can be selected to generate a given modified signal value S' using measurements taken at a distance of approximately 2-4 mm. M ; Measurement results S can be used MThe measurement result is within approximately 1-2 mm (e.g., 1.5 mm) of the position. For example, L can be a number from 0 to 4 (e.g., 1 or 2). For example, if a measurement result within 3 mm is used, and the interval between the measurement results is 1 mm, then L can be 1; if the interval is 0.5 mm, then L can be 2; if the interval is 0.25 mm, then L can be 4. However, this can depend on the configuration of the polishing equipment and processing conditions. The modified signal value S' is calculated in the following steps. M Other parameter values ​​(such as pad wear rate) can still be used.

[0066] For example, one or more hidden layers 520 may contain a number of hidden nodes 570 (i.e., "hidden nodes 570"), equal in number to the number of signal input nodes 544, with each hidden node 570 corresponding to a specific signal input node 544. Each hidden node 570 may be disconnected from (or have zero parameter values ​​for) the input nodes 544 corresponding to measurement results for some locations, where these locations are greater than a predetermined distance from the measurement location of the corresponding input node. For example, the Mth hidden node may be disconnected from (or have zero parameter values ​​for) the 1st to (ML-1)th input nodes 544 and (M+L+1) to Nth input nodes. Similarly, each output node 560 may be disconnected from (or have zero parameter values ​​for) the hidden nodes 570 corresponding to modified signals for some locations, where these locations are greater than a predetermined distance from the measurement location of the output node. For example, the Mth output node can be disconnected from the 1st to (ML-1)th hidden nodes 570 and the (M+L+1)th to Nth hidden nodes (or have zero parameter values ​​for these hidden nodes).

[0067] In some embodiments, the polishing apparatus 100 may use a static formula to determine the thickness at multiple locations (e.g., locations within edge regions) of a first set of substrates. These substrates can be used to generate training data, which is then used to train a neural network. The polishing apparatus 100 may then use a neural network 500 to generate a modified signal, which is used to determine the thickness at multiple locations (e.g., locations within edge regions) of a second set of substrates. For example, the polishing apparatus 100 may apply a static formula to determine the thickness values ​​of the first set of substrates and use the trained neural network 500 to generate a modified signal, which is used to determine the thickness values ​​of the second set of substrates.

[0068] Figure 6 This is a flowchart of an example process 600 for polishing substrate 10. Process 600 can be performed by polishing equipment 100.

[0069] Polishing apparatus 100 polishes (602) layers on substrate 10 and monitors (604) the layers during polishing to generate measured signal values ​​for different locations on the layers. Locations on the layers may include one or more locations within edge regions 203 of the substrate (corresponding to regions 422 / 426 of signal 401) and one or more locations within anchor regions 202 of the substrate (corresponding to regions 421 / 423 of signal 401). Anchor regions 202 are spaced apart from substrate edges 204 and located within the central region 201 of the substrate, and are therefore unaffected by distortions caused by substrate edges 204. However, anchor regions 202 may be adjacent to edge regions 203. Anchor regions 202 may also surround the inner region 205 of the central region 201. The number of anchor locations may depend on the measurement point size and measurement frequency of the in-situ monitoring system 160. In some embodiments, the number of anchor locations may not exceed a maximum value (such as a maximum of 4).

[0070] The polishing apparatus 100 generates (606) an estimated measurement of the thickness at each of the different locations based on the measured signals at these locations. This involves processing the measured signals through a neural network 500.

[0071] The input to the neural network 500 can be the raw (or updated) measured signals generated by the in-situ monitoring system 160 for different locations. In some embodiments, the device 100 updates each measured signal by standardizing the signal values. This standardization can increase the probability that at least some of the inputs 504 to the neural network system 500 fall within a certain range, thereby improving the quality of neural network training and / or the accuracy of the inferences made by the neural network 500.

[0072] The output of the neural network 500 is a modified signal, each modified signal corresponding to a measured input signal. If the measured signal is a standardized value, then the modified signal corresponding to the measured signal will also be a standardized value. Therefore, before using such modified signals to estimate the thickness of the substrate, the polishing equipment 100 may need to convert the modified signals into unstandardized values.

[0073] The polishing equipment 100 detects (608) the polishing endpoint and / or modifies the polishing parameters based on each estimated thickness measurement.

[0074] Figure 7 This is a flowchart of an example process 700 for generating an estimated measurement result of thickness using a neural network 500. Process 700 can be performed by a polishing device 100.

[0075] The polishing apparatus 100 identifies (702) anchor positions in a set of locations on the substrate and acquires (704) measured signals for each location in the set of locations. In some embodiments, the anchor positions are spaced apart from the edges of the substrate.

[0076] Polishing equipment 100 updates the measured signals by standardizing (706) each measured signal (e.g., by dividing each measured signal by the measured signal at the anchor position) based on the measured signals at the anchor positions. Polishing equipment 100 then processes (708) the updated measured signals through neural network 500 to generate a modified signal for each standardized measured signal, and transforms (710) the modified signal into a non-standardized value using the measured signal at the anchor positions (e.g., by multiplying each measured signal by the measured signal at the anchor position) to update the measured signals. Polishing equipment 100 then uses (612) the non-standardized modified signal to generate an estimated measurement of the thickness at each of these locations for neural network 500.

[0077] Figure 8 This is a flowchart of an example process 800 for training a neural network 500 to generate a modified signal from a measured signal group. Process 800 can be executed by a system of one or more computers configured to train the neural network 500.

[0078] The system acquires (802) estimated thickness measurements generated by a neural network 500 based on input values, which include the measured signal at each of a set of locations on the substrate. The system also acquires (804) ground truth measures of the thickness at each of these locations. The ground truth measures of the thickness can be generated using electrical impedance measurement methods, such as the four-point probe method.

[0079] The system calculates (806) the measurement error between the estimated thickness measurement and the reference actual thickness measurement, and updates one or more parameters of the neural network 500 based on the measurement error. For this purpose, the system can utilize a training algorithm using gradient descent with backpropagation.

[0080] Monitoring systems can be used in various polishing systems. Polishing pads or carrier heads (or both) can be moved to provide relative movement between the polishing surface and the substrate. Polishing pads can be circular (or some other shape) pads fixed to a platform, tapes extending between a supply roller and a take-up roller, or continuous tapes. Polishing pads can be attached to a platform, incrementally advanced on a platform between polishing operations, or continuously driven on a platform during polishing. The pads can be fixed to the platform during polishing, or fluid can be carried between the platform and the polishing pad during polishing. Polishing pads can be standard (e.g., polyurethane with or without fillers) coarse pads, soft pads, or bonded abrasive pads.

[0081] While the above discussion focuses on eddy current monitoring systems, these correction techniques can be applied to other types of monitoring systems that scan substrate edges, such as optical monitoring systems. Furthermore, although the discussion focuses on polishing systems, these correction techniques can be applied to other types of substrate processing systems, such as deposition or etching systems that include in-situ monitoring of substrate edges.

[0082] Several embodiments of the present invention have been described. However, it should be understood that various modifications can be made without departing from the spirit and scope of the invention. Therefore, other embodiments exist that fall within the scope of the following claims.

Claims

1. A method for polishing a substrate, comprising: The layers on the substrate are polished at the polishing station; During polishing at the polishing station, the layer is monitored using an in-situ monitoring system to generate a first plurality of measured signals for a first plurality of different locations on the layer; By processing the first plurality of measured signals via a neural network, an estimated thickness measurement result is generated for each of the first plurality of different locations, wherein the estimated thickness measurement result for the location depends on the measured signal at the location and one or more measured signals at one or more other locations, the processing including The measured signals from the first plurality of different locations are input into the plurality of corresponding input nodes of the neural network, and Receive multiple outputs corresponding to the estimated measurement results of the thickness from multiple output nodes of the neural network; as well as At least one of the following: detect the polishing endpoint or modify polishing parameters based on each estimated thickness measurement.

2. The method of claim 1, further comprising: For a second set of different locations on the layer, a second set of measured signals are acquired; as well as By using a static formula, for each of the second plurality of different locations, an estimated measurement result of the thickness at the location is generated based on the measured signal at the location, the static formula relating multiple values ​​of the measured signal to multiple values ​​of the estimated measurement result of the thickness.

3. The method of claim 1, further comprising: For a layer on a second substrate different from the said substrate, a third plurality of measured signals are acquired, each of the third plurality of measured signals corresponding to a position among a third plurality of different positions on the layer on the second substrate; as well as By using a static formula, for each of the third plurality of different locations, an estimated measurement result of the thickness at that location is generated based on the measured signal at that location, the static formula relating multiple values ​​of the measured signal to multiple values ​​of the estimated measurement result of the thickness.

4. The method of claim 1, further comprising: For each of one or more different locations in the layer, obtain a reference true measurement result for the thickness; Calculate the measurement error between the estimated thickness measurement result for each location and the reference true thickness measurement result for the corresponding location; as well as The parameters of the neural network are updated based on the measurement error.

5. The method of claim 4, wherein the reference true measurement result of the thickness is determined based on the four-point probe method.

6. A polishing system, comprising: A carrier for holding a substrate; Support member, the support member being used for polishing the surface; An in-situ monitoring system having sensors configured to generate a first plurality of measured signals at a first plurality of different locations on a layer on the substrate; An electric motor, used to generate relative motion between the sensor and the substrate; as well as Controller, the controller is configured to Receive the first plurality of measured signals from the in-situ monitoring system. By processing the first plurality of measured signals via a neural network, an estimated thickness measurement result is generated for each of the first plurality of different locations, wherein the estimated thickness measurement result for the location depends on the measured signal at that location and one or more measured signals at one or more other locations, and the neural network processing includes... The measured signals from the first plurality of different locations are input into the plurality of corresponding input nodes of the neural network, and Receive multiple outputs corresponding to the estimated measurement results of the thickness from multiple output nodes of the neural network; as well as Based on each estimated thickness measurement, the polishing endpoint is detected, polishing parameters are modified, or both are performed.

7. The polishing system of claim 6, wherein the controller is further configured to: generate an estimated thickness measurement of the location for each of a second plurality of different locations on the layer based on the measured signal for the location using a static formula, the static formula relating a plurality of values ​​of the measured signal to a plurality of values ​​of the estimated thickness measurement.

8. The polishing system of claim 6, wherein the controller is further configured to: generate an estimated thickness measurement of a location based on a measured signal for each of a third plurality of different locations on a layer on a second substrate different from the substrate, using a static formula, wherein the static formula correlates a plurality of values ​​of the measured signal with a plurality of values ​​of the estimated thickness measurement.

9. The polishing system of claim 6, wherein the in-situ monitoring system includes an eddy current sensor.

10. The polishing system of claim 6, wherein the one or more other locations include anchor locations, and determining each estimated thickness measurement includes: Each measured signal is standardized based on the measured signal for the anchor position to update the measured signal.

11. The polishing system of claim 10, wherein each estimated thickness measurement is a standardized value, and the controller is further configured to: use the measured signal for the anchor position to convert each estimated thickness measurement into a non-standardized value to update the estimated thickness measurement.

12. A computer storage medium encoded with instructions that, when executed by one or more computers, cause the one or more computers to perform an operation comprising: During the polishing of a layer on a substrate at a polishing station, a first plurality of measured signals from an in-situ monitoring system are received for a first plurality of different locations on the layer; By processing the first plurality of measured signals via a neural network, an estimated thickness measurement result is generated for each of the first plurality of different locations, wherein the estimated thickness measurement result for the location depends on the measured signal at that location and one or more measured signals at one or more other locations, and the neural network processing includes... The measured signals from the first plurality of different locations are input into the plurality of corresponding input nodes of the neural network, and Receive multiple outputs corresponding to the estimated measurement results of the thickness from multiple output nodes of the neural network; as well as At least one of the following: detect the polishing endpoint or modify polishing parameters based on each estimated thickness measurement.

13. The computer storage medium of claim 12, wherein the operation further comprises: Receive a second plurality of measured signals for a second plurality of different locations on the layer; as well as By using a static formula, for each of the second plurality of different locations, an estimated measurement of the thickness at that location is generated based on the measured signal at that location, the static formula relating multiple values ​​of the measured signal to multiple values ​​of the estimated measurement of the thickness.

14. The computer storage medium of claim 12, wherein the operation further comprises: Receive a third plurality of measured signals for a layer on a second substrate different from the substrate, each of the third plurality of measured signals corresponding to a position among a third plurality of positions on the layer on the second substrate; as well as By using a static formula, for each of the third plurality of different locations, an estimated measurement of the thickness at that location is generated based on the measured signal at that location, the static formula relating multiple values ​​of the measured signal to multiple values ​​of the estimated measurement of the thickness.

15. The computer storage medium as claimed in claim 12, wherein: The neural network includes one or more neural network layers, the one or more neural network layers including an input layer, an output layer, and one or more hidden layers; Each neural network layer consists of one or more neural network nodes; as well as Each neural network node is configured to process the input according to a set of parameters in order to produce an output.

16. The computer storage medium of claim 15, wherein the input to the neural network nodes in the input layer comprises: The measurement results of the wear degree of the pads in the polishing station.

17. The computer storage medium of claim 13, wherein the one or more other locations include anchor locations, and wherein determining each estimated measurement of thickness comprises: Each measured signal is standardized based on the measured signal for the anchor position to update the measured signal.

18. The computer storage medium of claim 17, wherein each estimated thickness measurement is a standardized value, and wherein the operation further comprises: Using the measured signal for the anchor location, each estimated thickness measurement is converted into a non-standardized value to update the estimated thickness measurement.

19. The computer storage medium of claim 12, wherein the operation further comprises: Receive a reference real measurement of the thickness for each of one or more different locations of the layer; Calculate the measurement error between the estimated thickness measurement result for each location and the reference true thickness measurement result for the corresponding location; as well as The parameters of the neural network are updated based on the measurement error.

20. The computer storage medium of claim 19, wherein updating the parameters of the neural network based on the measurement result error comprises: The gradient of the measurement error is backpropagated through multiple layers of the neural network.

Images