TWI492005B - Goodness of fit in spectrographic monitoring of a substrate during processing - Google Patents

Description

The suitability of the substrate for spectral monitoring during processing

This description generally relates to the spectral monitoring of substrates during chemical mechanical polishing.

The integrated circuit is usually formed on the substrate by successively depositing a conductive layer, a semiconductive layer or an insulating layer on the germanium wafer. One manufacturing step involves depositing a filler layer on a non-planar surface and planarizing the filler layer. For a particular application, the filler layer is planarized until the top surface of the patterned layer is exposed. A layer of electrically conductive filler, for example, may be deposited on the patterned insulating layer to fill the trenches or holes in the insulating layer. After planarization, portions of the conductive layer remaining between the raised patterns of the insulating layer form vias, plugs, and wires that provide a conductive path between the thin film circuits on the substrate. For other applications, such as oxide milling, the filler layer is planarized until a predetermined thickness remains on the non-planar surface. In addition, planarization of the substrate surface is generally necessary for photolithography.

Chemical mechanical polishing (CMP) is a recognized method of planarization. This planarization method typically requires mounting the substrate on a carrier or polishing head. The exposed surface of the substrate is typically placed against a rotating abrasive disc pad or belt pad. The polishing pad can be a standard pad or a fixed abrasive pad. The standard pad has a durable rough surface, while the fixed polishing pad has abrasive particles held in the receiving medium. The carrier head provides a controllable load on the substrate to cause it to push the polishing pad. A grinding liquid, such as a slurry having abrasive particles, is typically supplied to the surface of the polishing pad.

One problem in CMP is to determine if the polishing process is complete, that is, whether the substrate layer has been flattened to the desired flatness or thickness, or when the desired amount of material has been removed. Excessive grinding (removal of the conductive layer or film) results in an increase in circuit resistance. On the other hand, insufficient polishing of the conductive layer (too little removal) results in an electrical short. For example, variations in slurry composition, polishing pad state, relative speed between the polishing pad and the substrate, and load on the substrate can result in variations in material removal rates between the substrates or between different regions of the substrate. Moreover, variations in the initial thickness of the substrate layer between the substrates or between different regions of the substrate can result in a change in the amount required to remove the target thickness. These changes result in changes in the time required to reach the end of the grinding. Therefore, the polishing end point cannot be reliably determined based only on the polishing time.

In a general aspect, a computer implemented method includes obtaining, by an in-situ optical monitoring system, a sequence of current spectra from which each current spectrum of the current spectrum of the sequence is a spectrum of light reflected from a substrate having An outermost layer subjected to grinding and at least one underlying layer; comparing each of the current spectra with a plurality of reference spectra from the first reference spectral library and determining a first best matching reference spectrum to produce a first sequence first best Matching a reference spectrum; determining a first fitness for the first sequence; and determining a polishing endpoint based on the first sequence and the first fitness.

In another general aspect, a computer implemented method includes obtaining a sequence of current spectra from an in situ optical monitoring system, each current spectrum from the current spectrum of the sequence being a spectrum of light reflected from a substrate, the substrate Having an outermost layer subjected to grinding and at least one underlying layer; comparing each of the current spectra with a plurality of reference spectra from a plurality of reference spectral libraries; determining which library provides the best fit to the current spectrum of the sequence; and based on the sequence The current spectrum and the library that provides the best fit for the current spectrum of the sequence determine a polishing endpoint.

Implementation of either of these two methods can include one or more of the following features. Comparing the current spectrum with a plurality of reference spectra from a second reference spectral library and determining a second best matching reference spectrum to generate a second sequence of second best matching reference spectra, and determining the second a second fitness of the sequence, and the endpoint may be determined based on the first sequence, the second sequence, the first fitness, and the second fitness. Determining the polishing end point can include determining whether the first best matching reference spectrum indicates an end point, and if so, determining whether the first fitness level is better than the second fitness level, and if so, terminating the end point. Determining the polishing end point can include determining whether the second best matching reference spectrum indicates an end point, and if so, determining whether the second best fit is better than the first fitness level, and if so, terminating the end point. Determining a first index value for each of the first best matching reference spectra to generate a sequence of first index values, and determining the second fitness level can include determining one of each of the second best matching reference spectra The two index values are used to generate a sequence of second index values. The first function can fit the first index value of the sequence, and the second function can fit the second index value of the sequence. The first function and the second function can be linear functions. Determining the first fitness may include determining a fitness of the first index value of the sequence for the first function, and determining the second fitness may include determining a fitness of the second index value of the sequence for the second function. Determining the first fitness level may include determining a sum of squares of differences between the first index value of the sequence and the first function, and determining the second fitness level may include determining a second index value of the sequence and the second function The sum of the squared differences. The sequence first index value may form a first index trace, and wherein the sequence second index value may form a second index trace. Determining whether the first best matching reference spectrum indicates an end point may include determining whether an index of the first best matching reference spectrum is a target index. Determining whether the second best matching reference spectrum indicates an end point may include determining whether an index of the second best matching reference spectrum is a target index. It may be determined whether the first fitness for the first sequence is better than the second fitness for the second sequence. The plurality of reference spectra from the first reference spectral library may represent a substrate having a volt layer under the first thickness, and the plurality of reference spectra from the second reference spectral library may represent substrates having volt layers under different second thicknesses. Determining a first best matching reference spectrum can include determining which reference spectrum from the first reference spectral library has a minimum difference from the current spectrum, and determining the second best matching reference spectrum can include determining which of the second reference spectral libraries The reference spectrum has the smallest difference from the current spectrum.

In general, another aspect of the subject matter described in this specification can be embodied in a computer implemented method comprising receiving a first sequence of reflected light from a first region of a substrate. spectrum. A second sequence of reflected light may be received from a second region of the substrate. Each of the current spectra from the current spectrum of the first sequence and a plurality of reference spectra from a first reference spectral library can be compared to produce a first sequence of best matching reference spectra. Each current spectrum from the current spectrum of the second sequence can be compared to a plurality of reference spectra from a second reference spectrum library to produce a second sequence of best matching reference spectra. The second reference spectral library can be different from the first reference spectral library. Other embodiments of this aspect include corresponding systems, devices, and computer program products.

These and other embodiments may include one or more of the following features as appropriate. The first reference spectral library and the second reference spectral library can be predetermined.

In another aspect, a computer program product embodied in a computer readable medium can be operated to cause a data processing device to perform operations comprising the steps of the above methods.

In general, an aspect of the subject matter described in this specification can be embodied in a computer implemented method that includes receiving a first sequence of reflected light from a first region of a substrate. . A second sequence of reflected light may be received from a second region of the substrate. Each current spectrum from the current spectrum of the first sequence can be compared to a first plurality of reference spectra from a first plurality of reference spectral libraries to produce a plurality of first sequence best matching reference spectra. Each current spectrum from the current spectrum of the second sequence can be compared to a second plurality of reference spectra from a second plurality of reference spectral libraries to produce a plurality of second sequence best matching reference spectra. A plurality of first fitness levels may be determined for the plurality of first sequences to best match the reference spectrum. A plurality of second fitness levels may be determined for the plurality of second sequences to best match the reference spectrum. Other embodiments of this aspect include corresponding systems, devices, and computer program products.

These and other embodiments may include one or more of the following features as appropriate. Some of the first plurality of reference spectral libraries may be identical to the second plurality of reference spectral libraries. All of the first plurality of reference spectral libraries are identical to the second plurality of reference spectral libraries. None of the first plurality of reference spectral libraries may be identical to the second plurality of reference spectral libraries.

Generating a plurality of first sequence best matching reference spectra may include comparing each of the current spectra from the current spectrum of the first sequence with a plurality of reference spectra from the first reference spectral library and determining a first intermediate fitness; comparing from the first Comparing each current spectrum of the current spectrum of the sequence with a plurality of reference spectra from the second reference spectrum library and determining a second intermediate fitness; comparing the first intermediate fitness with the second intermediate fitness; based on the first intermediate fit First selecting one of the first reference spectral library or the second reference spectral library by comparing the second intermediate fitness; and determining a first sequence best matching reference spectrum based on the first selection.

Generating a plurality of second sequence best matching reference spectra can include comparing each current spectrum from the current spectrum of the second sequence to a plurality of reference spectra from the first reference spectral library and determining a third intermediate fitness; comparing from the second Comparing each current spectrum of the current spectrum of the sequence with a plurality of reference spectra from the second reference spectrum library and determining a fourth intermediate fitness; comparing the third intermediate fitness with the fourth intermediate fitness; based on the third intermediate fit Selecting one of the first reference spectral library or the second reference spectral library by a second comparison with the fourth intermediate fitness; and determining a second sequence best matching reference spectrum based on the second selection.

The first selection and the second selection may be determined within a predetermined period of grinding. The predetermined period of grinding may include the first twenty seconds of grinding. The method can further include determining a first polishing endpoint of the first region based on the first sequence best matching reference spectrum and a corresponding first fitness; and determining the second region based on the second sequence best matching reference spectrum and the corresponding second fitness The second grinding end point.

As used in this specification, the term "substrate" can include, for example, a product substrate (eg, which can include multiple memory or processor dies), a test substrate, a bare substrate, and a gated substrate. The substrate can be at various stages of integrated circuit fabrication, for example, the substrate can be a bare wafer, or it can include one or more deposited and/or patterned layers. The term "substrate" can include discs and rectangular sheets.

Possible advantages of embodiments of the invention may include the following. The endpoint detection system can be less sensitive to changes between the substrate or the underlying layer or regions of the substrate in the pattern, and thus can improve the endpoint system for detecting the desired endpoint of the substrate for each substrate or for each region of the substrate. reliability. Therefore, the inter-wafer and in-wafer thickness uniformity can be improved.

The details of one or more embodiments of the invention are set forth in the drawings and the claims Other features, aspects, and advantages of the invention will be apparent from the embodiments, drawings and claims.

During polishing, substrates having different patterns and different underlying layer thicknesses may pass through the polishing apparatus. If spectral monitoring is being used, these changes can result in unreliable identification of the measured spectra from the spectral library to the measured spectra.

To compensate for this, multiple libraries can be used, with different libraries representing different patterns and different underlying layer thicknesses. Next, a sequence of measured spectra is compared to reference spectra from the plurality of banks, and a library is provided that provides the best fit for the sequence, and the best fit library can be used for polishing rate or endpoint determination.

Additionally, the substrate, particularly the element substrate, can have different regions with different features (eg, different feature density or underlying layer thickness). Thus, during spectral monitoring performed in-situ during milling, the metrology spectra of some regions may not be reliably matched to reference spectra established based on data from other regions.

This problem can be solved by using multiple banks representing different regions within the substrate. A sequence of reflected light sequences can be measured for each of a plurality of regions of the substrate, and spectra from the sequences and reference spectra from different spectral libraries can be compared for different regions to produce the most useful endpoint determination Good match reference spectrum.

Referring to FIG. 1, a substrate 10 can include a wafer 12, an outermost layer 14 to be subjected to grinding, and one or more underlying layers 16 between the outermost layer 14 and the wafer 12, in the underlying layer 16. Some are usually patterned. For example, the outermost layer 14 and a immediately adjacent underlying layer 16 can both be dielectric. For example, the outermost layer 14 can be an oxide and the adjacent underlying layer 16 can be a nitride. Other layers, such as other conductive layers and dielectric layers, may be formed between the immediately adjacent underlying layer and the substrate.

A potential problem in the detection of spectral endpoints during chemical mechanical polishing, in particular where the outermost layer 14 and the underlying layer 16 are both dielectric, is that the thickness of the underlying layer can be from different substrates or substrates. Change in different areas. A substrate may have a plurality of regions, such as a central region, an intermediate region, and an edge region. For example, on a 300 mm wafer, the central region can extend from the center to a radius of 50 mm, the intermediate region can extend from a radius of 50 mm to about 100 mm and the edge can extend from about 100 mm to about 150 mm. In some embodiments, the substrate has more or fewer regions than the three regions mentioned.

Thus, a substrate having the same thickness on the outermost layer (or a region on the substrate) can actually reflect different spectra depending on the underlying layer. Thus, for example, if the underlying layers have different thicknesses, the target spectrum used to trigger the polishing endpoints of some of the substrates (or regions of the substrate) may not function properly for other substrates (or regions of the substrate). However, it is possible to compensate for this effect by comparing the spectrum obtained during grinding with a plurality of spectra representing changes in the underlying layer.

Due to changes other than the thickness of the underlying layer, such as a change in the initial thickness of the outermost layer subjected to grinding, a change in the thickness of the outermost layer during grinding (eg, due to different polishing rates in each region), Variations in the optical properties of the environment, changes in the pattern of the underlying layer (eg, line width (eg, metal or polysilicon line width)), or changes in the composition of the layer, may also inherently exist in the use of a substrate relative to another A substrate or a region of the substrate is between the reference spectra determined relative to the other region. However, it is equally possible to compensate for this effect by comparing the spectrum obtained during grinding with a plurality of spectra, which represent other variations between the substrates.

In addition, multiple reference spectral libraries may be used to compensate for variations. Within each bank are a plurality of reference spectra representing a substrate (or region) having a thickness variation of the outermost layer but having other similar features (eg, similar underlying layer thickness). Between the banks, other variations (such as variations in the thickness of the underlying layer) may be indicated, for example, different libraries include reference spectra representing substrates (or regions) having different underlying layer thicknesses.

2 is a cross-sectional view showing an example of a polishing apparatus 20 operable to polish a substrate 10. The polishing apparatus 20 includes a rotatable disc-shaped platform 24 on which a polishing pad 30 is located. The platform is operable to rotate about an axis 25. For example, a motor can rotate a drive shaft 22 to rotate the platform 24.

An optical inlet 36 is provided through one of the pads by including a perforation (i.e., through a hole in the polishing pad) or a solid window. Although in some embodiments the solid window can be supported on the platform 24 and protrude into one of the perforations in the polishing pad, the solid window can be secured to the polishing pad. The polishing pad 30 is typically placed on the platform 24 such that the perforation or window covers an optical head 53 located in a recess 26 in the platform 24. Thus, the optical head 53 has an optical entrance through the perforation or window to a substrate being ground. The optical head is further described below.

Grinding device 20 includes a combined slurry/rinsing arm 39. During grinding, the arm 39 can be operated to dispense a grinding fluid 38, such as a slurry. Alternatively, the grinding apparatus includes a grouting port operable to dispense the slurry onto the polishing pad 30.

The polishing apparatus 20 includes a carrier head 70 that is operable to secure the substrate 10 against the polishing pad 30. The carrier head 70 is suspended from a support structure 72 (e.g., a rotating rack) and coupled to a carrier head rotation motor 76 by a carrier drive shaft 74 such that the carrier head can rotate about a shaft 71. Additionally, the carrier head 70 can be oscillated laterally in a radial slot formed in the support structure 72. In operation, the platform is rotated about its central axis 25 and the carrier is rotated about its central axis 71 and laterally displaced across the top surface of the polishing pad.

As discussed below, the polishing apparatus also includes an optical monitoring system that can be used to determine the end of the grind, or whether to adjust the grind rate (or the adjustment of the grind rate). The optical monitoring system includes a light source 51 and a light detector 52. Light passes from the source 51 through the optical inlet 36 in the polishing pad 30, impinges and is reflected back from the substrate 10 via the optical inlet 36 and travels to the photodetector 52.

A bifurcated optical cable 54 can be used to transfer light from the source 51 to the optical inlet 36 and back to the photodetector 52 from the optical inlet 36. The bifurcated optical cable 54 can include a "trunk" 55 and two "branches" 56 and 58.

As mentioned above, the platform 24 includes a recess 26 in which the optical head 53 is located. The optical head 53 holds one end of the mains 55 of the bifurcated optical cable 54, which is configured to deliver light to and from the surface of the substrate being polished. The optical head 53 can include one or more lenses or windows that overlie the end of the bifurcated optical cable 54. Alternatively, the optical head 53 can only secure the end of the trunk 55 adjacent the solid window in the polishing pad.

The platform includes a removable in-situ monitoring module 50. The home position monitoring module 50 can include one or more of the following components: a light source 51, a light detector 52, and circuitry for transmitting signals to and receiving signals from the light source 51. For example, the output of the detector 52 can be a digital electronic signal that passes through one of the rotational shaft couplers (eg, a slip ring) of the drive shaft 22 to a controller 60 (such as a computer) of the optical monitoring system. Similarly, the light source can be turned "on" or "off" in response to a control command from the controller through the rotary coupler to the digital electronic signal of module 50.

The in-situ monitoring module can also secure the respective ends of the branch portions 56 and 58 of the bifurcated optical cable 54. The light source is operable to transmit light that is conveyed via the branches 56 and exits the end of the mains 55 located in the optical head 53 and impinges on the substrate being ground. Light reflected from the substrate is received at the end of the main line 55 in the optical head 53 and delivered to the photodetector 52 via the branch 58.

Light source 51 is operable to emit white light. In an embodiment, the emitted white light comprises light having a wavelength of from 200 to 800 nanometers. Suitable light source for xenon or mercury xenon lamps.

The photodetector 52 can be a spectrometer. A spectrometer is basically an optical instrument for measuring the intensity of light on a portion of an electromagnetic spectrum. Suitable for the spectrometer as a grating spectrometer. A typical output of a spectrometer is the intensity of light as a function of wavelength (or frequency).

Light source 51 and photodetector 52 are coupled to a computing component (e.g., controller 60) that is operable to control its operation and receive its signals. The computing component can include a microprocessor located adjacent to the polishing device, such as a personal computer. In terms of control, the computing component can, for example, synchronize the activation of light source 51 with the rotation of platform 24.

As shown in FIG. 3, when the platform is rotated, the computer can cause the light source 51 to start just before the substrate 10 passes over the home monitoring module and end up emitting a series of flashes just after the substrate 10 has passed the home monitoring module (depicted Each of the points 301-311 represents the location of the light from the in-situ monitoring module that is struck and reflected. Alternatively, the computer may cause the light source 51 to begin just before the substrate 10 passes over the home monitoring module and end the continuous emission of light just after the substrate 10 has passed the home monitoring module. In either case, the signal from the detector can be integrated during the sampling period to produce a spectral measurement at the sampling frequency. The sampling frequency can be from about 3 to 100 milliseconds. Although not shown, each time the substrate 10 passes over the monitoring module, the alignment of the substrate with the monitoring module can be different from the previous one. The spectra are obtained from different radii on the substrate via one rotation of the platform. That is, some spectra are obtained from a position closer to the center of the substrate and some closer to the edge. In addition, a sequence of spectra can be obtained over time via multiple rotations of the platform.

In operation, the computing device can receive, for example, a signal carrying information describing the spectrum of light received by the photodetector 52 for a particular flash or detector of the light source. Therefore, this spectrum is the spectrum measured in situ during grinding.

Without being limited to any particular theory, due to variations in the thickness of the outermost layer, the spectrum of light reflected from substrate 10 progresses as the milling progresses, thereby producing a sequence of spectra that change over time. In addition, a particular spectrum is presented by a particular layer stack thickness.

The computing component can process the signal to determine the end of the grinding step. In particular, the computing component can perform a logical determination based on the measurement spectrum when the end point has been reached.

In short, the computing component can compare the measured spectrum to a plurality of reference spectra and can use the comparison to determine when the endpoint has been reached.

As used herein, a reference spectrum is a predetermined spectrum that is produced prior to polishing a substrate. The reference spectrum can have a predetermined correlation with the value of the substrate properties, such as the thickness of the outermost layer (ie, defined prior to the grinding operation). In other or additional instances, the reference spectrum may have a predetermined association with a value indicative of the time in the polishing process, which is expected to occur at this time assuming the actual polishing rate follows the expected polishing rate.

The reference spectrum can be empirically generated, for example, by measuring the spectrum from a test substrate having a known layer thickness, or by theoretically generating a reference spectrum. For example, to determine the reference spectrum, the spectrum of the "configuration" substrate having the same pattern as the product substrate can be pre-ground at the measurement station. The substrate properties (eg, the thickness of the outermost layer) can also be measured by pre-grinding the same measurement table or different measurement tables. Next, the placement substrate is polished while collecting the spectrum. For each spectrum, a value is recorded indicating the time at which the spectrum was collected during the polishing process. For example, the value can be an elapsed time, or the number of platform rotations. The substrate can be overgrinded (i.e., ground beyond the desired thickness) such that the spectrum of light reflected from the substrate can be obtained when the target thickness is achieved. Next, the spectrum and properties of the configuration substrate (eg, the thickness of the outermost layer) can be measured by post-grinding at the measurement station.

The configuration substrate can be periodically removed from the grinding system as appropriate and its properties and/or spectra measured at the measurement station prior to returning to grinding. A value indicating the time at which the spectrum is measured at the measuring station in the polishing process can also be recorded.

The reference spectrum can be stored in a library. The reference spectrum in this library represents a substrate having a plurality of different outer layer thicknesses.

Multiple banks may be built from different configuration substrates that differ in features other than the thickness of the outermost layer (eg, in the underlying layer thickness, the underlying layer pattern, or the outer or underlying layer composition).

The measured thickness and the collected spectrum are used to select one or more spectra from the collected spectra that are determined to be present by the substrate when the substrate has an associated thickness. In particular, linear interpolation can be performed using the measured pre-ground film thickness and post-grinding substrate thickness (or other thickness measured at the measurement station) to determine the time and corresponding spectrum exhibited when the target thickness is achieved. The or the spectra that are determined to be present when the target thickness is achieved are designated as the or the target spectra.

In addition, assuming a uniform polishing rate, the measured pre-polished film thickness and post-grinding substrate thickness (or other thickness measured at the measuring station) can be used based on the time at which the spectrum is collected and the time input of the measured spectrum. Linear interpolation between ) calculates the thickness of the outermost layer for each spectrum collected in situ. Due to the initial planarization, the polishing rate may not be uniform from the beginning to the end of the grinding operation; in this case, if the progress of the polishing rate is known, the thickness can still be calculated. In addition, it can be assumed that the rate will likely be uniform for the end of the grinding.

In addition to being judged on an empirical basis, some or all of the reference spectra may be calculated theoretically, for example, using an optical model of the substrate layer. For example, and an optical model can be used to calculate the spectrum for a given outer layer thickness D. The value indicative of the time at which the spectrum will be collected in the polishing process can be calculated, for example, by assuming that the outer layer is removed at a uniform polishing rate. For example, the time Ts (Ts = (D0 - D) / R) of a specific spectrum can be simply calculated by assuming the initial thickness D0 and the uniform polishing rate R. As another example, linear interpolation between the measured time T1, T2 based on the pre-polished thickness D1 and the post-grinding thickness D2 (or other thickness measured at the measuring station) for the thickness D of the optical model can be performed. (Ts=T2-T1*(D1-D)/(D1-D2)).

As used herein, a reference spectral library is a collection of reference spectra representing a substrate that shares common properties (other than the thickness of the layer). However, the common properties shared in a single library can vary across multiple reference spectral libraries. For example, two different banks may include a reference spectrum representing a substrate having two different underlying layer thicknesses.

As discussed above, spectra of different banks can be generated by grinding a plurality of "configuration" substrates having different substrate properties (eg, underlying layer thickness or layer composition) and collecting spectra; spectra from a configuration substrate can provide a The first library and the spectrum from another substrate having a different underlying layer thickness provides a second bank. In other or additional cases, the reference spectra of the different banks may be calculated according to theory. For example, the optical spectrum of the first library may be used to calculate the spectrum of the first library, and may be used with a different thickness. The optical model of the layer calculates the spectrum of the second library.

In some embodiments, each reference spectrum is assigned an index value. This index can be a value indicating the time at which the reference spectrum is expected to be observed in the polishing process. The spectra can be indexed such that each spectrum in a particular library has a unique index value. Indexing can be performed to rank the index values in the order in which the spectra are measured. The index value can be selected to vary monotonically with the grinding process, for example, increasing or decreasing. In particular, the index value of the reference spectrum can be selected such that it forms a linear function of the number of time or platform rotations. For example, the index value can be proportional to the number of platform rotations. Thus, each index number can be an integer, and the index number can represent the expected platform rotation at which the associated spectrum will occur.

The reference spectrum and its associated index can be stored in a library. The bank can be implemented in the memory of the computing elements of the polishing apparatus. The index of the target spectrum can be specified as the target index.

During the grinding, an index trace can be generated for each bank. Each index trace includes a sequence index that forms one of the traces, each particular index of the sequence being associated with a particular metrology spectrum. For an index trace of a given library, a particular index in the sequence is generated by selecting an index of the reference spectrum from a given library that most closely matches the particular measurement spectrum.

As shown in FIG. 4, the index 80 corresponding to each measurement spectrum can be rotated according to time or platform rotation. For example, a polynomial function of known order (eg, a first order function (ie, line)) is fitted to the index number of the survey using a robust line fit. The place where the online and target index meets defines the end time or rotation. For example, the first order function 82 is fitted to the data points shown in FIG.

Without being limited to any particular theory, some libraries can predict the appropriate endpoint more accurately than other libraries because they more closely match the measured data. For example, among the plurality of banks representing substrates having different underlying layer thicknesses (or regions within the substrate), the library that most closely matches the thickness of the underlying layer of the measured substrate (or region within the substrate) The best match should be provided. Thus, the benefit of the present invention is a more accurate endpoint detection system achieved by utilizing multiple reference spectral libraries. In particular, different reference spectral libraries can be used for each region of the substrate. In addition, each region can have multiple libraries of different reference spectra.

For example, Figure 4 is a diagram illustrating an example index trace from a spectral monitoring system showing a good data fit corresponding to a first region on a substrate. In contrast, FIG. 5 is a diagram illustrating an example index trace from a spectral monitoring system showing weaker data adhesion to a second region on the substrate. The example index traces of Figures 4 and 5 represent index traces generated using the same reference spectral library. In contrast, the index number of the mapping in Figure 5 has a larger amount of difference from the associated line of weakness than the difference between the index line associated with the index number of the mapping in Figure 4; . Therefore, it may be advantageous to use different reference spectral libraries for different regions on the substrate.

In some embodiments, the spectra are obtained at more than one radial location of the substrate. For each spectral measurement, the radial position on the substrate can be determined, and the spectral measurements can be placed in a region based on its radial position (eg, a radial region). As described above, a substrate can have a plurality of regions, such as a central region, an intermediate region, and an edge region. The position of the obtained spectrum can be determined, for example, by the method described in U.S. Patent No. 7,097,537, or U.S. Patent No. 7,018,271, the disclosure of which is incorporated herein by reference.

As described above, the measured spectra from each region (or the average of the spectra from the region obtained from a single scan of the sensor across the substrate for each region) and a plurality of reference spectra are compared A reference spectrum in one or more of the libraries, and a comparison with the spectral library determines the corresponding index number. A corresponding index number for each region can be used to generate an index trace, and the index trace can be used to determine fitness.

Figure 6 shows a method 600 for determining the end of a grinding step. The substrate from one of the batches of substrates is ground (step 602) and the following steps are performed for each platform rotation (and for each region of the substrate where multiple banks are being used for each region). One or more spectra are measured to obtain a current spectrum for a current platform rotation (step 604). It is determined that a first best matching reference spectrum stored in a first spectral library is best suited to the current spectrum (step 606). It is determined that a second best matching reference spectrum stored in a second spectral library is best suited to the one of the current spectra (step 608). More specifically, for each bank that is being used for the substrate and/or region, a reference spectrum that best matches the current spectrum is determined. Determining an index of the first best matching reference spectrum from the first library that is best suited for the current spectrum (step 610) and adding it to one of the first index traces associated with the first library (step 612). An index from the second library that best fits the second best matching reference spectrum for the current spectrum is determined (step 614) and added to the second index trace associated with the second library (step 616) ). More specifically, for each library, an index of each best matching reference spectrum is determined and added to the index trace of the associated library. The first line is fitted to the first index trace (step 620) and the second line is fitted to the second index trace (step 622). More broadly, for each index trace, a line can be fitted to the index trace. These lines can be fitted using a robust linear fit.

The index of the first best matching spectrum may match or exceed the target index (step 624) and the index trace associated with the first spectral library may have an optimum fit for the line of weakness associated with the first spectral library. Degree (step 626), or the index of the second best matching spectrum matches or exceeds the target index (step 624) and the index trace associated with the second spectral library has an association with the second spectral library The best fit of the line of weakness (step 626) is referred to as the end point (step 630). Larger, it can be referred to as the end point when the index trace that best fits its associated fit line matches or exceeds the target index.

Although the two libraries are discussed above, this technique can be used by three or more libraries. In addition, some, all, or none of the libraries may be shared between regions, for example, some, all, or none of the libraries for one region may be used for another region.

Also, instead of comparing the index value itself with the target index, the value of the close line and the target index at the current time can be compared. That is, the self-linear function calculates the value for the current time (in this case, it does not need to be an integer) and compares this value with the target index.

Using method 600, for example, different reference spectral libraries can be used to determine the polishing endpoints for different regions of the substrate. In particular, a reference spectral library that produces index traces with the best fit for a particular region is used. In these and other embodiments, some regions may use the same reference spectral library, while some regions may use different reference spectral libraries. In some embodiments, a subset of the plurality of reference spectral libraries may be predetermined (eg, selected by the user) to limit the number of banks for each region. For example, two or more reference libraries may be predetermined for use in each region. In some embodiments, a particular reference spectral library can be identified for each region based on fitness. For example, during a predetermined period of time during the polishing process (eg, the first 10-20 seconds of grinding), a particular reference to be used for each region may be selected based on a reference spectral library that produces the best fit during the predetermined time period. Spectral library (for example, the best library for a region).

Other embodiments are possible. For example, although two libraries are discussed above, this technique can be used with three or more libraries. As another example, some, all, or none of the libraries may be shared between regions, for example, some, all, or none of the libraries for one region may be used for another region. As yet another example, exactly one reference spectral library can be predetermined for use in each region such that different reference spectral libraries are used for each region. For this embodiment, the reference library is not selected based on fitness; instead, the reliability of the endpoint can be improved simply by using different reference banks of different regions.

Determining whether the index trace associated with the spectral library has the best fitness for the linear function associated with the library may include (relatively, the difference between the index trace associated with the associated strain line and another library) For example, it is determined whether the index trace of the associated spectral library has a difference from the minimum amount of the associated difference line, for example, the lowest standard deviation, the maximum correlation, or other variation. In one embodiment, the fitness is determined by calculating the sum of the squared differences between the index data points and the linear function; the library with the lowest sum of squares has the best fit.

If one of the index traces reaches the target index but is not optimally fit, the system can wait until the index trace is optimally fit, or the index trace reaches the target index as best suited.

Although only two libraries and two index traces are discussed above, this concept can be applied to more than two libraries that will provide more than two index traces. In addition, instead of being referred to as the end point when the index of the trace matches the target index, the time that can be calculated by crossing the target index with respect to the line of the trace is referred to as the end point. In addition, it will be possible to reject the index trace with the worst fit before the end point (eg, after about 40% to 50% or 60% of the expected grinding time) in order to reduce processing.

Obtaining the current spectrum can include measuring at least one spectrum of light that reflects the surface of the substrate being polished (step 604). Multiple spectra may be measured, as appropriate, for example, from a single rotation of the platform, for example, at points 301-311 (Fig. 3) to obtain spectra measured at different radii on the substrate. If multiple spectra are measured, a subset of one or more of the spectra can be selected for use in the endpoint detection algorithm. For example, the spectra measured at sample locations near the center of the substrate (eg, at points 305, 306, and 307 shown in FIG. 3) can be selected. The spectrum measured during the current platform rotation is treated as appropriate to enhance accuracy and/or accuracy.

Determining the difference between each of the selected measured spectra and each of the reference spectra (step 606 or 610) may include calculating the difference as the sum of the intensity differences over a range of wavelengths. that is,

Wherein a and b are lower and upper limits of the wavelength range of the spectrum, and _{when the current} I ([lambda]) and the _reference I ([lambda]) for a given wavelength, respectively, of the current intensity and the spectral intensity of the reference spectrum. Alternatively, the difference can be calculated as a mean square error, ie:

Wherein for each region of the substrate or for the substrate, there are a plurality of current spectra that are rotated from a given platform, and a best match can be determined between each of the current spectra and each of the reference spectra of a given library. . Each selected current spectrum is compared to each reference spectrum. For example, assuming that the current spectra e , f, and g and the reference spectra E , F, and G , the matching coefficients can be calculated for each of the following combinations of the current spectrum and the reference spectrum: e and E , e and F , e and G , f and E , f and F , f and G , g and E , g and F, and g and G. Regardless of which matching coefficient indicates the best match (eg, is the smallest), the reference spectrum is determined, and thus the index is determined.

Determining whether the index trace associated with the spectral library has the best fit for the line of weakness associated with the spectral library (step 620 or 624) can include determining which library has the data point containing the index trace and is suitable for The sum of the squared differences between the lines of resistance associated with the spectral library. For example, the sum of the squared differences between the data points represented in Figures 4 and 5 and their respective associated resistance lines.

In some embodiments, the expected end time of an area, such as a central area, is determined. The grinding rate in the other regions is then adjusted as appropriate to achieve its desired end point at the same time as the expected end time for the selected region (e.g., the central region). The polishing rate can be adjusted, such as by increasing or decreasing the pressure in the corresponding region in the carrier head. In some carrier heads, such as the carrier heads described in U.S. Publication No. 2005-0211377, the carrier head has an adjustable pressure zone. It can be assumed that the change in the polishing rate is proportional to the change in pressure (eg, a simple Prestonian model). Additionally, a control model for grinding the substrate can be developed that takes into account the effects of the plate or head rotational speed, the secondary effects of different head pressure combinations, the grinding temperature, the slurry flow rate, or other parameters that affect the polishing rate.

Referring to Figure 7, if a particular profile is required (such as a uniform thickness across the surface of the substrate), the slope of the polishing rate can be monitored (as indicated by the change in index number over time) and if the index trace fit indicates the spectrum The measurement is reliable (for example, the fitness is less than the predetermined threshold) and the polishing rate is adjusted. After the grinding stabilization period 705, a spectrum is obtained at the central region 710, at the edge region 715, and in the middle of the intermediate region 720. Here, the regions are circular or annular regions. Each spectrum is related to its respective index. This process is repeated via a number of platforms, or repeated over time, and the rate of polishing at each of central region 710, intermediate region 720, and edge region 715 is determined. The polishing rate is indicated by the slope of the line obtained by mapping the index 730 (y-axis) according to the number of rotations 735 (x-axis). If any of the rates is calculated to be faster or slower than the other rates, the rate in the region can be adjusted if the index trace fitness indicates that the spectral measurements are reliable. Here, the adjustment is based on the end point C _{E of the} central region 710. For some embodiments, if the polishing rate is within an acceptable range, no adjustment is needed. A similar grinding end point EDP is known from grinding a similar substrate having similar grinding parameters or from using the differential method described above. T ₁ during the polishing process of the first polishing time, the polishing rate of the intermediate region 720 decrease the polishing rate increases and the edge region. Without adjusting the polishing rate at the intermediate region 720, the intermediate region will be ground faster than the rest of the substrate, and the intermediate region will be ground at an excessive polishing rate of M _A . Without adjusting the polishing rate of 715 ₁ T at the edge region, the edge region will be less than the polishing rate of 715 E _u.

At a subsequent time (T ₂ ) during the grinding process, the rate can be adjusted again as appropriate. The goal in this polishing process is to finish the polishing when the substrate has a flat surface or an oxide layer that spans a relatively flat surface. One method of determining the amount to adjust the polishing rate is to adjust the rate such that the index of each of the center region, the intermediate region, and the edge region is equal at the approximate polishing end point EDP. Therefore, the polishing rate at the edge region needs to be adjusted, and the central region and the intermediate region are ground at the same rate as before T ₂ . If the EDP is an approximation, the grinding can be terminated when the index at each region is at the desired position, i.e., at the same index for each location.

During the polishing process, it is preferred to only change the polishing rate a few times, such as four, three, two or only once. Adjustments can be made near the beginning of the grinding process, in the middle or towards the end. Associating the spectrum with the index number establishes a linear comparison of the grinding at each of the regions and simplifies the calculations necessary to determine how to control the polishing process and eliminates complex software or processing steps.

A method can be applied during the endpoint process to limit the portion of the library that is searched for matching spectra. The library typically includes a broader spectrum of spectra than would be obtained when the substrate was polished. This broader range illustrates the spectrum obtained from the thicker starting outermost layer and the spectrum obtained after overgrinding. The library search is limited to a library spectrum of a predetermined range during substrate polishing. In some embodiments, the current rotation index N of the substrate being ground is determined. N can be determined by searching all library spectra. For spectra obtained during subsequent rotations, the library can be searched for in the range of degrees of freedom of N. That is, if the index number is considered to be N during a rotation, then during the subsequent rotation for a later X rotation, where the degree of freedom is Y, it will be from (N+X)-Y to (N+X)+ Y searches for this range. For example, if the matching index is considered to be 8 and the degree of freedom is selected to be 5 under the first grinding rotation of the substrate, only the spectrum corresponding to the index number 9±5 is checked for the spectrum obtained during the second rotation. Used for matching. When applying this method, the same method can be applied independently to all the libraries currently in use in the endpoint detection process.

Embodiments of the invention and all of the functional operations described in this specification can be performed on digital electronic circuits or computer software, firmware or hardware (including structural members and structural equivalents thereof disclosed in this specification) or Implemented in the portfolio. Embodiments of the present invention may be implemented as one or more computer program products, that is, one or more computer programs embodied in an information carrier (eg, in a machine readable storage medium) for use by a data processing device (For example, a programmable processor, a computer or a plurality of processors or computers) performs or controls the operation of the data processing apparatus. Any form of programming language (including compiled or interpreted language) can be written into a computer program (also known as a program, software, software application or code) and can be deployed in any form, including as a stand-alone program or as Modules, components, sub-normals or other units suitable for use in a computing environment. The computer program does not have to correspond to the file. The program may be stored in a portion of a file in which other programs or materials are stored, or in a single file dedicated to the program, or in a plurality of coordinated files (eg, storing one or more modules, subprograms, or portions of code) Archive)). Computer programs can be deployed to be executed on one computer or on multiple computers that are at one site or distributed across multiple sites and interconnected by a communication network.

The processes and logic flows described in this specification can be performed by one or more computer programs executing one or more programmable processors by performing operations on input data and generating output. The processes and logic flows may also be performed by dedicated logic circuitry, such as a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC), and the apparatus may be implemented as the dedicated logic circuitry.

The above grinding apparatus and method can be applied to a variety of grinding systems. The polishing pad or carrier head or both can be moved to provide relative motion between the abrasive surface and the substrate. For example, the platform can run on a track instead of rotating. The polishing pad can be a circular (or some other shape) pad that is fastened to the platform. Some aspects of the endpoint detection system can be applied to linear polishing systems, for example, where the polishing pad is a linear moving continuous or disc belt. The abrasive layer can be a standard (eg, a polyurethane with or without a filler) abrasive material, a soft material, or a fixed abrasive material. The term relative positioning is used; it should be understood that the abrasive surface and substrate can be fixed in a vertical orientation or some other orientation.

Specific embodiments of the invention have been described. Other embodiments are within the scope of the following patent claims. For example, the acts described in the scope of the claims can be performed in a different order and still achieve desirable results.

10. . . Substrate

12. . . Wafer

14. . . Outermost layer

16. . . Underlying layer

20. . . Grinding device

twenty two. . . Drive shaft

twenty four. . . Platform/rotary disc shape platform

25. . . Central axis/axis

26. . . Groove

30. . . Abrasive pad

36. . . Optical entrance

38. . . Grinding liquid

39. . . Combined slurry / rinse arm / arm

50. . . In-situ monitoring module/module

51. . . light source

52. . . Light detector / detector

53. . . Optical head

54. . . Bifurcated optical cable

55. . . Trunk

56. . . Branch/branch section

58. . . Branch/branch section

70. . . Carrier head

71. . . Central axis/axis

72. . . supporting structure

74. . . Carrying drive shaft

76. . . Carrier head rotation motor

301. . . point

302. . . point

303. . . point

304. . . point

305. . . point

306. . . point

307. . . point

308. . . point

309. . . point

310. . . point

311. . . point

600. . . method

602. . . step

604. . . step

606. . . step

608. . . step

610. . . step

612. . . step

614. . . step

616. . . step

620. . . step

622. . . step

624. . . step

626. . . step

630. . . step

705. . . Grinding stabilization cycle

710. . . Central region

715. . . Edge area

720. . . Intermediate area

730. . . index

735. . . Number of rotations

C _E . . . end

EDP. . . Grinding end point

E _u . . . rate

M _A . . . Excessive grinding rate

T ₁ . . . First grinding time

T ₂ . . . Later time

Figure 1 shows an example cross section of a portion of a substrate.

Figure 2 is a cross-sectional view showing an example of a polishing apparatus.

Figure 3 is a top plan view of one example of a rotating platform illustrating the position of the in-situ measurement.

Figure 4 is a diagram showing an example index trace from a spectral monitoring system showing a good data fit.

Figure 5 is a diagram showing an example index trace from a spectral monitoring system showing a weaker fit.

Figure 6 is a flow chart of an embodiment for determining the end point of the grinding.

Figure 7 illustrates an example plot of the grinding process versus time for a process to adjust the polishing rate.

The same component symbols and symbols in the various drawings indicate the same components.

600. . . method

602. . . step

604. . . step

606. . . step

608. . . step

610. . . step

612. . . step

614. . . step

616. . . step

620. . . step

622. . . step

624. . . step

626. . . step

630. . . step

Claims (25)

A computer-implemented substrate monitoring method comprising the steps of: obtaining a sequence of current spectra by an in-situ optical monitoring system, wherein each current spectrum from the current spectrum of the sequence is a spectrum of light reflected from a substrate, the substrate having An outermost layer subjected to grinding and at least one underlying layer; comparing each of the current spectra with a plurality of reference spectra from a first reference spectral library and determining a first best matching reference spectrum to produce a first sequence Best matching reference spectra; comparing each of the current spectra with a plurality of reference spectra from a second reference spectral library and determining a second best matching reference spectrum to produce a second sequence of second best matching reference spectra; a first fitness for the first sequence; determining a second fitness for the second sequence; and determining a first based on the first sequence, the second sequence, the first fitness, and the second fitness Grinding the end point. The method of claim 1, wherein the step of determining the polishing end point comprises the step of: determining whether the first best matching reference spectrum indicates an end point, and if so, determining whether the first fitness level is greater than the second Better fit, and if so, it is called an end point. The method of claim 1, further comprising the steps of: determining a first index value for each of the first best matching reference spectra to generate a sequence of first index values, comprising fitting a first function Fitting a first index of the sequence, determining a second index value for each of the second best matching reference spectra to generate a sequence of second index values, and A second function fits the second index value of the sequence. The method of claim 3, wherein the first function is a first linear function and the second function is a second linear function. The method of claim 3, wherein determining the first fitness comprises the steps of: determining a first index value of the sequence for one of the first functions, and wherein determining the second fitness comprises the following steps: A second index value of the sequence is determined to be suitable for one of the second functions. The method of claim 5, further comprising the step of determining whether the first fitness for the first sequence is better than the second fitness for the second sequence. The method of claim 1, wherein the plurality of reference spectra from the first reference spectral library represent a substrate having the first thickness of the underlying layer, and the plurality of substrates from the second reference spectral library The reference spectrum represents a substrate having the underlying layer of a different second thickness. The method of claim 1, wherein determining a first best matching reference spectrum comprises the steps of: determining which reference spectrum from the first reference spectral library has a minimum difference from the current spectrum, and wherein determining a The two best matching reference spectra include the step of determining which reference spectrum from the second reference spectral library has the smallest difference from the current spectrum. A computer-implemented substrate monitoring method includes the steps of: receiving a first sequence of reflected light from a first region of a substrate; receiving a second sequence of reflected light from a second region of the substrate Spectrum; comparing each current spectrum from the current spectrum of the first sequence with a first plurality of reference spectra from a first plurality of reference spectral libraries to generate a plurality of first sequence best matching reference spectra; comparing from the first Each of the current spectra of the two sequences of the current spectrum and the second plurality of reference spectra from a second plurality of reference spectral libraries to generate a plurality of second sequence best matching reference spectra; determining the best for the plurality of first sequences Matching a plurality of first fitness levels of the reference spectrum; and determining a plurality of second fitness levels for the plurality of second sequences to best match the reference spectrum. The method of claim 9, wherein the first plurality of reference spectral libraries are all identical to the second plurality of reference spectral libraries. The method of claim 9, wherein none of the first plurality of reference spectral libraries is the same as the second plurality of reference spectral libraries. The method of claim 9, wherein generating the plurality of first sequence best matching reference spectra comprises the steps of comparing each current spectrum from the current spectrum of the first sequence with a plurality of samples from a first reference spectrum library Reference spectra and determining a first intermediate fitness; comparing each current spectrum from the current spectrum of the first sequence with a plurality of reference spectra from a second reference spectrum library and determining a second intermediate fitness; Comparing the first intermediate fitness with the second intermediate fitness; selecting one of the first reference spectral library or the second reference spectral library based on comparing the first intermediate fitness with the second intermediate fitness; And determining a first sequence best matching reference spectrum based on the first selection; and wherein generating the plurality of second sequence best matching reference spectra comprises the steps of: comparing each current spectrum from the current spectrum of the second sequence with Comparing the plurality of reference spectra of the first reference spectrum library and determining a third intermediate fitness; comparing each current spectrum from the current spectrum of the second sequence with a plurality of reference spectra from the second reference spectrum library and determining a first a middle intermediate fitness; comparing the third intermediate fitness with the fourth intermediate fitness; selecting the first reference spectral library or the second reference based on comparing the third intermediate fitness with the second intermediate fitness One of the spectral libraries, and based on the second selection, determines a second sequence to best match the reference spectrum. The method of claim 12, wherein the first selection and the second selection are determined within a predetermined period of polishing. The method of claim 13, further comprising the steps of: determining a first polishing endpoint of the first region based on the first sequence best matching reference spectrum and a corresponding first fitness; and based on the second Sequence best matching reference spectrum and a corresponding second fitness A second grinding end point of one of the second regions is determined. A computer-implemented substrate monitoring method includes the steps of: receiving a first sequence current spectrum of reflected light from a first region of a substrate; receiving a second sequence current spectrum of reflected light from a second region of the substrate Comparing each of the current spectra from the current spectrum of the first sequence with a plurality of reference spectra from a first reference spectral library to produce a first sequence of best matching reference spectra; and comparing each of the current spectra from the second sequence A current spectrum and a plurality of reference spectra from a second reference spectral library to produce a second sequence of best matching reference spectra, the second reference spectral library being different from the first reference spectral library. The method of claim 15, wherein the first reference spectral library and the second reference spectral library are predetermined. A computer-implemented substrate monitoring method comprising the steps of: obtaining a sequence of current spectra by an in-situ optical monitoring system, wherein each current spectrum from the current spectrum of the sequence is a spectrum of light reflected from a substrate, the substrate having An outermost layer subjected to grinding and at least one underlying layer; comparing each of the current spectra with a first plurality of reference spectra from one of the first libraries of the plurality of reference spectral libraries to determine a first best matching reference spectrum, and Comparing each of the current spectra with a second plurality of reference spectra from a second bank of the plurality of reference spectral libraries to determine a second best matching reference spectrum; Determining which library provides a preferred fitness for the current spectrum of the sequence; and determining a polishing endpoint based on the current spectrum of the sequence and providing the library that best fits one of the current spectra of the sequence. A computer program product, as embodied on a computer readable medium, operable to cause a data processing device to perform an operation comprising: obtaining an array of current spectra from an in situ optical monitoring system, Each current spectrum of the current spectrum of the sequence is a spectrum of light reflected from a substrate having a top layer subjected to grinding and at least one underlying layer; comparing each of the current spectra with a library from a first reference spectrum A plurality of reference spectra and determining a first best matching reference spectrum to generate a first best matching reference spectrum of the first sequence; comparing each of the current spectra with a plurality of reference spectra from a second reference spectral library and determining one a second best matching reference spectrum to generate a second best matching reference spectrum of the second sequence; determining a first fitness for the first sequence; determining a second fitness for the second sequence; and based on The first sequence, the second sequence, the first fitness, and the second fitness determine a polishing endpoint. The computer program product of claim 18, wherein determining the polishing end point comprises: determining whether the first best matching reference spectrum indicates an end point, and if so, determining whether the first fitness level is greater than the second Better fit, and if so, it is called an end point. For example, if the computer program product of claim 18 is applied, further The method includes: determining a first index value for each first best matching reference spectrum to generate a sequence of first index values, fitting a first function to the first index value of the sequence, for each second best The matching reference spectrum determines a second index value to produce a sequence of second index values, and a second function fits the sequence second index value. The computer program product of claim 20, wherein the first function is a first linear function and the second function is a second linear function. The computer program product of claim 20, wherein determining the first fitness comprises the following operation: determining a first index value of the sequence is suitable for the first function, and wherein determining the second fitness includes the following Operation: Determine the suitability of the second index value of the sequence for one of the second functions. A computer program product, as embodied on a computer readable medium, operable to cause a data processing device to perform an operation comprising: obtaining an array of current spectra from an in situ optical monitoring system, Each current spectrum of the current spectrum of the sequence is a spectrum of light reflected from a substrate having a top layer subjected to grinding and at least one underlying layer; comparing each of the current spectra with one of a plurality of reference spectral libraries a first plurality of reference spectra of the first bank to determine a first best matching reference spectrum, and comparing each of the current spectra with a second plurality of reference spectra from one of the second libraries of the plurality of reference spectral libraries Determining a second best matching reference spectrum; Determining which library provides a preferred fitness for the current spectrum of the sequence; and determining a polishing endpoint based on the current spectrum of the sequence and providing the library that best fits one of the current spectra of the sequence. A computer program product encoded on a computer readable medium, the computer program product being operative to cause a data processing device to perform an operation comprising: receiving a reflected light from a first region of a substrate a sequence current spectrum; receiving a second sequence of reflected light from a second region of the substrate; comparing each current spectrum from the current spectrum of the first sequence with one from a first plurality of reference spectral libraries a plurality of reference spectra to generate a plurality of first sequence best matching reference spectra; comparing each current spectrum from the current spectrum of the second sequence with a second plurality of reference spectra from a second plurality of reference spectral libraries to generate a plurality of second sequences optimally matching the reference spectra; determining a plurality of first fitnesss for the plurality of first sequences to best match the reference spectra; and determining a plurality of first matching of the plurality of second sequences for the best matching reference spectra Two fitness. A computer program product encoded on a computer readable medium, the computer program product being operative to cause a data processing device to perform an operation comprising: receiving a reflected light from a first region of a substrate a sequence of current light Generating a second sequence of reflected light from a second region of the substrate; comparing each of the current spectra from the current spectrum of the first sequence with a plurality of reference spectra from a first reference spectral library to produce a The first sequence best matches the reference spectrum; and compares each of the current spectra from the current spectrum of the second sequence with a plurality of reference spectra from a second reference spectral library to produce a second sequence of best matching reference spectra, the first The second reference spectral library is different from the first reference spectral library.