WO2001032363A1 - Method of determining performance characteristics of polishing pads - Google Patents

Method of determining performance characteristics of polishing pads Download PDF

Info

Publication number
WO2001032363A1
WO2001032363A1 PCT/US2000/028196 US0028196W WO0132363A1 WO 2001032363 A1 WO2001032363 A1 WO 2001032363A1 US 0028196 W US0028196 W US 0028196W WO 0132363 A1 WO0132363 A1 WO 0132363A1
Authority
WO
WIPO (PCT)
Prior art keywords
site
wafer
polishing
wafers
polishing pads
Prior art date
Application number
PCT/US2000/028196
Other languages
French (fr)
Inventor
Kevin A. Keller
Gerald A. Whitman, Jr.
Original Assignee
Memc Electronic Materials, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Memc Electronic Materials, Inc. filed Critical Memc Electronic Materials, Inc.
Publication of WO2001032363A1 publication Critical patent/WO2001032363A1/en

Links

Classifications

    • BPERFORMING OPERATIONS; TRANSPORTING
    • B24GRINDING; POLISHING
    • B24BMACHINES, DEVICES, OR PROCESSES FOR GRINDING OR POLISHING; DRESSING OR CONDITIONING OF ABRADING SURFACES; FEEDING OF GRINDING, POLISHING, OR LAPPING AGENTS
    • B24B37/00Lapping machines or devices; Accessories
    • B24B37/11Lapping tools
    • B24B37/20Lapping pads for working plane surfaces
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B24GRINDING; POLISHING
    • B24BMACHINES, DEVICES, OR PROCESSES FOR GRINDING OR POLISHING; DRESSING OR CONDITIONING OF ABRADING SURFACES; FEEDING OF GRINDING, POLISHING, OR LAPPING AGENTS
    • B24B49/00Measuring or gauging equipment for controlling the feed movement of the grinding tool or work; Arrangements of indicating or measuring equipment, e.g. for indicating the start of the grinding operation
    • HELECTRICITY
    • H01ELECTRIC ELEMENTS
    • H01LSEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
    • H01L22/00Testing or measuring during manufacture or treatment; Reliability measurements, i.e. testing of parts without further processing to modify the parts as such; Structural arrangements therefor
    • H01L22/20Sequence of activities consisting of a plurality of measurements, corrections, marking or sorting steps

Definitions

  • This invention relates to polishing pads for polishing semiconductor wafers, and more particularly to determining performance characteristics of a production lot of polishing pads using discriminant analysis.
  • Semiconductor wafers are generally prepared from a single crystal ingot sliced into individual wafers.
  • the wafers are subjected to several processing operations to reduce the thickness of the wafer, remove damage caused by the slicing operation, and create a highly reflective surface.
  • a lapping operation is performed on the front and back surfaces of the wafer using an abrasive slurry and a set of rotating lapping plates. The lapping operation reduces the thickness of the wafer to remove surface damage induced by the slicing operation and to make the opposing side surfaces of each wafer flat and parallel.
  • the wafers Upon completion of the lapping operation, the wafers are subjected to a chemical etching operation to reduce further the thickness of the wafer and remove mechanical damage produced by the prior processing operations.
  • One side surface of each wafer (often called the "front" side of the wafer) is then polished using a polishing pad, a colloidal silica slurry (polishing slurry) and a chemical etchant to ensure that the wafer has a highly reflective, damage-free surface.
  • This invention relates in particular to the polishing process.
  • Semiconductor wafers used as starting materials for the fabrication of integrated circuits must meet certain surface flatness requirements. Such wafers must be particularly flat to print circuits on them by, for example, an electron beam-lithographic or photolithographic process. Wafer flatness in the focal point of the electron beam delineator or optical printer is important for uniform imaging in the electron beam-lithographic and photolithographic processes. The flatness of the wafer surface directly influences device line width capability, process latitude, yield and throughput. The continuing reduction in device geometry and increasingly stringent device fabrication specifications force manufacturers of semiconductor wafers to prepare increasingly flatter wafers.
  • Wafers can be characterized for flatness in terms of a global flatness variation parameter (for example, total thickness variation ("TTV”)) or in terms of a local site flatness variation parameter (e.g., Site Total Indicated Reading (“STIR”) or Site Focal Plane Deviation (“SFPD”)) as measured against a reference plane of the wafer (e.g., Back Reference Center Focus (“SBIR”) or Site Best Fit Reference Plane (“SFQR”)).
  • TTV total thickness variation
  • SBIR Back Reference Center Focus
  • SBQR Site Best Fit Reference Plane
  • polishing pad performance between production lots has been variable, due to minor variations in the supplier's manufacturing process.
  • the quality of the pads is typically consistent because each pad is stamped from the same raw material. Therefore, each production lot of polishing pads must be separately assessed to determine its quality and usefulness.
  • Grade I semiconductor material is expensive to manufacture, and its use in testing pad quality places the polished semiconductor material at risk of flatness degradation which may render the wafer unsuitable for use. Therefore, any reduction in the amount of Grade I material placed at risk during the testing process is beneficial.
  • a new production lot of pads is used in the wafer polishing process.
  • their quality is unknown.
  • the new pads are broken in, if the quality of wafer flatness remains high, the pads are considered good and not analyzed further.
  • pad quality is evaluated.
  • a testing and analyzing method for determining the performance characteristics of a production lot of polishing pads is disclosed.
  • the first step requires polishing a plurality of semiconductor wafers with a selected number of polishing pads from the production lot of polishing pads. At least a portion of the surface area of each of the plurality of semiconductor wafers is divided into a plurality of sites. At least one wafer characteristic from each site is measured. The wafer characteristic from each site is incorporated into a discriminant function selected to predict the performance characteristics of a production lot of polishing pads by the quantitative level of the data obtained from a plurality of semiconductor wafers polished by the selected number of polishing pads.
  • FIG. 1 is a schematic of a 150 mm wafer divided into sites.
  • Corresponding reference characters indicate corresponding parts throughout the several views of the drawings.
  • the present method is adapted to efficiently determine if a production lot of polishing pads is of good or bad quality.
  • An early determination of pad quality is essential to semiconductor wafer production because it ensures proper polishing and reduces the amount of material polished by pads of bad quality.
  • the present method preferably polishes 20 to 40 wafers with pads of unknown quality from the same production lot. Various sites on the wafers are then measured for flatness and the results are inserted into a discriminant function developed for the present method. The results of the discriminant function show whether the polishing pads, and the lot from which they were manufactured, is of good or bad quality.
  • Discriminant analysis is a statistical tool designed to categorize objects into classes. By creating a discriminant function based on a discriminant analysis, the function can serve as a method for assigning objects to their proper class.
  • pertinent types of observations from a particular group of similar objects must be identified. These might include dimensions, tolerances, quality measures, etc.
  • the classes of interest must be defined, as the goal of the discriminant analysis is to place each object into its proper class. These classes might be good and bad, high quality and low quality, accept or do not accept, etc. These classes represent the ultimate goal of the classification process. New data observations can then be applied to the discriminant function, predicting in which class the data belongs.
  • discriminant function For every p qualitative levels, or classes, there exists (p-1 ) discriminant functions. In the preferred embodiment, two classes, good and bad, were selected, producing one discriminant function. For "good” and “bad” classes, the discriminant function will predict if a given set of new data is "good” or "bad.”
  • the discriminant analysis compiles these multiple observations and transforms the multivariate observations into univariate observations. These univariate observations separate the objects belonging in different classes to a greater degree than the multivariate observations. More specifically, linear combinations of the relevant multivariate observations are mathematically manipulated to create the univariate observations. The linear combinations are specifically selected in the discriminant analysis to maximize the squared distance between the mean of the observations from each class. By maximizing the squared distance between the observations from each class, the univariate distributions enhance the impact of those observations that most clearly and easily categorize the objects into classes. This process also reduces the impact of those observations that do not help differentiate between the classes.
  • the univariate observations are useful tools for separating objects into defined classes where the variety and number of observations are great and an understanding of which observations best predict class membership is difficult.
  • the discriminant function initially, two sets of wafers were distributed on 14 polishers and polished. The first set was polished with polishing pads from a good production lot. The second set was polished with polishing pads from a bad production lot. The polishing pads characterized as "good” or "bad” were characterized as such by using prior art methods for determining quality. Site flatness measurements were recorded and flatness data calculated for all wafers. A selection of 24 wafers from each class, good and bad, were randomly selected to be the data points used in establishing the discriminant function.
  • the mathematical manipulations required to establish the discriminant function are set out in this specification after the description of the method used to apply the function. By inserting the remaining wafer data into the discriminant function, the accuracy of the discriminant function was confirmed. The discriminant function correctly predicted whether the wafers were polished on a good lot or bad lot of polishing pads.
  • the first step in applying the method of determining the quality of a production lot of polishing pads is polishing the wafers.
  • four polishing pads from a production lot of unknown quality are selected at random.
  • the polishing pads are then used to polish a small number of wafers, approximately 20 to 40. Because of the discriminant function's accuracy in determining pad quality, assessing polishing pad production lot quality places only these 20 to 40 Grade I wafers at risk.
  • the polished wafer surfaces are divided into discrete sites (Fig. 1 ). Global parameters, measurements taken over the entire wafer surface, are useful to describe the overall shape of the wafer surface, but more specific descriptions of wafer flatness are available from measurements taken at specific sites on the surface of the wafer.
  • the wafer is divided into areas of equal size, called blocks or sites (Fig. 1 ). Sites that rest on the edge of the wafer, having only some of their area covering wafer material, are called partial sites.
  • utilization of partial sites in the analysis is called partials active, while disregarding measurements from the partial sites is called partials inactive.
  • the method and discriminant function disclosed herein apply the partials active methodology, but the use of partials inactive is also contemplated by the present invention.
  • the STIR values from all wafers are averaged, creating an average STIR value for each wafer site.
  • the discriminant function may also be applied with the site data from a single wafer, rather than an average, yielding a score for each wafer. Yet averaging the STIR values from several wafers helps diminish the variability among the wafers, yielding better predictive results from the model.
  • the wafer parameter calculations are complete. Finally, the average STIR value for each site is incorporated into the discriminant function developed for determining production lot quality. The specific mathematical manipulations required for developing the discriminant function are disclosed below.
  • the completed discriminant function is a multivariate equation where the score assigned to the wafer set is equal to a 43-term equation.
  • the equation is generally described as follows, where n corresponds to a given site on the wafers, k n is a coefficient associated with each site, STIR n is a STIR value, or average STIR value, calculated for a particular site, and K is a constant.
  • the resulting total is the score for the set of wafers. If the score is greater than zero, the polishing pads belong to a good production lot. Conversely, if the score is less than zero, the polishing pads belong to a bad production lot. For example, in applying the preferred embodiment to 24 wafers polished by pads from a bad polishing lot, the poor quality of the pads was easily recognizable because the wafer score was less than zero. Applying the preferred embodiment to 24 wafers polished by pads from a good polishing lot, the good quality of the pads is easily recognizable because the score was greater than zero. As the scores increase from zero, the quality of the polishing pad lot increases. For example, a pad lot scoring 4 would be of greater quality than a pad lot scoring 2. Because the discriminant function yields a quantitative result, a numerical score, the quality of the production lot of polishing pads can be readily ascertained with little data interpretation and no qualitative judgments.
  • the method described in the present invention is applicable to discriminant functions having more than two qualitative levels, or classes. For example, rather than choosing two classes, the present invention could divide polishing pad production lot quality into four classes: superb, good, fair, and poor. With four classes, a discriminant analysis of the pads would yield three discriminant functions.
  • the discriminant analyses described herein are similar to the analysis and description in the text of Applied Multivariate Statistical Analysis, the disclosure of which is incorporated by reference. R.A. JOHNSON & D. W. WICHERN, Applied Multivariate Statistical Analysis 53-56, 62, 470-82, 509-510 (2d ed., Prentice Hall International 1988).
  • the data manipulations required to initially establish the discriminant function for this application are hereinafter described.
  • data must be collected from wafers polished on good and bad quality pads. This data should be in the form of STIR values for each of the 42 sites on each wafer. From this data, the mean, standard deviation, and coefficient of correlation will each be calculated.
  • the mean for each site is designated x n , where n denotes the site number.
  • the standard deviation is designated ⁇ n , where n denotes the site number.
  • the coefficient of correlation between the two sets of data is designated p n m , where n denotes the site number corresponding to the first data class and m denotes the site number corresponding to the second data class.
  • a vector of sample averages is constructed from the data for each class.
  • two vectors are calculated, each being a 42 row by 1 column array, representing the mean STIR values from each site.
  • x ⁇ bar and x 2 bar each represent vector arrays containing the mean values for wafer sites 1 to 42; y bar representing the mean STI R values for the wafers polished on good pads and x 2 bar representing the mean STIR values for the wafers polished on bad pads.
  • the covariance matrix is a 42 by 42 matrix, defined as follows:
  • Wafer Score 27.6541 STIR, - 34.428 x STIR 2 + . . . + 61.2205 x STIR 42
  • the final data manipulation relates to shifting the intercept of the discriminant function by adding a constant term. It is desirable to have positive wafer scores for good production pads and negative wafer scores for bad production pads. Accordingly, the constant s, equivalent to -32.0343, was added to the equation to decrease the raw scores such that wafer scores from good pads would remain positive while wafer scores from bad pads would be negative. In doing so, the quality of the polishing pads is readily apparent from the sign of the scores. Additionally, as the score increases above zero, the quality and performance of the pads increase. With the final discriminant function developed, new data may be inserted into the function as described above, yielding scores which accurately and efficiently classify the polishing pads. The discriminant function coefficients may be redefined by collecting and incorporating more data, but such a process is not typically required unless another process variable changes significantly. In view of the above, it will be seen that the several objects of the invention are achieved and other advantageous results attained.

Abstract

A testing and analyzing method for determining the performance characteristics of a production lot of polishing pads. The first step requires polishing a plurality of semiconductor wafers with a selected number of polishing pads from the production lot of polishing pads. At least a portion of the surface area of each of the plurality of semiconductor wafers is divided into a plurality of sites. At least one wafer characteristic from each site is measured. The wafer characteristic from each site is incorporated into a discriminant function selected to predict the performance characteristics of a production lot of polishing pads by the quantitative level of the data obtained from a plurality of semiconductor wafers polished by the selected number of polishing pads.

Description

METHOD OF DETERMINING PERFORMANCE CHARACTERISTICS OF POLISHING PADS
Background of the Invention
This invention relates to polishing pads for polishing semiconductor wafers, and more particularly to determining performance characteristics of a production lot of polishing pads using discriminant analysis.
Semiconductor wafers are generally prepared from a single crystal ingot sliced into individual wafers. The wafers are subjected to several processing operations to reduce the thickness of the wafer, remove damage caused by the slicing operation, and create a highly reflective surface. In conventional wafer shaping processes, a lapping operation is performed on the front and back surfaces of the wafer using an abrasive slurry and a set of rotating lapping plates. The lapping operation reduces the thickness of the wafer to remove surface damage induced by the slicing operation and to make the opposing side surfaces of each wafer flat and parallel. Upon completion of the lapping operation, the wafers are subjected to a chemical etching operation to reduce further the thickness of the wafer and remove mechanical damage produced by the prior processing operations. One side surface of each wafer (often called the "front" side of the wafer) is then polished using a polishing pad, a colloidal silica slurry (polishing slurry) and a chemical etchant to ensure that the wafer has a highly reflective, damage-free surface. This invention relates in particular to the polishing process.
Semiconductor wafers used as starting materials for the fabrication of integrated circuits must meet certain surface flatness requirements. Such wafers must be particularly flat to print circuits on them by, for example, an electron beam-lithographic or photolithographic process. Wafer flatness in the focal point of the electron beam delineator or optical printer is important for uniform imaging in the electron beam-lithographic and photolithographic processes. The flatness of the wafer surface directly influences device line width capability, process latitude, yield and throughput. The continuing reduction in device geometry and increasingly stringent device fabrication specifications force manufacturers of semiconductor wafers to prepare increasingly flatter wafers. Wafers can be characterized for flatness in terms of a global flatness variation parameter (for example, total thickness variation ("TTV")) or in terms of a local site flatness variation parameter (e.g., Site Total Indicated Reading ("STIR") or Site Focal Plane Deviation ("SFPD")) as measured against a reference plane of the wafer (e.g., Back Reference Center Focus ("SBIR") or Site Best Fit Reference Plane ("SFQR")). A more detailed discussion of the characterization of wafer flatness can be found in F. SHIMURA, Semiconductor Silicon Crystal Technology 191-195 (Academic Press 1989).
Because a wafer's flatness is of paramount importance, production variables affecting wafer flatness are closely monitored. The quality of the polishing pad used to polish the wafer, for example, directly affects the flatness characteristics of the polished wafer. Therefore, an accurate assessment of the polishing pad quality, and particularly identification of bad lots, is important to semiconductor wafer production. Historically, polishing pad performance between production lots has been variable, due to minor variations in the supplier's manufacturing process. In contrast, within a production lot, the quality of the pads is typically consistent because each pad is stamped from the same raw material. Therefore, each production lot of polishing pads must be separately assessed to determine its quality and usefulness. To assess the quality of a particular production lot of wafers, several pads must be tested on Grade I semiconductor material to assess the overall quality of the pad. Grade I semiconductor material is expensive to manufacture, and its use in testing pad quality places the polished semiconductor material at risk of flatness degradation which may render the wafer unsuitable for use. Therefore, any reduction in the amount of Grade I material placed at risk during the testing process is beneficial.
Assessing the quality of a production lot of wafers traditionally has been a qualitative task, based upon the wafer shape and other flatness measures. According to the prior methodology, once a current production lot of good polishing pads is exhausted, a new production lot of pads is used in the wafer polishing process. When these new pads are installed on the polishers, their quality is unknown. As the new pads are broken in, if the quality of wafer flatness remains high, the pads are considered good and not analyzed further. Alternately, if wafer quality appears to suffer and substandard polishing pads are suspected, pad quality is evaluated. By evaluating the flatness of the wafers with different flatness measures, as described above, the questionable production lot is characterized as bad or marginal. This in-use type evaluation process is time consuming and costly. Because the pads are used in production before their quality is appreciated, they place many Grade I wafers at risk of flatness degradation and possible loss before a problem is identified. This problem is exacerbated if quality problems associated with the pads are difficult to detect and go undetected for significant periods. Once the lot quality is determined to be bad, the lot is rejected and returned to the pad manufacturer. Marginal production lots are subject to further pad testing and further destruction of Grade I material until a determination of lot quality can be made. Because this process is time consuming and subjects many Grade I wafers to risk of flatness degradation and potential loss, a more efficient and less costly methodology for determining pad quality is necessary.
Summary of the Invention
Among the several objects of this invention may be noted the provision of an improved methodology that reduces how much Grade I semiconductor material is placed at risk of flatness degradation while testing polishing pad quality; the provision of such a methodology that speeds the reaction time to identify poorly performing pad production lots; the provision of such a methodology that increases the accuracy of the determination of pad production lot quality; the provision of such a methodology that decreases the number of polishers required to determine pad quality; the provision of such a methodology that employs a directive, proactive, methodology which actively determines the quality of a production lot of polishing pads; the provision of such a methodology that uses quantitative, rather than qualitative, measurements to determine pad production lot quality; and the provision of such a methodology that creates an easily understandable grading system, allowing quick comparison of production lot quality to predetermined standards.
Generally, a testing and analyzing method for determining the performance characteristics of a production lot of polishing pads is disclosed. The first step requires polishing a plurality of semiconductor wafers with a selected number of polishing pads from the production lot of polishing pads. At least a portion of the surface area of each of the plurality of semiconductor wafers is divided into a plurality of sites. At least one wafer characteristic from each site is measured. The wafer characteristic from each site is incorporated into a discriminant function selected to predict the performance characteristics of a production lot of polishing pads by the quantitative level of the data obtained from a plurality of semiconductor wafers polished by the selected number of polishing pads.
Other objects and features will be in part apparent and in part pointed out hereinafter.
Brief Description of the Drawings
FIG. 1 is a schematic of a 150 mm wafer divided into sites. Corresponding reference characters indicate corresponding parts throughout the several views of the drawings.
Description of the Preferred Embodiments
Generally, the present method is adapted to efficiently determine if a production lot of polishing pads is of good or bad quality. An early determination of pad quality is essential to semiconductor wafer production because it ensures proper polishing and reduces the amount of material polished by pads of bad quality. The present method preferably polishes 20 to 40 wafers with pads of unknown quality from the same production lot. Various sites on the wafers are then measured for flatness and the results are inserted into a discriminant function developed for the present method. The results of the discriminant function show whether the polishing pads, and the lot from which they were manufactured, is of good or bad quality.
Discriminant analysis is a statistical tool designed to categorize objects into classes. By creating a discriminant function based on a discriminant analysis, the function can serve as a method for assigning objects to their proper class. To perform a discriminant analysis, pertinent types of observations from a particular group of similar objects must be identified. These might include dimensions, tolerances, quality measures, etc. Then, the classes of interest must be defined, as the goal of the discriminant analysis is to place each object into its proper class. These classes might be good and bad, high quality and low quality, accept or do not accept, etc. These classes represent the ultimate goal of the classification process. New data observations can then be applied to the discriminant function, predicting in which class the data belongs. For every p qualitative levels, or classes, there exists (p-1 ) discriminant functions. In the preferred embodiment, two classes, good and bad, were selected, producing one discriminant function. For "good" and "bad" classes, the discriminant function will predict if a given set of new data is "good" or "bad."
To determine in what class each object belongs, relevant observations are recorded for each object. The discriminant analysis compiles these multiple observations and transforms the multivariate observations into univariate observations. These univariate observations separate the objects belonging in different classes to a greater degree than the multivariate observations. More specifically, linear combinations of the relevant multivariate observations are mathematically manipulated to create the univariate observations. The linear combinations are specifically selected in the discriminant analysis to maximize the squared distance between the mean of the observations from each class. By maximizing the squared distance between the observations from each class, the univariate distributions enhance the impact of those observations that most clearly and easily categorize the objects into classes. This process also reduces the impact of those observations that do not help differentiate between the classes. By maximizing the impact of those observations that most clearly separate the objects into their proper class, the univariate observations are useful tools for separating objects into defined classes where the variety and number of observations are great and an understanding of which observations best predict class membership is difficult. To establish the discriminant function initially, two sets of wafers were distributed on 14 polishers and polished. The first set was polished with polishing pads from a good production lot. The second set was polished with polishing pads from a bad production lot. The polishing pads characterized as "good" or "bad" were characterized as such by using prior art methods for determining quality. Site flatness measurements were recorded and flatness data calculated for all wafers. A selection of 24 wafers from each class, good and bad, were randomly selected to be the data points used in establishing the discriminant function. The mathematical manipulations required to establish the discriminant function are set out in this specification after the description of the method used to apply the function. By inserting the remaining wafer data into the discriminant function, the accuracy of the discriminant function was confirmed. The discriminant function correctly predicted whether the wafers were polished on a good lot or bad lot of polishing pads.
The first step in applying the method of determining the quality of a production lot of polishing pads is polishing the wafers. In the preferred embodiment, four polishing pads from a production lot of unknown quality are selected at random. The polishing pads are then used to polish a small number of wafers, approximately 20 to 40. Because of the discriminant function's accuracy in determining pad quality, assessing polishing pad production lot quality places only these 20 to 40 Grade I wafers at risk.
Secondly, the polished wafer surfaces are divided into discrete sites (Fig. 1 ). Global parameters, measurements taken over the entire wafer surface, are useful to describe the overall shape of the wafer surface, but more specific descriptions of wafer flatness are available from measurements taken at specific sites on the surface of the wafer. To measure site data, the wafer is divided into areas of equal size, called blocks or sites (Fig. 1 ). Sites that rest on the edge of the wafer, having only some of their area covering wafer material, are called partial sites. According to conventional measurement methods, utilization of partial sites in the analysis is called partials active, while disregarding measurements from the partial sites is called partials inactive. The method and discriminant function disclosed herein apply the partials active methodology, but the use of partials inactive is also contemplated by the present invention. The steps described herein were applied to 150 mm diameter wafers, and the disclosed method equally applies to any size wafer. For a 150 mm wafer, division into square sites measuring 20 mm per side creates an array of 42 sites, including the partial sites, as shown in Fig. 1. Site selection excludes additional possible sites, beyond the chosen 42, near the wafer's edge where only a small portion of the wafer surface lies within the site area. Sites 20 mm square were selected for convenience due to previous testing and familiarity with sites of that size, but the invention is equally applicable to sites of any size. Variations in site size inversely influence the number of sites and hence the number of terms in the discriminant function.
Each individual site is compared with an ideal reference plane lying parallel to the wafer surface. The reference plane is determined for each wafer, representing a theoretical wafer surface for a perfectly flat wafer. For each site on the wafer, measurements are taken to determine the highest point on the wafer above the reference plane and the lowest point below the reference plane. From this data, a Site Total Indicated Reading ("STIR") is calculated for each site. A STIR value is defined as the absolute value of the difference between the highest point on the surface above the reference plane and the lowest point on the surface below the reference plane, measured for a particular site. The numerical values printed within each wafer site of Fig. 1 are examples of STIR values for a particular wafer, measured in microns and included for illustrative purposes only. For each site, the STIR values from all wafers are averaged, creating an average STIR value for each wafer site. The discriminant function may also be applied with the site data from a single wafer, rather than an average, yielding a score for each wafer. Yet averaging the STIR values from several wafers helps diminish the variability among the wafers, yielding better predictive results from the model. Once the average STIR value is calculated for each of the 42 sites, the wafer parameter calculations are complete. Finally, the average STIR value for each site is incorporated into the discriminant function developed for determining production lot quality. The specific mathematical manipulations required for developing the discriminant function are disclosed below. Without describing those manipulations here, the completed discriminant function is a multivariate equation where the score assigned to the wafer set is equal to a 43-term equation. The equation is generally described as follows, where n corresponds to a given site on the wafers, kn is a coefficient associated with each site, STIRn is a STIR value, or average STIR value, calculated for a particular site, and K is a constant.
Wafer Score = kn x STIR„ + κ n=\ For each n, the corresponding kn is shown in the table below
Figure imgf000009_0001
The constant K is equivalent to -32.0343. The development of each of the 43 terms is discussed infra. Each of the first 42 terms of the equation corresponds to an average STIR value calculated for a particular site, multiplied by the discriminant function coefficient {kn) developed for that particular site. The 43rd term of the equation is an additional constant. Each of the 43 terms is calculated and their results are subsequently summed. The following equation shows the discriminant function with the coefficients and constant represented numerically: Wafer Score =
27.6541 STIR, - 34.428 * STIR2 + . . . + 61.2205 * STIR42 - 32.0343
The resulting total is the score for the set of wafers. If the score is greater than zero, the polishing pads belong to a good production lot. Conversely, if the score is less than zero, the polishing pads belong to a bad production lot. For example, in applying the preferred embodiment to 24 wafers polished by pads from a bad polishing lot, the poor quality of the pads was easily recognizable because the wafer score was less than zero. Applying the preferred embodiment to 24 wafers polished by pads from a good polishing lot, the good quality of the pads is easily recognizable because the score was greater than zero. As the scores increase from zero, the quality of the polishing pad lot increases. For example, a pad lot scoring 4 would be of greater quality than a pad lot scoring 2. Because the discriminant function yields a quantitative result, a numerical score, the quality of the production lot of polishing pads can be readily ascertained with little data interpretation and no qualitative judgments.
In an alternate embodiment, a wafer score may be calculated for each wafer, creating a set of 20 to 40 numerical scores for a selection of polishing pads. The 20 to 40 scores are then plotted on a bar chart having score ranges along the x-axis and the number of wafers in that score range determining the height of the bar along the y-axis. The data set from each class should yield a normal distribution of wafer scores. If the distribution of scores is greater than zero, polishing pads from a good production lot polished the wafers. Conversely, if the distribution of scores is less than zero, polishing pads from a bad production lot polished the wafers.
The method described in the present invention is applicable to discriminant functions having more than two qualitative levels, or classes. For example, rather than choosing two classes, the present invention could divide polishing pad production lot quality into four classes: superb, good, fair, and poor. With four classes, a discriminant analysis of the pads would yield three discriminant functions. The discriminant analyses described herein are similar to the analysis and description in the text of Applied Multivariate Statistical Analysis, the disclosure of which is incorporated by reference. R.A. JOHNSON & D. W. WICHERN, Applied Multivariate Statistical Analysis 53-56, 62, 470-82, 509-510 (2d ed., Prentice Hall International 1988).
The data manipulations required to initially establish the discriminant function for this application are hereinafter described. First, data must be collected from wafers polished on good and bad quality pads. This data should be in the form of STIR values for each of the 42 sites on each wafer. From this data, the mean, standard deviation, and coefficient of correlation will each be calculated. For calculation of the mean, the data from the good polishing pads should be analyzed separately from the data from the bad polishing pads. The mean for each site is designated xn, where n denotes the site number. For calculating the standard deviation and the coefficient of correlation, the data from both the good and bad polishing pads should be analyzed together. The standard deviation is designated σn, where n denotes the site number. The coefficient of correlation between the two sets of data is designated pn m, where n denotes the site number corresponding to the first data class and m denotes the site number corresponding to the second data class.
Once the raw data is calculated, a vector of sample averages is constructed from the data for each class. With 42 variables from each wafer, two vectors are calculated, each being a 42 row by 1 column array, representing the mean STIR values from each site. In the following arrays, x} bar and x2 bar each represent vector arrays containing the mean values for wafer sites 1 to 42; y bar representing the mean STI R values for the wafers polished on good pads and x2 bar representing the mean STIR values for the wafers polished on bad pads.
Figure imgf000012_0001
Assuming both class populations have the same covariance matrix, as is required in any discriminant function analysis, the covariance matrices for each class are combined to form a single population variance-covariance matrix. The single population variance-covariance matrix is designated Spooled. The covariance matrix is a 42 by 42 matrix, defined as follows:
Figure imgf000012_0002
pooled
Figure imgf000012_0003
With Spooled in hand, the inverse of S led is calculated, employing standard techniques for matrix inversion, yielding \Spooled Λ.
With [Spooled]' Xj bar, and x2 bar, the discriminant function is readily calculated. The discriminant function for two classes is generally defined as follows:
Figure imgf000013_0001
Applying the data from this application to the general discriminant function, the equation yields a three-term matrix multiplication. A 1 row by 42 column matrix ( x7 bar minus x2 bar) is first multiplied by a 42 row by 42 column matrix ( [Spooled]'1) and then multiplied by a 42 row by one column matrix (an xn for each site). The resulting equation is the 42 term equation described below:
Wafer Score = 27.6541 x7 - 34.428 * x2 + . . . + 61.2205 x x 42
Applying the equation to the present embodiment by substituting a more descriptive variable forx,., the equation yields:
Wafer Score = 27.6541 STIR, - 34.428 x STIR2 + . . . + 61.2205 x STIR42
The final data manipulation relates to shifting the intercept of the discriminant function by adding a constant term. It is desirable to have positive wafer scores for good production pads and negative wafer scores for bad production pads. Accordingly, the constant s, equivalent to -32.0343, was added to the equation to decrease the raw scores such that wafer scores from good pads would remain positive while wafer scores from bad pads would be negative. In doing so, the quality of the polishing pads is readily apparent from the sign of the scores. Additionally, as the score increases above zero, the quality and performance of the pads increase. With the final discriminant function developed, new data may be inserted into the function as described above, yielding scores which accurately and efficiently classify the polishing pads. The discriminant function coefficients may be redefined by collecting and incorporating more data, but such a process is not typically required unless another process variable changes significantly. In view of the above, it will be seen that the several objects of the invention are achieved and other advantageous results attained.
When introducing elements of the present invention or the preferred embodiment(s) thereof, the articles "a", "an", "the" and "said" are intended to mean that there are one or more of the elements. The terms "comprising", "including" and
"having" are intended to be inclusive and mean that there may be additional elements other than the listed elements.
As various changes could be made in the above without departing from the scope of the invention, it is intended that all matter contained in the above description and shown in the accompanying drawings shall be interpreted as illustrative and not in a limiting sense.

Claims

WHAT IS CLAIMED IS:
1. A testing and analyzing method for determining the performance characteristics of a production lot of polishing pads, comprising the steps of: polishing a plurality of semiconductor wafers with a selected number of polishing pads from the production lot of polishing pads; dividing at least a portion of the surface area of each of the plurality of semiconductor wafers into a plurality of sites; measuring at least one wafer characteristic from each site; and incorporating said wafer characteristic from each site into a discriminant function selected to predict the performance characteristics of a production lot of polishing pads by the quantitative level of the data obtained from a plurality of semiconductor wafers polished by the selected number of polishing pads.
2. A method according to claim 1 , comprising dividing the surface area of each of the plurality of semiconductor wafers into a plurality of sites of the same area and shape.
3. A method according to claim 2, comprising dividing the surface area of each of the plurality of semiconductor wafers into a plurality of sites forming a grid of sites.
4. A method according to claim 1 , wherein said step of incorporating said wafer characteristic comprises determining flatness for each site.
5. A method as set forth in claim 4 wherein determining flatness for each site comprises calculating a Site Total Indicated Reading ("STIR") for each site, the STIR value being defined as an absolute value of a difference between a highest point of a site above a reference plane and a lowest point of a site below the reference plane, the reference plane representing a theoretical wafer surface for a perfectly flat wafer and lying generally parallel to the wafer surface.
6. A method according to claim 1 , wherein said step of polishing said wafers comprises polishing fewer than 40 wafers.
7. A method according to claim 6, wherein said step of polishing said wafers comprises polishing at least 20 wafers.
8. A method according to claim 1 , wherein said step of incorporating said wafer characteristic from each site into a discriminant function is preceded by a step of calculating an average wafer characteristic for each site from the plurality of wafers and subsequently incorporating the average from each site into the discriminant function.
PCT/US2000/028196 1999-11-03 2000-10-12 Method of determining performance characteristics of polishing pads WO2001032363A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US09/432,928 US6293139B1 (en) 1999-11-03 1999-11-03 Method of determining performance characteristics of polishing pads
US09/432,928 1999-11-03

Publications (1)

Publication Number Publication Date
WO2001032363A1 true WO2001032363A1 (en) 2001-05-10

Family

ID=23718142

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2000/028196 WO2001032363A1 (en) 1999-11-03 2000-10-12 Method of determining performance characteristics of polishing pads

Country Status (2)

Country Link
US (1) US6293139B1 (en)
WO (1) WO2001032363A1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6293139B1 (en) 1999-11-03 2001-09-25 Memc Electronic Materials, Inc. Method of determining performance characteristics of polishing pads
WO2019060582A1 (en) * 2017-09-25 2019-03-28 Applied Materials, Inc. Semiconductor fabrication using process control parameter matrix

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE19905737C2 (en) * 1999-02-11 2000-12-14 Wacker Siltronic Halbleitermat Method for producing a semiconductor wafer with improved flatness
US6344416B1 (en) * 2000-03-10 2002-02-05 International Business Machines Corporation Deliberate semiconductor film variation to compensate for radial processing differences, determine optimal device characteristics, or produce small productions
GB0019293D0 (en) * 2000-08-07 2000-09-27 Federal Mogul Friction Product Magnetic clamping arrangement
US6605159B2 (en) * 2001-08-30 2003-08-12 Micron Technology, Inc. Device and method for collecting and measuring chemical samples on pad surface in CMP
US7160173B2 (en) * 2002-04-03 2007-01-09 3M Innovative Properties Company Abrasive articles and methods for the manufacture and use of same
US6723574B1 (en) * 2002-09-26 2004-04-20 Lam Research Corporation Method for quantifying uniformity patterns and including expert knowledge for tool development and control
US7089081B2 (en) * 2003-01-31 2006-08-08 3M Innovative Properties Company Modeling an abrasive process to achieve controlled material removal
US6939200B2 (en) * 2003-09-16 2005-09-06 Hitachi Global Storage Technologies Netherlands B.V. Method of predicting plate lapping properties to improve slider fabrication yield
US6918815B2 (en) * 2003-09-16 2005-07-19 Hitachi Global Storage Technologies Netherlands B.V. System and apparatus for predicting plate lapping properties to improve slider fabrication yield
US7908023B2 (en) * 2008-01-14 2011-03-15 International Business Machines Corporation Method of establishing a lot grade system for product lots in a semiconductor manufacturing process
JP7219009B2 (en) * 2018-03-27 2023-02-07 株式会社荏原製作所 SUBSTRATE HOLDING DEVICE AND DRIVE RING MANUFACTURING METHOD
CN115632007B (en) * 2022-12-06 2023-02-28 广州粤芯半导体技术有限公司 Method for monitoring wafer flatness

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5664987A (en) * 1994-01-31 1997-09-09 National Semiconductor Corporation Methods and apparatus for control of polishing pad conditioning for wafer planarization
EP0803327A2 (en) * 1996-04-26 1997-10-29 MEMC Electronic Materials, Inc. Apparatus and method for shaping polishing pads
US5872633A (en) * 1996-07-26 1999-02-16 Speedfam Corporation Methods and apparatus for detecting removal of thin film layers during planarization

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4417478A (en) 1981-11-30 1983-11-29 Western Electric Co., Inc. Method for determining lead frame failure modes using acoustic emission and discriminant analysis techniques
US5317901A (en) * 1985-12-10 1994-06-07 United Technologies Corporation Nondestructive test for coated carbon-carbon composite articles
FR2634551B1 (en) * 1988-07-20 1990-11-02 Siderurgie Fse Inst Rech METHOD AND DEVICE FOR IDENTIFYING THE FINISH OF A METAL SURFACE
US4955383A (en) 1988-12-22 1990-09-11 Biofield Corporation Discriminant function analysis method and apparatus for disease diagnosis and screening
US5005205A (en) 1990-01-12 1991-04-02 International Business Machines Corporation Handwriting recognition employing pairwise discriminant measures
US5508915A (en) 1990-09-11 1996-04-16 Exxon Production Research Company Method to combine statistical and engineering techniques for stuck pipe data analysis
US5377451A (en) * 1993-02-23 1995-01-03 Memc Electronic Materials, Inc. Wafer polishing apparatus and method
US5450326A (en) 1994-07-06 1995-09-12 Harris Corporation Graphical display discriminant factor indicator for anomaly identification in semiconductor manufacture batch process
US5762536A (en) * 1996-04-26 1998-06-09 Lam Research Corporation Sensors for a linear polisher
US6293139B1 (en) 1999-11-03 2001-09-25 Memc Electronic Materials, Inc. Method of determining performance characteristics of polishing pads

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5664987A (en) * 1994-01-31 1997-09-09 National Semiconductor Corporation Methods and apparatus for control of polishing pad conditioning for wafer planarization
EP0803327A2 (en) * 1996-04-26 1997-10-29 MEMC Electronic Materials, Inc. Apparatus and method for shaping polishing pads
US5872633A (en) * 1996-07-26 1999-02-16 Speedfam Corporation Methods and apparatus for detecting removal of thin film layers during planarization

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ROTHE H ET AL: "Discrimination of surface properties using BRDF-variance estimators as feature variables", SPECIFICATION AND MEASUREMENT OF OPTICAL SYSTEMS, BERLIN, GERMANY, 14-16 SEPT. 1992, PROCEEDINGS OF THE SPIE - THE INTERNATIONAL SOCIETY FOR OPTICAL ENGINEERING, 1993, USA, PAGE(S) 152 - 162, ISSN: 0277-786X, XP002157653 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6293139B1 (en) 1999-11-03 2001-09-25 Memc Electronic Materials, Inc. Method of determining performance characteristics of polishing pads
WO2019060582A1 (en) * 2017-09-25 2019-03-28 Applied Materials, Inc. Semiconductor fabrication using process control parameter matrix

Also Published As

Publication number Publication date
US6293139B1 (en) 2001-09-25

Similar Documents

Publication Publication Date Title
US6293139B1 (en) Method of determining performance characteristics of polishing pads
US5999003A (en) Intelligent usage of first pass defect data for improved statistical accuracy of wafer level classification
US7365830B2 (en) Wafer flatness evaluation method, wafer flatness evaluation apparatus carrying out the evaluation method, wafer manufacturing method using the evaluation method, wafer quality assurance method using the evaluation method, semiconductor device manufacturing method using the evaluation method and semiconductor device manufacturing method using a wafer evaluated by the evaluation method
US7805258B2 (en) System and method for film stress and curvature gradient mapping for screening problematic wafers
US6393602B1 (en) Method of a comprehensive sequential analysis of the yield losses of semiconductor wafers
US6507800B1 (en) Method for testing semiconductor wafers
US8065109B2 (en) Localized substrate geometry characterization
US20020002415A1 (en) Defect analysis method and process control method
US7042564B2 (en) Wafer inspection methods and an optical inspection tool
US7174281B2 (en) Method for analyzing manufacturing data
CN103714191B (en) 2D/3D analysis for abnormal tools and stage diagnosis
US7386420B2 (en) Data analysis method for integrated circuit process and semiconductor process
US6389366B1 (en) Methods for identifying sources of patterns in processing effects in manufacturing
US7991497B2 (en) Method and system for defect detection in manufacturing integrated circuits
US7062415B2 (en) Parametric outlier detection
US20030170920A1 (en) Method of estimating post-polishing waviness characteristics of a semiconductor wafer
US6503767B2 (en) Process for monitoring a process, planarizing a surface, and for quantifying the results of a planarization process
CN100463136C (en) Chemical mechanical polishing and method for manufacturing semiconductor device using the same
US7330800B1 (en) Method and apparatus for selecting sites for sampling
CN113611626B (en) Method for detecting etching depth of silicon groove on line
US6789033B2 (en) Apparatus and method for characterizing features at small dimensions
US6853873B1 (en) Enhanced throughput of a metrology tool
EP1500996B1 (en) Inspection data analysis program, inspection tools, review apparatus and yield analysis apparatus
US6287173B1 (en) Longer lifetime warm-up wafers for polishing systems
US7220605B1 (en) Selecting dice to test using a yield map

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): CN JP KR SG

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE

121 Ep: the epo has been informed by wipo that ep was designated in this application
DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
122 Ep: pct application non-entry in european phase
NENP Non-entry into the national phase

Ref country code: JP