The CMP process induces complex patterns of radial thickness variation that are typified by local maxima and minima. This nonuniformity can be seen in Figure 1, which depicts pre- and post-CMP thickness data measured on a blanket oxide film. The pre-CMP film is fairly flat with a thickness standard deviation on the order of 1% of the overall film thickness, whereas the post-CMP measurements vary widely. The physics of these polish rate variations have not been satisfactorily explained, but it is generally accepted that they are created by several mechanisms, including the stresses placed on the wafer by the carrier and pad, slurry transport and chemical effects, and the relative motion of the wafer carrier and platen.^3–6

Because CMP processes are typically optimized and controlled based on blanket film polishing results, obtaining accurate and representative measurements on this film layer is essential.^7,8 Most such measurements are made using standard area or polar maps, or a subset or variation of a standard setup. Another popular technique is the line or diameter scan. However, because these methods were primarily designed to handle films with basically monotonic radial thickness variations, such as those generated in deposition and etching processes, they are often inadequate when applied to the significantly higher-order radial thickness variations that are typically obtained in CMP processing.²

Figure 1: Pre- and post-CMP thickness data illustrating the nonuniformity induced by the CMP process.

Figure 2: Representations of 200-mm wafers illustrating some site selection patterns used in CMP metrology recipes: (a) a polar map, (b) an area map, (c) a typical subset of the 49-point polar map used to generate a 13-site measurement for a process qualification, and (d) the proposed radial pattern. All x and y coordinates are in millimeters.

Some examples of typical site selection patterns are given in Figure 2. While they all provide an areally averaged measurement on the global scale, the radial, polar, and area maps are quite different in appearance. In the polar map (Figure 2a), information is gathered only at the center point and along three discrete wafer radii. Although it contains points that seem to be much more evenly distributed than those of the polar map, the area map (Figure 2b) results in information being collected only along seven discrete radii. The modified polar pattern (Figure 2c), which is used for process qualification, suffers from the same lack of comprehensive radial information as the standard polar map and is further compromised by reduced azimuthal information and a bias toward the wafer center.

This article describes an alternative methodology for a radially based, areally averaged site selection scheme (depicted in Figure 2d) that captures the full range of thickness variation patterns typically encountered in CMP processes. Examples of the proposed method's capabilities on wafers that had undergone oxide or metal polishing are compared with those of the traditional site selection methods, and a process optimization case study is presented.

The proposed alternative measurement scheme uses a radial map composed of a number of points that represent equivalent concentric annular areas of the wafer. Only one point is associated with each discrete concentric annulus, which maximizes the amount of radial information that is obtained, and because each point is associated with an equivalent area, the measurement is areally averaged. In addition, the points are rotated around the wafer to maximize the azimuthal resolution as well.

The design for a given site selection scheme using the proposed methodology is generated as follows. For a wafer map consisting of n total points, including one center point, the wafer is divided into n regions of equal area, with the center region being a circle and the remaining regions being annuli, or rings. The area of each region is defined as

Using the design method outlined above, a site map can be generated that accurately captures the full range of radial and azimuthal variation in polish rate behavior while also providing an areally averaged measurement. By collapsing the two-dimensional data onto a line and considering the points as a function of radius, a line scan also can be generated, providing added information about the azimuthal variation.

A series of blanket film measurement experiments were performed at Rodel's Materials Integration Center in Phoenix using either a PM200 Gemini CMP tool from Peter Wolters of America (Plainville, MA) or an IPEC472 from Speedfam-IPEC (Chandler, AZ). Multiple consumable sets and process settings were used in these experiments, and redundant measurements were taken on the same wafer to compare the capabilities of the various site selection methods.

Oxide polishing was done using Rodel Klebosol 1501/50 colloidal silica slurry and IC1000 perforated or K-groove pads; tungsten polishing was done using an experimental variation of Rodel MSW2000 slurry and IC1000 K-groove pads; and copper polishing was done using an experimental abrasiveless slurry and IC1000 K-x,y groove pads. Oxide film measurements were made on an OptiProbe 2600 from Therma-Wave (Fremont, CA), and copper and tungsten measurements were made on a CDE ResMap from Creative Design Engineering (Cupertino, CA).

The polish rate results presented below represent the differences between pre- and postpolish thickness values. Calculated from the polish rate mean and standard deviation, polish rate uniformity performance is expressed as a percent; for example, a polish profile with a mean of 2500 Å/min and a standard deviation of 250 Å/min would be said to have a uniformity of 10%.

Figure 3: Example of the design process for the radial measurement pattern with n = 13. The blue symbols represent the same points as the red symbols aligned on the x-axis, rotated 7.5°.

n 3-mmEdge Exclusion 5 -mmEdge Exclusion 10-mmEdge Exclusion x-axis y-axis x-axis y-axis x-axis y-axis 1 0.00 0.00 0.00 0.00 0.00 0.00 2 0.00 –28.00 0.00 –27.42 0.00 –25.98 3 0.00 39.60 0.00 38.78 0.00 36.74 4 –48.50 0.00 –47.50 0.00 –45.00 0.00 5 56.00 0.00 54.85 0.00 51.96 0.00 6 44.27 –44.27 43.36 –43.36 41.08 –41.08 7 –48.50 48.50 –47.50 47.50 –45.00 45.00 8 52.39 52.39 51.31 51.31 48.61 48.61 9 –56.00 –56.00 –54.85 –54.85 –51.96 –51.96 10 0.00 –84.00 0.00 –82.27 0.00 –77.94 11 88.55 0.00 86.72 0.00 82.16 0.00 12 0.00 92.87 0.00 90.96 0.00 86.17 13 –97.00 0.00 –95.00 0.00 –90.00 0.00

Table II: Uniformity statistics for the center-fast and center-slow polish profiles illustrated in Figure 8.

Oxide Polishing. The optimum polish profile in oxide CMP with a hard carrier and a passive carrier ring on a 200-mm wafer is typified by a relatively flat region in the center of the wafer, a local maximum between the 60- and 80-mm radii, a sharp drop in rate beyond ~90 mm, and a severe increase in rate within a few millimeters of the wafer edge.^1–3,7 This type of polish behavior is particularly difficult to capture with conventional site selection patterns, which either do not measure in the regions where the majority of the polish rate variation occurs or bias the measurement by heavily weighting discrete radii. The problem is exacerbated when a very small edge exclusion is used, since the most severe variations tend to occur close to the edge.

An extreme example of this phenomenon can be seen in Figure 4, which depicts polish rate as a function of radius for four different site selection patterns. Three of the patterns have 49 points, while the area scan includes 52 points; all use a 3-mm edge exclusion. The data were collected using a wafer carrier with a concave profile, yielding a reduction in effective pressure applied to the wafer center and creating a center-slow polish rate. In the figure, two-dimensional data have been collapsed onto lines, which are spline fits with l = 1000 in the cases of the radial map, polar map, and area map, and l = 1 in the case of the line scan.

Figure 4: Polish rate as a function of radius for a single oxide-polished wafer, measured using four different patterns: (a) radial and polar, and (b) area and line scan.

The polar and area patterns, which measure at three discrete radial points plus a center point and seven discrete radial points, respectively, offer significantly less information than the line scan and radial maps. Because of the gaps in their radial coverage, both the polar and area schemes have the potential for significantly misrepresenting polish rate behavior, and, in fact, the polar map does an extremely poor job of capturing the actual behavior in this case. Although the line scan does provide a comprehensive picture of the actual polish behavior with respect to radius, it is not an areally averaged measurement (since the points are spaced at even radial intervals, the information collected at the center of the wafer is weighted more heavily than that at the edge), and it does not provide any azimuthal information. In contrast, the polish profile illustrated in Figure 4 is certainly not optimized, but it does serve to illustrate the divergence of thickness information caused by the site selection pattern.

Some additional insight into the relative biasing effects of three of the site selection methods is presented in Figure 5, which shows the number of points at a given radius as a function of the radius squared, where the square of the radius is proportional to the area. Although all three recipes represent an area-averaged measurement in the global sense, severe local variations in measurement density can be seen in the figure. With the radial map, the measurement points are evenly distributed with respect to wafer area; with the area map, the points are fairly evenly distributed with slight local variations; but with the polar map, the measurement becomes severely biased toward the wafer edge, to the extent that nearly half of the thickness information is gathered at the outermost radii. When coupled with the radial polish rate variations typically observed in CMP, these sampling variations often give rise to data misinterpretations.

Figure 5: Measurement point density as a function of the square of the radius for the area, polar, and radial patterns. (Note that R2 is proportional to wafer area.)

A more realistic process scenario is illustrated in Figure 6, which plots polish rate and uniformity performance as a function of wafer backside pressure. Commonly used to control uniformity in CMP, backside pressure acts to increase the polish rate of the wafer center relative to the edge. The data in the figure were collected using a wafer carrier with a flat profile using the four site selection patterns discussed above. In each data set, the outlier in closest proximity to the wafer notch was excluded from the polar data. Very significant differences between the results for the various patterns can be seen in the figure, especially those for the radial and polar maps. When compared with the radial map data, the polar map data underestimate the polish rate by 7.3–9.9% and overestimate uniformity by 49–112%.

Copper and Tungsten Polishing. In metal polishing, the biasing effects of heavily weighting the wafer edge tend not to be as severe as they are with oxide polishing, since the edge exclusion on metal film measurements is generally larger than the corresponding oxide measurement. Figure 7, which presents data generated in a copper-polishing process as a function of backside pressure, reveals that polar mapping can underestimate the polish rate by as much as 3% and overestimate uniformity by as much as 32%. Similar effects were observed in measurements taken on tungsten-polished wafers. The metal-polishing data did not exhibit the same degree of difference between patterns that were observed in the similar measurements for oxide polishing for two reasons: the maximum polish rate coincided roughly with one of the measurement rings on the polar map, and the edge exclusion for these measurements was set to 10 mm. Nevertheless, some differences in the polish profiles for the various patterns were evident.

Figure 6: Polish rate and uniformity data as a function of backside pressure for oxide-polished wafers, measured using four different patterns.

In addition, future increases in yield requirements will necessitate fuller use of the outer wafer edge. As technology advances and edge effects in metal deposition become less significant, the issues that are evident in oxide polishing will be found in metal polishing as well.

The goal of the optimization project described here was to develop a polishing process that could meet certain strict polish rate and uniformity criteria, including maintaining an upper limit of polish rate uniformity using an edge exclusion of 6 mm or, if possible, 4 mm.

A designed experiment was performed in which wafer down force and backside pressure were varied. Redundant measurements on the same wafer were taken using radial maps with 3-, 4-, and 6-mm edge exclusions and area maps with 4- and 6-mm edge exclusions. Figure 8 illustrates two polish rate profiles—center fast (top) and center slow (bottom)— that were observed in these experiments.

Figure 7: Polish rate and uniformity data as a function of backside pressure for copper-polished wafers, measured using four different patterns.

The data shown represent the profiles with the most extreme differences between the 4- and 6-mm area mapping data, which were most significant for the outermost points. In most cases, the 96-mm ring represents the minimum polish rate over the entire wafer; beyond the 96-mm radius, leading-edge effects are dominant and the polish rate increases significantly. This phenomenon can be seen clearly in the bottom data set in Figure 8, where the polish rate using the radial pattern is greater at 97 mm than at 96 mm. Table II gives the uniformity data for the two polish profiles in Figure 8, along with comparable data measured using the radial pattern and 4- and 6-mm edge exclusions.

When a very small edge exclusion is used on wafers with severe edge effects, even a normally robust site selection pattern such as an area map can yield misleading process optimization information. For example, Figure 9 illustrates response surface plots for uniformity as a function of wafer down force and normalized backside pressure. Although all of the plots indicate that uniformity is optimized at a high down force and low backside pressure, there are some striking differences between the 4-mm radial and area responses and between the 4- and 6-mm responses. In addition, only the 4-mm area plot indicates a local minimum in uniformity at low down force and backside pressure. The scaling of down force and backside pressure is identical on all three plots; the scaling of uniformity is not.

Figure 8: Polish rate as a function of radius for oxide-polished wafers from a process optimization experiment.

Measurement Pattern Profile Type Area,6-mm EdgeExclusion Radial, 6-mm EdgeExclusion Area,4-mm EdgeExclusion Radial,4-mm EdgeExclusion Radial,3-mm EdgeExclusion Center fast 4.86 2.64 12.55 6.01 8.03 Center slow 6.48 6.73 10.29 7.75 8.00

Table I: Example of sites for a radial map for n = 13.

Figure 9: Response surface plots representing model fits to redundant measurements of wafers from a process optimization experiment using three mapping patterns: (a) an area map with a 6-mm edge exclusion and (b) area and radial maps with a 4-mm edge exclusion. The surface represents the model fit of the polish rate standard deviation as a function of wafer down force (DF) and backside pressure (BP).

When the optimized process was run on customer wafers, significant variations resulting from local biasing and site placement effects were observed. The 4-mm area pattern yielded a mean polish rate of 2612 Å/min compared with 2712 Å/min for the 4-mm radial pattern, resulting in an underestimation of polish rate of about 4%. The mean uniformity from the 4-mm area pattern was 9.32% versus 4.93% for the 4-mm radial pattern, resulting in an uniformity overestimation of 108%. The 6-mm radial and area patterns yielded mean polish rates of 2730 and 2692 Å/min, respectively, and mean uniformities of 2.71 and 3.45%, respectively.

Traditional site selection patterns for measuring wafer thickness were not designed to capture the complex radial polish rate variations typically encountered in CMP processing. Severe biasing of the data can result from heavily weighting discrete radii, which often do not coincide with the wafer regions where most thickness variation occurs or, conversely, may overrepresent an area with extreme local thickness variation. A proposed alternative site selection methodology, the radial pattern, has demonstrated an ability to overcome these challenges. The pattern provides an areally averaged data set that accurately and efficiently captures both the radial and azimuthal variations in wafer thickness caused by CMP.

The author would like to thank Regina Cunningham, Gerald Travis, Robert Green, and Peggy McElroy at the Rodel Materials Integration Center for their help in performing the polishing experiments, and Andrey Zagrebelny of Cypress Semiconductor, which provided the wafers. A version of this article was originally published in the proceedings of the Sixth International Chemical Mechanical Planarization for ULSI Multilevel Interconnection Conference (CMP-MIC), held March 7–9, 2001, in Santa Clara, CA.