Thin Solid Films 518 (2010) 1994–2000

Contents lists available at ScienceDirect

Thin Solid Films

j ourna l homepage: www.e lsev ie r.com/ locate / ts f

Pad flattening ratio, coefficient of friction and removal rate analysis during silicondioxide chemical mechanical planarization

H. Lee a,⁎, Y. Zhuang a,b, M. Sugiyama e, Y. Seike c, M. Takaoka c, K. Miyachi c, T. Nishiguchi d,H. Kojima d, A. Philipossian a,b

a Department of Chemical and Environmental Engineering, University of Arizona, Tucson, Arizona 85721, USAb Araca Incorporated, Tucson, AZ 85750, USAc Asahi Sunac Corporation, 5050 Asahimae-Cho, Owariasahi, Aichi Pref. 488-8688, Japand Production Engineering Research Laboratory, Hitachi Limited, 292 Yoshida-cho, Totsuka-ku, Yokohama, 244-0817, Japane Department of Chemical Engineering and Materials Science, University of Minnesota, Minneapolis, Minnesota 55455, USA

⁎ Corresponding author. Tel.: +1 520 626 9368; fax:E-mail address: Hyo.sang.Lee@Novellus.com (H. Lee

0040-6090/$ – see front matter © 2009 Elsevier B.V. Adoi:10.1016/j.tsf.2009.07.193

a b s t r a c t

a r t i c l e i n f o

Article history:Received 5 November 2007Received in revised form 17 May 2009Accepted 31 July 2009Available online 13 August 2009

Keywords:Pad flattening ratioCoefficient of frictionRemoval rateChemical mechanical planarization

Pad flattening ratio (PFR) was investigated as a non-destructive pad surface analysis method on the IC1000 K-groove pad during silicon dioxide chemical mechanical planarization. The PFR defines as the ratio of the brightarea to the total image area. A series of marathon polishing runswere performed under ex-situ diamonddisc padconditioning, ex-situ high-pressure micro jet (HPMJ) pad conditioning and no pad conditioningmethods wherePFR analysis was performed to estimate the amount of pad surface flattening or glazing under these conditions.With no conditioning, PFR increased rapidly to 40% indicating severe glazing. With ex-situ diamond disc padconditioning, PFR remained relatively constant below 12%, suggesting a rough and stable pad surface for pol-ishing, whilewith ex-situ HPMJ conditioning it increased gradually and stabilized at the value of about 23%. Real-time analysis of friction force and silicon dioxide removal rate showed a clear correlation among the PFR, thecoefficient of friction and the removal rate during the silicon dioxide chemical mechanical planarization.

© 2009 Elsevier B.V. All rights reserved.

1. Introduction

Chemical mechanical planarization (CMP) has been widely used inthe integrated circuitmanufacturing industry to achieve local and globalsurface planarity through combined chemical and mechanical means.During the CMP process, the slurry reactswith thewafer surface to forma softened surface film which is subsequently removed by mechanicalabrasion among the wafer, the slurry abrasives and the pad surface. Padsurfaceplays a crucial role in thepolishingprocess, andpadconditioningis required to restore the microscopic roughness on the pad surface toachieve constant and uniform material removal rate across the wafer[1–8]. With insufficient conditioning, the pad becomes flattened orglazedwith particles clogging thepores of thepad and forminga layer ofslurry residue andwafer particulates, leading to lowermaterial removalrates and higher defects on the wafer surface [1–9]. On the other hand,conditioning causes wear of the pad such that excessive pad condi-tioning shortens pad life and increases cost of ownership [9]. Therefore,it is necessary to have a technique to monitor the pad surface conditionduring the CMP process to achieve desirable polishing performancewith appropriate amount of pad conditioning. Currently, stylus proõ-filometry, atomic force microscopy (AFM), scanning electron micros-copy (SEM) and white light interferometry (WLI) are used as the

+1 520 621 6048.).

ll rights reserved.

primary techniques to characterize pad surface [1–7,9]. While stylusprofilometry and AFM require tip contactwith the pad surface, SEM andWLI provide non-contact pad surface analysis. However, all these tech-niques are destructive methods that require samples to be cut from thepolishing pad for analysis and thus cannot be easily integrated into thepolishing process to provide real-time pad surface characterization.

In this study, a developed pad flattening ratio (PFR) tool is used toprovide convenient and non-destructive pad surface characterizationduring the CMP process [10]. The principle of operation of the tool andits main constituents are described later in this paper. Fifty thermallygrown silicon dioxide wafers are polished on IC1000 K-groove pad un-der the ex-situ diamond disc pad conditioning, ex-situ high-pressuremicro jet (HPMJ) pad conditioning and no pad conditioning methods.PFR analysis is performed on the pad at multiple intervals during themarathon runs and the results are compared. Real-time friction forceand silicon dioxide removal rate are alsomeasured during the polishingprocess, and the correlation among the PFR, the coefficient of friction(COF) and the silicon dioxide removal rate is investigated.

2. Pad flattening ratio apparatus

The PFR apparatus is a non-destructive pad surface monitoring toolthat takes an image of the pad surface and calculates the degree of thesurface flatness. The working principle of the PFR apparatus is shownin Fig. 1. A fiber-light guide is used to introduce vertical incidence beam

1995H. Lee et al. / Thin Solid Films 518 (2010) 1994–2000

produced from the halogen light source to the polishing pad surface asshown in Fig. 1(a). A rough pad surface results in highly scattered re-flecting light while a flat surface leads to low dispersed reflecting lightas shown in Fig. 1(b) and (c). A macro lens (magnification 0.5×) with acoaxial incident illuminator allows the light that is reflected from thepad surface to be detected by the CCD camera. As the flat portion of thepad surface generates stronger reflecting light signal, it appears brighterin the pad image. The CCD camera sends the pad image to the computerwhere the image is binarized. The computer uses the Adobe Photoshopsoftware to calculate the ratio of the bright area pixels (Abright) to thetotal image area pixels (Atotal) to obtain the PFR as

PFR =Abright

Atotalð1Þ

In the Adobe Photoshop software, the darkest area corresponds tothe gray scale value of zero while the brightest area corresponds to thegray scale value of 255. The mean gray scale value (128) is selected asthe threshold value to determine the PFR. For instance, if the gray scalevalue of the pixel area is equal or above 128, it is counted as a ‘brightarea’, and if it is below128, it is counted as a ‘dark area’. In this study, thePFR value is used to indicate the extent of the pad surface that wasflattened or glazed during the polishing process.

3. Experimental details

All experimentswere performed using a scaled polisher. Details of thepolishing apparatus aswell as its unique ability to acquire real-time shearforce data critical for determining the COF are described elsewhere [11–

Fig. 1. Working principl

13]. IC1000K-groovepads (without subpad)wereused topolish 100-mmsilicon wafers. The surface film of the silicon wafer consisted of 6000 Åthermally grown silicon dioxide. Fujimi PL-4217 slurry containing 12.5%byweight silica abrasiveswasused, and theflowratewas kept constant at80 ml/min. The polishing pressure and the sliding velocity weremaintained at 20.7 kPa and 0.64 m/s, respectively. For each pad, 50 siliconwafers were polished for 2 min each. Between wafer polishes, ex-situdiamond disc pad conditioning, or ex-situ HPMJ pad conditioning, or nopad conditioning was performed to investigate the pad surface underdifferent conditioning methods. A 5-cm diameter, 100-grit diamond discwas used for the ex-situ diamond disc pad conditioning, with a rotationrate of 30 rpmand a sweep rate of 0.33 Hz. The conditioning pressurewas3.5 kPa and the conditioning timewas 2 min. The HPMJ pad conditioningtechnique used a high-pressuremicro jet to spray KOH solution (having apHof ~11) for 10 s at a pressure of 10 MPa onto the polishing pad to cleanthe residual slurry off the pad surface, remove embedded slurry particlesand re-establish pad asperity heights. Details of the HPMJ padconditioning technique can be found elsewhere [14]. The PFR tool tookthe pad surface image after the wafer polishing and the pad flatteningratio was calculated. A 7×8mm2 pad sample under the wafer carriercenter was selected for the PFR analysis as shown in Fig. 2. An air gunattached to the PFR tool was used to dry the selected pad surface beforethe PFR tool took the pad surface image. The pad sample surface was alsoanalyzed by a JEOL JSM-T300 scanning electron microscope (SEM). TheSEM accelerating voltage used was 5 kV.

Prior towafer polishing, the IC1000 K-groove padswere conditionedat the pressure of 3.5 kPa for 30 min using ultra pure water by the 100-grit diamond disc, which rotated at 30 rpm and oscillated at 0.33 Hz.After 30-minute pad seasoning, the PFR tool took the initial pad surface

e of PFR apparatus.

Fig. 2. Pad sample location for PFR analysis.

1996 H. Lee et al. / Thin Solid Films 518 (2010) 1994–2000

image before wafer polishing. The pads were then polished by dummysilicon wafers using Fujimi PL-4217 slurry until the measured shearforcebetween thewafer and thepad stabilized. Thepadswere thenusedto polish 50 silicon wafers under the ex-situ diamond disc pad con-ditioning, ex-situ HPMJ pad conditioning and no pad conditioning, res-pectively. During the polishing process, one thousand shear force datawere collectedevery secondand their valueswere averaged anddividedby the applied normal force to obtain the average COF. The silicondioxide film thickness was measure by a Filmetrics F20V15 optical in-strument before and after polishing to obtain the removal rate.

4. Results and discussions

4.1. PFR analysis

Fig. 3 shows pad surface images and their corresponding pad flat-tening ratios provided by the PFR tool under different pad conditioningmethods during the 50-wafer marathon. It should be noted that thereare no bright areas in the initial pad surface images, and their respectivemeasured PFR values are zero, indicating a rough pad surface beforewafer polishing. Under ex-situ HPMJ pad conditioning and no pad con-ditioning, the percentage of the bright area in the pad surface imagesincrease dramatically with the number of polished wafers. In compar-ison, during ex-situ diamond disc pad conditioning, the percentage ofthe bright area in thepad surface image remains almost unchangedafter31-wafer polishing, and it increases gradually after 40-wafer polishes.The complete sets of the PFR values under different pad conditioningmethods during the 50-wafer polishing process are shown in Fig. 4.Under no pad conditioning, the PFR value increases significantly fromabout 4% after 10-wafer polishing to 27% after 19-wafer polishing andcontinues to increase to about 40% at the endof 50-wafer polishing. Thisindicates the pad surface is glazed rapidly under no pad conditioning.For the ex-situ HPMJ pad conditioning, the PFR values increases grad-ually and stabilizes at about 24%, indicating part of the pad surface isflattened or glazed at the end of 50-wafer polishing. During the first 15-wafer polishing, the PFR values for both the ex-situ HPMJ padconditioning and no pad conditioning increase gradually, suggestingthat the pad asperities collapse from the fresh pad surface under bothconditions. During thefirst 15-waferpolishing, thePFRvalues for theex-situHPMJ pad conditioning are slightly higher than no pad conditioningfor some wafers. This is not expected because the pad asperities arebelieved to be partially re-established by the ex-situ HPMJ pad condi-tioning compared with no pad conditioning. Further study will becarried out to examine the PFR behavior and pad surface at the begin-ning stage of polishing under the ex-situ HPMJ pad conditioning. InFig. 4, under the ex-situ diamond disc pad conditioning, the PFR valueremains below 5% after 37-wafer polishing and increases to about 12%

at theendof 50-waferpolishing, suggestinga relatively roughand stablepad surface during the polishing process.

Fig. 5 shows the SEM images and corresponding pad flatteningratio value under no pad conditioning and the ex-situ HPMJ pad con-ditioning. No pad sample was taken for SEM analysis under the ex-situdiamond disc conditioning. Under no pad conditioning, the SEM imagesindicate that part of the pad surface is glazed after 19-wafer polishingand the corresponding PFR value is about 27%. The extent of glazecontinues to increase with the number of polished wafers as the PFRvalue increases to about 36% at the end of 50-wafer polishing. In com-parison, under the ex-situ pad conditioning, the extent of glaze is lesssignificant and the pad surface topography was similar after 40-waferpolishing. The corresponding PFR value reaches about 24% after 40-wafer polishing and remains relatively stable. The above results showthat the PFR analysis is qualitatively consistent with the SEM analysis.While the correlation between the PFR value and the exact percentageof the pad surface that is flattened or glazed is to be established in thefuture research, current PFR results haveprovided a goodanalysis on theextent of flatness and glazing of the pad surface under different condi-tioning methods.

The above PFR results are consistent with the previous studies onthe pad surface characterization in chemical mechanical planarization[1–9]. Under no pad conditioning, the pad asperities collapse rapidlyand the pores in the pad surface are quickly clogged by the slurryresidues, resulting in a highly flattened or glazed pad surface. In com-parison, the diamond disc pad conditioning regenerates the pad as-perities, removes the slurry residues and breaks up the glazed area,maintaining the appropriate surface roughness for polishing. Whilethe pad surface under the ex-situ HPMJ pad conditioning is not as roughas that under the ex-situ diamond disc pad conditioning, the ex-situHPMJ pad conditioning does provide a rougher pad surface comparedwith nopad conditioning. Future researchwill be carried out to improvethe HPMJ pad conditioning technique to achieve the same pad surfacegenerated by the diamond disc pad conditioning.

4.2. COF analysis

Fig. 6 shows the COF results under different pad conditioningmeth-ods during the 50-wafer marathon. With ex-situ diamond disc con-ditioning, the COF decreases slightly in the initial polishing process andquickly stabilizes after 5-wafer polishing around 0.23. In comparison,with ex-situ HPMJ pad conditioning, the COF decreases significantlyfrom 0.26 to 0.18 after 10-wafer polishing and then stabilizes. This sug-gests that the pad asperities are partially revived under the ex-situHPMJpad conditioning and a relatively stable pad surface is achieved duringpolishing. Experiments were designed to determine pad cut rate byHPMJpad conditioning anddiamonddisc pad conditioning. For this test,only pad conditioning was performed for 7h and then, pad cut ratewas determined by weight loss of pad before and after pad condition-ing. Results indicated that pad cut rate by HPMJ pad conditioning(0.011 µm/min) was approximately four times less than that of dia-mond disc pad conditioning (0. 046 µm/min). Decreasing from 0.26 to0.18, theCOFundernopad conditioning is very close to theex-situHPMJpad conditioning during the first 25-wafer polishing. This is notexpected because the PFR values under no pad conditioning are signif-icantly higher than the ex-situ HPMJ pad conditioning from 17-waferto 25-wafer polishing as shown in Fig. 4. The exact cause needs furtherinvestigation. On the other hand, different from the ex-situ HPMJ padconditioning, the COF under no pad conditioning continues to decreaseafter 25-wafer polishing and fluctuates between 0.11 and 0.13 after 33-wafer polishing. In this wafer number region, the trend of COF is con-sistent with the PFR trend shown in Fig. 4.

Fig. 7 shows the correlation between the COF and the PFR underdifferent pad conditioning methods during the 50-wafer marathon.While the data exhibit certain extent of scattering when the PFR valuesare less than 25%, in general the COF is inversely proportional to the

Fig. 3. Pad surface image and corresponding pad flattening ratio value under different pad conditioning methods during 50-wafer polishing process.

1997H. Lee et al. / Thin Solid Films 518 (2010) 1994–2000

Fig. 4. Pad flattening ratio (PFR) under different pad conditioning methods during 50-wafer polishing process.

Fig. 5. SEM images under no conditioning and the ex-situ HPMJ conditioning methods during 50-wafer polishing process.

Fig. 6. Coefficient of friction (COF) under different pad conditioning methods during 50-wafer polishing process.

1998 H. Lee et al. / Thin Solid Films 518 (2010) 1994–2000

Fig. 7. Correlation between COF and PFR under different pad conditioning methods(threshold value of 256 was used to calculate PFR value).

Fig. 9. Silicon dioxide removal rate under different pad conditioning methods during50-wafer polishing process.

1999H. Lee et al. / Thin Solid Films 518 (2010) 1994–2000

PFR, indicating that the extent of pad flatness and glazing has a sig-nificant effect on the COF during the polishing process. As the padsurface is flattened or glazed, the number of pad asperities makingmechanical contact with the wafer decreases significantly, resulting ina low shear force and coefficient of friction.

In order to investigate the effect of threshold value on PFR value,threshold value was increased to 266 (+10) and decreased to 246(−10) to re-calculate PFR values. And then, the correlation betweenCOF and re-calculated PFR was analyzed and shown in Fig. 8. It is notedthat all data obtained under different pad conditioning methods (HPMJ

Fig. 8. Effect of threshold value on correlation between COF and PFR. (a) Thresholdvalue of 246 was used to calculate PFR value. (b) Threshold value of 266 was used tocalculate PFR value.

pad condition, no conditioning, and diamond disk conditioning) wereshown in graphs without notation. The average PFR values with thre-shold values of 246, 256, and 266 are 13.6, 13.3, and 12.9, respectively. Itsuggests that PFR value is dependent of the threshold value. However,with re-calculated PFR values followed by changed the threshold value(256±10), correlation between COF and PFR did not change signifi-cantly in this study. Further investigation is needed to classify thesensitivity of the threshold value to the PFR value. In this study, 256wasused as threshold value to calculate PFR value.

4.3. Removal rate analysis

Fig. 9 shows the silicon dioxide removal rate results under differ-ent pad conditioningmethods during the 50-wafer marathon. Removalrate behaves similarly to coefficient of friction shown in Fig. 7. Underthe ex-situ diamond disc pad conditioning, the removal rate decreasesslightly in the initial polishing process and quickly stabilizes after 5-wafer polishing around 18 Å/s. In comparison, under the ex-situ HPMJpad conditioning, the removal rate decreases significantly from 20 to10 Å/s after 30-wafer polishing and stabilizes. Under no pad condition-ing, the removal rate decreases more significantly from 20 to 5 Å/s after33-wafer polishing and stabilizes. These results are consistent with theprevious reports that showed a logarithmic decay of removal rate with-out pad conditioning and a stabilized removal rate with diamond discpad conditioning [4,6–8].

Fig. 10 shows the correlation between the silicon dioxide removalrate and the PFR under different pad conditioning methods during the

Fig. 10. Correlation between silicon dioxide removal rate and PFR under different padconditioning methods.

Fig. 11. Correlation between silicon dioxide removal rate and COF under different padconditioning methods.

2000 H. Lee et al. / Thin Solid Films 518 (2010) 1994–2000

50-wafer polishing process. When the PFR value is less than 10%, theremoval rate data are relatively scattered. In this PFR region, the ex-situdiamond disc and HPMJ pad conditioning give a higher removal ratecompared with no pad conditioning, indicating that the extent of padflatness and glazing is not the only factor that can affect removal rate inthis region. Stein et al. pointed out that the replacement of old slurryinside the pores of the pad surface can also affect the removal rate [7].When the PFR values are less than 10%, with the diamond disc andHPMJ pad conditioning, the old slurry inside the pores of the pad surfaceis removed or replaced with fresh slurry while under no pad condi-tioning, the old slurry stays inside the pores and mixes with the freshslurry leading to the drop in removal rate. COF values associated withdiamond disc and HPMJ conditioning are slightly higher than that ofno conditioning, which also contributes to their higher removal ratesobserved in this PFR region. On the other hand, when PFR is larger than10%, removal rate decreases linearly with increasing PFR, indicating theextent of pad flatness and glazing has a dominant effect on the removalrate in this region.

Fig. 11 shows the correlation between the silicon dioxide removalrate and the COF under different pad conditioningmethods during the50-wafer polishing process. The silicon dioxide removal rate increaseslinearly with COF in the range of parameters considered in this paper.This is consistent with previous studies on the interlayer dielectricCMP using different groove pads [13,15].

5. Conclusions

In this study, the PFR tool was used as a non-destructive pad sur-face analysis tool to investigate the pad surface under the ex-situ dia-mond disc conditioning, ex-situHPMJ conditioning and no conditioningschemes during silicon dioxide chemical mechanical planarization. Re-sults showed that PFR increased rapidly to about 40% under no pad

conditioning, indicating the pad was flattened or glazed quickly duringthe polishing process. Under the ex-situ diamond disc pad conditioning,the PFR remained below 5% after 37-wafer polishing and increased toabout 12% at the endof 50-wafer polishing, suggesting a relatively roughand stable pad surface during polishing. PFR under the ex-situ HPMJpad conditioning increased gradually and stabilized at the valueof about23%, generating a surface that was less rough than that generated bythe diamond disc. Real-time friction force wasmeasured during polish-ing and the general trend showed that the coefficient of frictionwas in-versely proportional to the PFR value.When the PFR valuewas less than10%, the ex-situ diamond disc and HPMJ pad conditioning gave a highersilicon dioxide removal rate compared with no pad conditioning, sug-gesting that the extent of pad flatness and glazing was not the onlyfactor to affect the removal rate in this region. When the PFR value waslarger than 10%, the removal rate decreased linearly with the increaseof the PFR, indicating the extent of pad flatness and glazing had a dom-inant effect on the removal rate. A linear relationship was observedbetween the silicon dioxide removal rate and the coefficient of friction,which was consistent with the previous studies on the interlayer di-electric CMP using different groove pads. The PFR technique providesa good qualitative analysis on the extent of flatness and glazing of thepad surface and is consistent with the SEM analysis. In the future re-search, the correlation between the PFR value and the exact percent-age of the pad surface that is flattened or glazed during CMP processeswill be investigated to provide an accurate quantitative pad surfaceanalysis.

Acknowledgement

The authors wish to express their gratitude to the NSF/SRC Engi-neering Research Center for Environmentally Benign SemiconductorManufacturing for the financial support.

References

[1] L.J. Borucki, T. Witelski, C. Please, P.R. Kramer, D. Schwendeman, J. Eng. Math. 50(2004) 1.

[2] J. McGrath, C. Davis, J. Mater. Processing Technol. (2004) 666–153/154.[3] B.J. Hooper, G. Byrne, S. Galligan, J. Mater. Processing Technol. 123 (2002) 107.[4] L. Borucki, J. Eng. Math. 43 (2002) 105.[5] H. Liang, F. Kaufman, R. Sevilla, S. Anjur, Wear 211 (1997) 271.[6] K. Achuthan, J. Curry, M. Lacy, D. Campbell, S.V. Babu, J. Electron. Mater. 25 (1996)

1628.[7] D. Stein, D. Hetherington, M. Dugger, Stout T., J. Electron. Mater. 25 (1996) 1623.[8] W. Li, D.W. Shin, M. Tomozawa, S.P. Murarka, Thin Solid Films 270 (1995) 601.[9] H. Lu, B. Fooks, Y. Obeng, S.Machinski, K.A. Richardson, Mater. Charact. 49 (2002) 35.[10] H. Kojima, T. Nishiguchi (Eds.), CMP Consumables, Proceeding 8th International

Chemical–Mechanical Planarization for ULSI Multilevel Interconnection Confer-ence, Marina Del Ray, U.S.A, February 19–21, 2003, p. 37.

[11] A. Philipossian, E. Mitchell, J. Electrochem. Soc. 150 (2003) G854.[12] A. Philipossian, E. Mitchell, Jpn. J. Appl. Phys. 42 (2003) 7259.[13] A. Philipossian, S. Olsen, Jpn. J. Appl. Phys. 42 (2003) 6371.[14] Y. Seike, D. DeNardis, M. Takaoka, K. Miyachi, A. Philipossian, Jpn. J. Appl. Phys. 44

(2005) 1225.[15] D. Rosales-Yeomans, T. Doi, M. Kinoshita, T. Suzuki, A. Philipossian, J. Electrochem.

Soc. 152 (2005) G62.