Determining the Effects of Slurry Surfactant, Abrasive Size, and Abrasive Content on the Tribology and Kinetics of Copper CMP

Transcription

1 Journal of The Electrochemical Society, G299-G /2005/152 4 /G299/6/$7.00 The Electrochemical Society, Inc. Determining the Effects of Slurry Surfactant, Abrasive Size, and Abrasive Content on the Tribology and Kinetics of Copper CMP Z. Li, a, *,z K. Ina, b P. Lefevre, c, ** I. Koshiyama, b and A. Philipossian a, ** a Department of Chemical and Environmental Engineering, University of Arizona, Tucson, Arizona 85721, USA b Fujimi Incorporated, Kagamigahara, Gifu Prefecture , Japan c Fujimi Corporation, Tualatin, Oregon 97062, USA G299 The effects of slurry surfactant, abrasive size, abrasive content, wafer pressure, and sliding velocity on frictional and kinetics attributes of copper chemical mechanical planarization were studied. While abrasive content did not affect the tribological mechanism of the process, abrasive size was shown to be a significant factor. Surfactant-containing formulations were also shown to dramatically reduce coefficient of friction COF. At low pressures and velocities, the removal rate was independent of surfactant content, abrasive diameter, and abrasive concentration, while at high pressures and velocities, surfactant-containing slurries caused an increase in removal rate. Slurries containing a larger abrasive increased removal rate. No correlation was observed between the removal rate and COF. Instead the removal rate was shown to loosely correlate with the variance of the frictional force, thus suggesting that the rapid formation and extinction of the copper oxide layer as captured by the variance of frictional forces i.e., stick-slip was the rate-determining step The Electrochemical Society. DOI: / All rights reserved. Manuscript submitted November 10, 2003; revised manuscript received November 21, Available electronically March 14, Slurry formulation is a critical factor in ensuring chemical mechanical planarization CMP processes with superior yield, manufacturability, cost of ownership, and environmental friendliness. Formulations focused on lowering the coefficient of friction COF, and thus improving pad life, will lead to reduced pad consumption, higher tool availability, and significant capital cost avoidance due to the fact that reinstalling, preparing, and requalifying pads can take away hours of otherwise useful production time from a polishing tool. Furthermore, lowering the COF will help reduce chances of low-k dielectric delamination during copper CMP, which has been identified as one of the major integration challenges for the 70 nm and beyond technologies. 1,2 In addition, reducing abrasive content of the slurry without compromising performance will lead to lower solid wastes and will lessen the burden on copper waste treatment. This study aims to understand the effect of slurry surfactant, abrasive diameter, and abrasive concentration on the frictional and removal rate attributes of copper CMP. Experimental Experiments were performed on a scaled version of a Speedfam- IPEC 472 polisher using Rodel s IC-1000 k-groove pad. Details of the experimental apparatus are described elsewhere. 3,4 A computer synchronized the friction table to the polishing process so that the real-time friction data could be obtained. Copper disks, having a nominal diameter of 100 mm, and a purity of at least 99.99% acted as substrates. A precise balance, with a readability of 0.01 mg, was used to measure the weight of the disk before and after polishing to calculate the removal rate. Previous studies 3,5 have shown the reproducibility in determining the COF and removal rate to be approximately 0.05 and 250 A/min, respectively. Previous studies 6 had shown that copper disks were viable alternatives to copper-deposited wafers for tribological and kinetic investigations. Eight hydrogen peroxide based slurries containing colloidal silica abrasives of varying diameters and contents were tested. One-half of the formulations also contained equal amounts of a proprietary surfactant. Table I summarizes various features of the eight slurry formulations. Prior to data acquisition, the pad was conditioned for 30 min using ultrapure water at a nominal diamond disk pressure of 0.5 psi, * Electrochemical Society Student Member. ** Electrochemical Society Active Member. z zhonglin@ .arizona.edu rotational velocity of 30 rpm, and sweep frequency of 20/min. Pad conditioning was followed by a 5 min pad break-in with the subject slurry. Experiments were conducted with in situ conditioning at the same disk velocity and oscillation frequency as above. Three settings of applied wafer pressures, p 1.5, 2.0, and 2.5 psi and relative pad-wafer velocity, U 0.46, 0.63, and 1.09 m/s were used resulting in nine different combinations of p U. In all cases, the slurry flow rate was kept constant at 80 cm 3 /min. Theoretical Approach Sommerfeld number. To determine the dominant tribological mechanism during CMP, Stribeck curves are presented using a dimensionless grouping of CMP-specific parameters, called the Sommerfeld number shown in Eq. 1 So = U 1 p In the above equation is the slurry viscosity, U is the relative pad-wafer average linear velocity, p is the applied wafer pressure, and is the effective slurry thickness in the pad-wafer region. Determination of U and are fairly straightforward as the latter can be measured experimentally for a given slurry, while the former depends on tool geometry and angular velocities of the wafer and the platen. Previous dual emission laser induced fluorescence DELIF experimental results 7,8 have shown that the slurry film thickness in the pad-wafer region ranged from 20 to 40 µm. This is very close to the value of pad roughness measured in this study by stylus profilometry. This film is considered to distribute the pressure and eliminate the effect caused by different grooves. Therefore the wafer pressure is defined as the applied down force divided by the wafer area. For simplicity, slurry film thickness, which varies only slightly with pressure and velocity, may be assumed to be equivalent to the extent of pad roughness Ra. This approximation resulted in the calculated Sommerfeld number to shift to the right or to the left i.e., increase or decrease with respect to the actual value of Sommerfeld number, however it had no effect on the overall trends of the individual Stribeck curves. The relative standard deviation for surface roughness was less than 10% for all the values measured. Coefficient of friction (COF) and the Stribeck curve. COF is defined as the ratio of shear to normal force

2 G300 Journal of The Electrochemical Society, G299-G Table I. Summary of slurry attributes. Slurry no. Abrasive diameter nm Abrasive content wt % Surfactant I 13 1 Yes II 13 1 No III 13 2 Yes IV 13 2 No V 35 1 Yes VI 35 1 No VII 35 2 Yes VIII 35 2 No COF = F shear 2 F normal The plot of the COF vs. the Sommerfeld number is known as the Stribeck curve and gives direct evidence of the extent of waferslurry-pad contact. When plotting the COF as a function of the Sommerfeld number three major modes of contact can be envisaged. 9 The first mode of contact is boundary lubrication where all solid bodies are in intimate contact with one another and the COF does not depend on the Sommerfeld number. The second mode is partial lubrication where the wafer and the pad are partially contacting each other. As the Stribeck curve transitions from boundary lubrication to partial lubrication, the slope of the line measuring the COF becomes negative. Finally, the hydrodynamic lubrication mode of contact occurs at larger values of the Sommerfeld number where the fluid film layer totally separates the pad and the wafer, and the COF once again becomes independent of Sommerfeld number albeit at a much lower value. For a given Stribeck curve, the average COF represents the arithmetic average of all COF values taken at various Sommerfeld numbers. It is unclear whether the pad and wafer are in intimate contact, fully separated by a hydrodynamic layer, or some combination of the two. Research groups have supported hydrodynamic polishing, and varying degrees of asperity contact Figure 2. Spectrum associated with a constant tap on the upper plate of the friction table with the pad and wafer totally disengaged. Spectral analysis. For a given polishing run, the measured total unidirectional shear force as a function of time can be broken up into two components a mean force component and a fluctuating force component as shown in Eq. 3 below F shear t = F + f t 3 Figure 1a is an example of the total force measurement at a sampling frequency of 1000 s obtained during a 1 s interval of a typical polishing run. For a 75 s polishing experiment, a total of 75 such plots are generated and analyzed for tribological attributes. The mean force, F, which represents the average of all 75,000 data points, is used in calculating the COF as defined in Eq. 2 and utilized in a previous report. 17 The COF, which is then used to construct Stribeck curves, is therefore totally independent of the fluctuations observed in Fig. 1a. For spectral analysis, the measured total unidirectional shear force function which includes the fluctuating component is converted into frequency domain via fast Fourier transformation FFT. 18 Figure 1b is an example of this transformation where the x axis represents signal frequency in Hertz and the y axis is an indication of the amplitude of the transformed function. In Fig. 1b, the area underneath the curve is the basis of a new parameter termed the interfacial interaction index denoted by. In the frequency domain, it is important to understand the cause or causes behind the presence of a particular peak at a particular frequency. For instance, Fig. 1b shows the presence of several peaks ranging in frequencies from 1 to 120 Hz. To determine whether the frequencies at which the peaks were observed were results of the resonance of the polisher or the result of wafer-slurry-pad interaction, a simple experiment was performed. With the wafer completely disengaged from the pad, the top plate of the friction table was tapped once with a constant force thus causing the polisher to vibrate. By recording the total unidirectional force, and by converting it into frequency domain, the intrinsic resonance of the tool could be quantified see Fig. 2 and compared to that of Fig. 1b. The spectrum in Fig. 2 clearly indicates that, in the absence of any pad-wafer interaction, the tool resonates at a dominant frequency of around 9-10 Hz, with multiple harmonic peaks occurring at frequency ranges of Hz as well as Hz. There is also a fundamental peak at about 60 Hz. By comparing this spectrum to the one shown in Fig. 1b, it becomes clear that the fundamental peak at 9 Hz along with its harmonic peaks at 18 and 36 Hz are direct results of the intrinsic resonance of the tool. On the other hand, peaks occurring around 1-3, 50-55, and Hz all seem to stem from pad-slurrywafer interactions. The lower frequency peak is believed to be due to the kinematics of the process including the rotational velocities of the platen, wafer, and the diamond conditioner which have been calculated to result in frequencies in the range of 0.5 to 2.5 Hz. The observed medium to high frequency peaks are unique to CMP processes using k-grooved pads and seems to correlate to the distribution of collision events between grooves and the leading edge of the wafer which have been calculated to range between 20 and 500 Hz. From Fig. 1b, is determined empirically based on transformed Figure 1. Shear force measured duringa1spolishing interval a, left and its associated spectrum b, right.

3 Journal of The Electrochemical Society, G299-G G301 Figure 3. Stribeck curves for slurries with 13 nm abrasives left and 35 nm abrasives right. data that are in the frequency domain, and it is essentially a measure of the range of forces encountered during polish. On a qualitative basis, the area under the curve in Fig. 1b can be envisaged to represent the total amount of mechanical energy caused by stick-slip phenomena. It must be noted that random variations in friction force measurements do not constitute stick-slip. When studying the tribology of CMP, stick-slip refers to a cyclic fluctuation in the magnitudes of frictional force and relative velocity between the wafer and the pad. It is usually associated with a relaxation oscillation that depends on a decrease of the COF with increasing sliding velocity. True stick-slip, in which each cycle consists of a stage of actual stick followed by a stage of overshoot i.e., slip, requires that the kinetic COF i.e., the parameter being measured in this study is lower than the static COF i.e., corresponding to the maximum friction force that must be overcome to initiate macroscopic motion between the wafer and pad. In this study, the above criterion is certainly met. Another form of stick-slip can be due to spatial periodicity of the friction coefficient along the path of contact i.e., pad grooves or microtrenches created on the surface of the pad due to the conditioner, or complex films forming and being abraded on the surface of the wafer. While further studies are underway to distinguish among various types of stick-slip occurring during CMP, this study combines all stick-slip phenomena into one entity represented by. Results and Discussion Stribeck curves. Figure 3 shows that regardless of abrasive diameter, changes in abrasive content in percent by weight have no effect on the tribological mechanism of the process. In the case of 13 nm abrasives, the dominant tribological mechanism is that of partial lubrication with the surfactant-containing slurries showing a slight tendency to transition to boundary lubrication at low Sommerfeld numbers. In the case of 35 nm abrasives, the dominant tribological mechanism is that of boundary lubrication. The significant changes in the tribological attributes of the process as a result of abrasive diameter are believed to be due to gross differences in abrasive concentrations i.e., number of particles per unit volume of slurry associated with the two types of abrasives and their weight contents in the slurry. Assuming spherical abrasives and following the calculation methodology set forth in Ref. 19, slurries with 1 and 2 wt % of 13 nm abrasives contain 5510 and particles per cubic centimeter, respectively, while those with 1 and 2 wt % of 35 nm abrasives contain only 2050 and 4100 particles per cubic centimeter, respectively. For a given abrasive content, the significantly higher number of 13 nm abrasives aid in the sliding action of the copper wafer against the polishing pad. Tests are underway to verify the above postulation by performing experiments under controlled abrasive concentration conditions. If verified, this would indeed be consistent with recently reported results where varying concentrations of fumed silica slurry were used to intentionally alter the tribological attributes of the interlevel dielectric ILD CMP processes. 13 Figure 4. Average value and total range of COF as a function of surfactant content, abrasive diameter, and abrasive content at low left and high right values of p U. COF and the removal rate. Ongoing work 5 has shown that copper CMP is Prestonian 20 at low values of p U, and non- Prestonian at higher values of p U where thermal effects dominate over mechanical effects. Therefore, the frictional and kinetic attributes of the eight slurries adopted in this study are best described in terms of the COF and material removal rate at the lowest and highest values of p U. Figure 4 summarizes the main effects of surfactant, abrasive diameter, and abrasive concentration on the COF. Regardless of the value of p U or the values of other parameters, surfactant-containing formulations reduce the average COF. This is due to the enhanced lubricating nature of the fatty components of the surfactant. Regarding abrasive diameter, at low values of p U, the COF remains unaffected i.e., 0.47 to 0.49 by an increase in abrasive diameter, however at high values of p U, the COF decreases to below 0.40 with decreasing abrasive diameter. This is likely due to the individual or combined effects of two phenomena as described below. By taking into account the contact model proposed by Brown 21 A c C 1/3 1/3 0 D p 4 where A c is the contact area between the wafer and the abrasive particle, C 0 is the abrasive concentration, and D p is the abrasive particle diameter. According this model, smaller particles have a larger surface area than larger particles, and a higher abrasive concentration also will increase contact area. For a given abrasive content, the abrasive concentrations associated with smaller diameter particles is 2.7 times more than larger diameter abrasives. Therefore smaller particles always get a larger contact area. When the contact area is large, the wafer is more likely to make contact with abrasive particles floating between wafer and pad interface and will not interact with the abrasive particles adsorbed on the pad surface. Therefore, at high velocities, both abrasive larger concentration and small

4 G302 Journal of The Electrochemical Society, G299-G Figure 6. Dependence of removal rate on COF for all experiments performed in this study. Figure 5. Average value and total range of removal rate as a function of surfactant content, abrasive diameter, and abrasive content at low left and high right values of p U. diameter will aid the lubrication process, thus reducing shear force. In all cases, changing the abrasive content of the slurry has minimal impact on the COF thus suggesting that in the absence of other issues i.e., interactions between abrasive concentration and other factors of interest, lower abrasives content formulation maybe favored. Figure 5 summarizes the main effects of surfactant, abrasive diameter, and abrasive concentration on the copper removal rate. Given a reproducibility of about 250 A/min for the copper removal rate, results indicate that, at low values of p U, the removal rate is more or less independent of surfactant content, abrasive diameter, and abrasive content. On the other hand, at high values of p U, surfactant-containing slurries cause an increase in the removal rate, and slurries containing larger abrasive increase removal rate. In spite of the significantly lower values of the COF observed with the surfactant-containing slurries, the corresponding values of copper removal at high values of p U, are significantly higher. This indicates that rate enhancement due to the surfactant is most likely chemical, rather than mechanical, in nature thus providing a pathway for increasing pad life and removal rate simultaneously. Also, the trends observed regarding the effect of abrasive content on material removal indicates an environmental benefit associated with some of the slurries tested in this study. The observation that slurries containing larger diameter abrasives increase the removal rate is currently under investigation. It is interesting to note that for all combinations of pressure and velocity, when all eight slurry formulations are taken into account, no correlation is observed between the removal rate and the COF see Fig. 6. This is in stark contrast to ILD polishing where the COF and removal rate are strongly correlated, 3 thus suggesting that factors other than the COF need to be considered in designing slurries with optimal copper removal rate characteristics. Spectral analysis and the interfacial interaction index. Figure 7 shows the values of corresponding to slurries I, II, V, and VI. Slurries I and II have identical values of abrasive content and abrasive diameter, however slurry I contains a surfactant, while slurry II does not. Similar attributes as above also apply to slurries V and VI. In addition, slurries I and II have higher abrasive concentrations compared to slurries V and VI. Results indicate that surfactantcontaining slurries are have lower values of compared to their unsurfactonated counterparts. One interesting characteristic of Fig. 7 is that slurries I and II have lower values of compared to slurries V and VI, respectively. This may be due to the significant difference in abrasive particle concentrations between the two pairs since high abrasive concentrations can reduce stick-slip by enhancing the lubricity of the system and by lowering the mean and variance of the frictional force. Figure 8 summarizes the main effects of surfactant, abrasive diameter, and abrasive concentration on. Results indicate increases, on average, by nearly two orders of magnitude at high, compared to low, values of p U. This is intuitive since higher values of p U indicate a greater amount of energy in the pad-wafer interface due to higher shaft work and hence a greater extent of stick-slip phenomena. In fact, Figure 7. Spectral comparison of various slurry formulations at 2.0 psi and 1.09 m/s.

5 Journal of The Electrochemical Society, G299-G G303 Figure 10. Spectral comparison of ILD and copper polish processes. Figure 8. Average value and total range of as a function of surfactant content, abrasive diameter, and abrasive content at low left and high right values of p U. for all combinations of pressure and velocity, when all eight slurry formulations are taken into account, the removal rate and seem to loosely correlate to one another as seen in Fig. 9. The observation that for copper polish, unlike ILD polish, the removal rate does not correlate to COF but rather to, suggests that it is the variance in the measured shear force, rather than its magnitude that drives copper removal. Further comparison of the force spectra between ILD and copper polish help shed light on the differences between these two processes. Figure 10 compares typical force spectra corresponding to a copper polish process 60 cm 3 /min of slurry VI at 2 wt % colloidal silica abrasive content, at a pressure Figure 9. Dependence of removal rate on for all experiments performed in this study. of 2.0 psi and a sliding velocity of 1.09 m/s with an ILD polish process 80 cm 3 /min of Fujimi PL-4217 slurry with 12.5 wt % fumed silica abrasive content, at a pressure of 4.0 psi and a sliding velocity of 0.31 m/s. In both cases, Rodel s IC-1000 pad was employed. The particular choices of abrasive content, sliding velocity, slurry flow rate, and wafer pressure were made to ensure a nearly constant value of between the two processes 4.33 for the ILD CMP process and 4.63 for the copper CMP process. Analysis of the spectra indicates that in spite of roughly equivalent values of between the two systems, the energy caused by hydrodynamic chattering associated with the copper polish process is shifted to higher frequencies in the range of 37 to 43 Hz compared to that of the ILD polish process. As reported previously, 9 and as it may be the case here, the higher frequency data appear to be due to chemical phenomena unique to copper CMP. These phenomena are most likely the rapid formation and abrasion of the copper oxide passivation layer during polish and manifest themselves in a series of cyclical processes which tend to cause a high-frequency periodicity in the coefficient of friction and hence its variance during polishing. Assuming this hypothesis to be true, the data suggest that the periodicity of the passivation film formation and abrasion for the family of Fujimi slurries to range between 10 ms in the case of Fig. 1b where the highest frequency peak was centered around 100 Hz and 25 ms in the case of Fig. 10 where the highest frequency is around 40 Hz. Based on the analysis above, it is likely that the loose correlation between removal rate and as reported in Fig. 9 is indeed governed by the rapid formation and extinction of the copper oxide layer as captured by the variance of the frictional attributes of the process associated with this phenomenon. Conclusions This paper reports on the effects of slurry surfactant, abrasive size, abrasive content, wafer pressure, and sliding velocity on frictional and kinetics attributes of copper CMP. While abrasive content did not affect the tribological mechanism of the process, abrasive size was shown to be a significant factor. Surfactant-containing formulations were also shown to dramatically reduce the coefficient of friction. At low pressures and velocities, the removal rate was inde-

6 G304 Journal of The Electrochemical Society, G299-G pendent of surfactant content, abrasive diameter, and abrasive concentration while at high pressures and velocities surfactantcontaining slurries caused an increase in removal rate, while slurries containing larger abrasives increased the removal rate. No correlation was observed between the removal rate and the COF. Instead the removal rate was shown to loosely correlate with the variance of the frictional force, thus suggesting that the rapid formation and extinction of the copper oxide layer as captured by the variance of frictional forces i.e., stick-slip was the rate-determining step. Acknowledgments The authors express their gratitude to the NSF/SRC Engineering Research Center for Environmentally Benign Semiconductor Manufacturing for their financial support. Special thanks are also due to Nikko Materials for their generous donation of the copper disk used in the experiments. The University of Arizona assisted in meeting the publication costs of this article. References 1. K. Cheemalapati, A. Chowdhury, V. Duvvuru, and Y. Li, in Proceedings of the CMP-MIC, Marina Del Rey, CA F. Sun, R. Zhou, M. Kason, and R. Martinez, in Proceedings of the CMP-MIC, Marina Del Rey, CA A. Philipossian and S. Olsen, Jpn. J. Appl. Phys., Part 1, 42, A. Philipossian and E. Mitchell, Jpn. J. Appl. Phys., Part 1, 42, Z. Li, L. Borucki, and A. Philipossian, J. Electrochem. Soc., 151, G Z. Li, S. Radar, P. Lefevre, K. Ina, and A. Philipossian, Abstract 899, The Electrochemical Society Meeting Abstracts, Vol , Orlando, FL, Oct 12-16, J. Coppeta, M.S. Thesis, Tufts University, Medford, MA J. Lu, C. Rogers, V. P. Manno, and A. Philipossian, J. Electrochem. Soc., 151, G K. Ludema, Friction, Wear, Lubrication: A Textbook in Tribology, CRC Press, Inc., Boca Raton, FL L. Cook, J. Non-Cryst. Solids, 120, S. Runnels and L. Eyman, J. Electrochem. Soc., 141, S. Soares, D. Baselt, J. Black, K. Jungling, and W. Stowell, Appl. Opt., 33, W. Tseng and Y. Wang, J. Electrochem. Soc., 142, L C Liu, B Dai, W Tseng, and C Yeh, J. Electrochem. Soc., 143, L. Cook, J. Wang, D. James, and A. Sethuraman, Semicond. Int., 11, R. Bhushan, R Rouse, and J. Lukens, J. Electrochem. Soc., 142, D. DeNardis, J. Sorooshian, M. Habiro, C. Rogers, and A. Philipossian, Jpn. J. Appl. Phys., Part 1, 42, E. Brigham and H. Oren, The Fast Fourier Transform and Its Applications, Prentice-Hall Inc., Inglewood Cliffs, NJ E. Paul and A. Philipossian, in Proceedings of the CMP-MIC Conference, Marina Del Rey, CA F. Preston, J. Soc. Glass Technol., 11, N. Brown, C. Baker, and R. Maney, SPIE Proc., 1981, 340.