Pad effects on slurry transport beneath a wafer during polishing Coppeta α, J., Racz χ, L., Philipossian δ,a., Kaufman ε, F., Rogers β, C., Affiliations: α= Research assistant, Tufts University, Department of Mechanical Engineering, Medford, MA. β = Associate Professor, Tufts University, Department of Mechanical Engineering, Medford, MA. χ = Assistant Professor, Tufts University, Department of Mechanical Engineering, Medford, MA. δ= Technology Manager, Intel Corporation, Santa Clara, CA. ε= CMP Engineer Fellow, Cabot Corporation, Aurora, Illinois Abstract: In this paper we study the effects of pad conditioning on slurry entrainment and mixing beneath a wafer. We examined three cases; polishing with an Embossed Politex pad, an unconditioned IC1000 pad, and a conditioned IC1000 pad. Cab-O-Sperse SC1 slurry was used in a 1:1.5 dilution with water. Mixing data shows that conditioning has a negligible effect on the rate of slurry entrainment and mixing, however conditioning has a large effect on the thickness of the slurry layer between the wafer and pad. The conditioned pad shows a marked increase in the slurry thickness as opposed to an unconditioned pad. That is, the conditioning moves more slurry under the wafer but the new slurry mixes with the old slurry at a rate independent of conditioning. In addition the gradients in slurry age beneath the wafer were compared among the three cases. The IC1000 pads supported a gradient in the inner third of the wafer only, while the Embossed Politex pad showed a linear gradient across the wafer implying it retains pockets of unmixed slurry in the embossed topography.
Introduction Little is known about what is occurring beneath a wafer during Chemical Mechanical Planarization (CMP) processes. Both chemical and mechanical mechanisms have been identified as contributing factors to the overall removal phenomenon. However, the relative contribution of each is still poorly understood and requires an understanding of the slurry transport beneath the wafer. Fluid behavior beneath the wafer dictates the polishing mechanism; either by transporting the chemical component or by allowing the pad to abraid the wafer 1-3. Factors such as fluid film thickness, slurry entrainment processes, slurry age and pad slurry carrying capacity are some of the fluid properties that are likely to influence polishing 4-5. We are studying these variables to develop a more complete understanding of the polishing process. In particular, for the purposes of this paper, we will be focusing on how pad conditioning affects the fluid flow beneath a wafer. Experimental Set Up Figure 1 shows our modified CMP set up used to study the slurry flow beneath a wafer. A tabletop Struers RotoPol-31 polisher is used to rotate a 12" polishing pad. The standard RotoPol head has been replaced with a 20" industrial rated drill press that both rotates and applies down force to a 3" wafer. Since we are measuring fluid parameters using an optical technique known as dual emission laser induced fluorescence or DELIF, the wafer must be transparent and a pure silicon wafer cannot be used. Instead, a glass wafer that is transparent to visible light is used. Two high-resolution spatially aligned 12 bit digital cameras are used to measure the fluorescence beneath the wafer. Weighted Traverse Drill Press Two Aligned 12 Bit Camera Struers RotoPol-31 100 RPM Three Way Solenoid Valve Tagged Slurry Slurry Figure 1. Experimental set up In situ or ex situ pad conditioning can be performed. A 3" diamond grit wafer corotating with the platen conditions the pad by periodically sweeping across the pad radius. Many of the polishing parameters including platen speed, down force, slurry
delivery, and conditioning speeds are computer controlled and monitored. In addition, the computer synchronizes the camera to the polishing process so that we can interrogate the wafer at any point in the mixing process. Technique To investigate the flow field, we use a non-intrusive optical technique known as DELIF 6. This technique uses the fluorescence from two different dyes each fluorescing at different wavelengths to measure mixing or temperature. The fluorescence from one dye contains the mixing, laser distribution, and film thickness information while the fluorescence from the second dye contains the laser distribution and film thickness information. Normalizing the fluorescence of the first dye by the fluorescence from the second dye causes the resulting ratio to be a function of mixing only. Figure 2 shows an example of how the technique works. Figure 2 (a) shows the blue (left) and green (right) fluorescence images of the liquid between a glass wafer and an Embossed Politex pad. Since the fluorescence intensity is proportional to the thickness of the liquid layer, the dark areas of the image represent the high points of the embossed pad while the bright areas represent the valleys of the pad. 2a. 2b. 2c. Figure 2. (a) Top view of blue and green fluorescence respectively on an Embossed Politex pad. (b) Intensity graph of ratio values from dividing the blue image by the green image. (c) Intensity values (left) and their respective ratio values (right) for an arbitrary cross-section of the fluorescence images.
Figure 2 (b) is an intensity plot of the ratio values of the blue image divided by the green image. Note that all variations in the intensity due to the pad topography are eliminated through normalization thus allowing us to measure slurry age on any topography. Figure 2 (c) shows an arbitrary row of the intensity values in each image (right) with their respective ratios values (right). The ratio values are constant within 5% even though the intensity values of each image vary by as much as 40%. This property of normalization allows us to monitor the dye concentration gradients or mixing while neglecting thickness variations in the slurry due to the pad and pad wafer dynamics. Our previous work 7 has shown that by injecting tagged commercially available slurry with two fluorescent dyes, slurry mixing, entrainment beneath the wafer, residence time on the pad, and film thickness can be measured. For the purposes of this paper, we are concerned with the slurry mixing only. The major limitation to the accuracy of this technique on our CMP set up is the IC1000 s inherent fluorescence. That is, under laser illumination the IC1000 pad fluoresces over wavelengths that overlap with the dye fluorescence we are trying to measure. Rodel has manufactured a dyed IC1000 pad, which reduces but does not eliminate the pad fluorescence. This background noise increases the uncertainty in our measurements to about 6%, which is shown in detail in the results section. Experimental Design Parameter Embossed Politex Pad IC1000 Pad Platen Speed 60 RPM 60 RPM Head Speed 60 RPM 60 RPM Down Force 4 psi 4 psi Head Position 12 o clock, 3.25 from pad center 12 o clock, 3.25 from center Injection Position 6 o clock, 1.25 from pad Pad center center Dye Concentration ~30 ppm ~30 ppm Injection Flow Rate 50ml/min 50 ml/min Conditioning NA None or 163 micron diamond grit Slurry Cab-O-Sperse SC1 in a 1:1.5 dilution with water Cab-O-Sperse SC1 in a 1:1.5 dilution with water We examine two types of pads for their entrainment and mixing properties. The first is the Embossed Politex pad. Although this pad is of little importance to oxide
polishing, it is the optimal material to study with our technique because it does not fluoresce. The second pad we studied is a flat IC1000 pad. This pad was studied both before and after conditioning with diamond grit. The parameters of both experiments are summarized in the table above. Conditioning for the IC1000 pad was performed using a 3 diamond grit wafer available from TBW Industries. The grit size used is 163 microns. The conditioning protocol involves sweeping the diamond grit wafer across the rotating pad while injecting slurry at a constant rate of 50 ml/min. The wafer is rotating about its own axis at 100 RPM while oscillating across the pad 8 times a minute. The pad is rotating at 60 RPM during conditioning. The downforce on the conditioning wafer is approximately 6 pounds. The cameras focused on a 1 X 0.5 interrogation region selected in the front center of the wafer. Exposure times were approximately 750 milliseconds. The cameras are capable of acquiring and saving a picture every 5 seconds. Therefore a time history of the mixing is constructed by repeatedly injecting new (-tagged) slurry onto the pad and acquiring images at different points during the mixing process. The duration of each pulse of tagged slurry was set to 15 seconds. Results Figure 3 (right) shows the entrainment characteristics of each pad in terms of the spatially averaged percentage of new slurry entrained in the interrogation region versus time. That is, what percentage of the total slurry under the wafer is tagged. On the left, a schematic of the relative component locations is included. The graph illustrates several important points. First the Politex pad pulls slurry under the wafer faster than either the conditioned or unconditioned IC1000 pads. This is due to the fact that slurry is carried in the valleys of the embossed pad, which prevents efficient mixing with the surrounding slurry. Therefore, a faster build up of new slurry beneath the wafer is observed as new unmixed slurry is carried beneath the wafer in the pad valleys. This entrainment mechanism is likely very different from a flat IC1000 pad that would spread and mix new slurry introduced onto the pad. Figure 3 also shows that the performed conditioning does not have a large effect on either the entrainment time or the amount of mixing with most differences falling within the uncertainty of the measurement. In addition, the accuracy of the technique on a Politex pad versus the accuracy on an IC1000 pad is evident. The Politex results are smoother and mixing points for a similar point in time after a pulse are closer together. Between run results in figure 3 differ by no more than 2% for the Politex case whereas the IC1000 results differ by about 4%. The standard deviation of each normalized image is 4% for the Politex pad and 6% for the IC1000 pad.
Wafer Interrogation Region Injection Position Polishing Pad Figure 3. Percentage of new slurry entrained beneath the wafer versus time. Even though conditioning does not affect the mixing characteristics of the IC1000 pad, the thickness of the slurry layer shows a strong dependence on conditioning. Figure 4 shows a top view of the fluorescence from a wafer subsection of both the conditioned and unconditioned pad. The dark circular region at the top of the image is the mounting post to the glass wafer. Fluorescence intensity is proportional to the brightness of the image shown. The fluorescence intensity increases by a factor of two after conditioning, all other parameters being equal. This correlates to a thicker slurry layer being carried beneath the wafer. We are currently working to correlate the fluorescence intensity into a slurry film thickness. Figure 4. Fluorescence images of the conditioned versus unconditioned pad Finally, it is interesting to observe the different slurry age gradients beneath the wafer that develop with different pad types. Figure 5 shows how the slurry composition changes across the wafer at a particular point in time. That is, an arbitrary line across the wafer (shown left) was selected from the interrogation region and analyzed along with the same line in each of the other two cases. This vertical line on the wafer coincides with the radial line of the pad radius. The y-axis is calculated by subtracting the mean slurry age
from the slurry age at each point along the line. Therefore, all zero values on the plot would indicate all of the slurry beneath the wafer is the same age or contains the same percentage of new fluid. The data was smoothed for clarity. The two IC1000 lines were taken 12 seconds after the beginning of a pulse of new slurry while the Politex pad was taken 8.1 seconds after the beginning of a pulse. These times were chosen to correspond to the maximum observed gradient in each case. As shown in figure 5, the IC1000 maintains a composition gradient only in the inner third of the interrogation region while the embossed Politex pad supports a linear composition fluctuation across the entire interrogation region. 20 10 IC1000 No Cond. IC1000 Cond. Embossed Politex 0 Wafer -10-20 0 100 200 300 400 500 Position (arbitary units) Figure 5. Slurry composition (age) gradient across wafer versus position Conclusion In conclusion, we demonstrated that the performed diamond conditioning has a negligible effect on slurry mixing within the uncertainty of the measurement. Conditioning did increase the intensity of the fluorescence by a factor of two indicating that conditioning increases the thickness of the slurry layer between the pad and wafer. We are currently working towards quantifying this thickness. Finally, we examined the slurry age or compositional gradients beneath the wafer and found that there were significant differences between the Embossed Politex pad and the IC1000 pad. The IC1000 pad showed a gradient in the slurry age (or percentage of new slurry) only in the inner third of the wafer, with the rest of the wafer containing a relatively constant slurry age. The Embossed Politex pad showed a linear gradient across the wafer. The difference between these two cases is explained in terms of the Politex pad mixing mechanism. The Politex pad traps fluid in the valleys of the pad, which reduces mixing of fluid from one part of the pad with fluid from another part of the pad. The IC1000 pad can trap fluid in much small pores on a much smaller scale and so this effect will not dominate the mixing and gradients will tend to be smeared out and reduced. This is why conditioning does not have a large effect on the IC1000 mixing characteristics.
Acknowledgements The authors would like to thank Intel and Cabot Corporations for their funding and support. Bibliography 1. Levert, J., Mess, F., Grote, L., Mykola, D., Cook, L., Danyluk, S., Slurry Film Thickness Measurements in Float and Semi-permeable and Permeable Polishing Geometries. Proceedings of the International Tribology Conference, Yokohama Japan 1995. 2. Cook, L., Chemical Processes in Glass Polishing. Journal of Non-Crystalline Solids, 120, 1990, pp.152-171. 3. Levert, J., Baker, R., Mess, F., Salant, R., Danyluk, S., Mechanisms of Chemical Mechanical Polishing of SiO 2 Dielectric on Integrated Circuits. Proceedings of the Annual Meeting of the Society of Tribologists and Lubrication Engineers - London, 1997. 4. Stavreva, Z., Zeidler, D., Plotner, M., Drescher, K., Characteristics in Chemical Mechanical Polishing of Copper: Comparison of Polishing Pads. Applied Surface Science, 108, (1997), pp. 39-44. 5. Iqbal, A., Sudipto, R., Pad Conditioning in Interlayer Dielectric CMP. Solid State Technology, June, 1997, pp.185-187. 6. Coppeta, J., Rogers, C., Fluorescence Imaging Normalization for Direct Planar Scalar Behavior Measurements. Experiments in Fluids. Accepted, publication date pending. 7. Coppeta, J., Rogers, C., Philipossian, A., Kaufman, F., B., Characterizing Slurry Flow During CMP Using Laser Induced Fluorescence. Proceedings of the Second International Chemical-Mechanical Polish for ULSI Multilevel Interconnection Conference, Santa Clara, Ca. February 1997.