Investigating Slurry Transport Beneath a Wafer during Chemical Mechanical Polishing Processes

Abstract

Since its introduction into integrated circuit (IC) manufacturing by IBM Corporation in the mid-1980s, chemical mechanical pla-narization (CMP) has become a key enabling technology in the semiconductor industry. All of the major IC manufacturers including Intel, Motorola, and IBM now incorporate CMP in the production of their chips. In addition to IC production, CMP applications have spread into other manufacturing processes including dynamic ran-dom access memory (DRAM) chips, hard drives, and modem chips. Given the magnitude of capital invested in this technology, there is a large impetus to develop a fundamental understanding of the process. Research with this goal in mind is being performed; how-ever, an overall understanding of the process remains elusive be-cause of the multidisciplinary nature of CMP. Researchers have focused on individual aspects of the process, such as slurry chem-istry, 1-5 wafer-pad dynamics, 6-8 mechanisms, 9-14 and numerical simulations of the slurry fluid mechanics. 15-17 There has been, how-ever, little experimental research regarding slurry fluid mechanics. Several researchers have commented on the importance of slurry flow and slurry distribution beneath the wafer although the impor-tance of slurry flow has not been experimentally demonstrated. Stavreva et al. 18 discussed how the pad's ability to transport slurry could affect the polishing rate and uniformity during copper CMP. Parikh found that slurry flow rate had a large effect on the polishing uniformity on an orbital polisher. 19 Ali et al. 20 stated that the slurry composition, flow rate, and direction of slurry impingement onto the polishing pad all play important roles in interdielectric removal rates. Singer 21 reported that the manner in which slurry is transport-ed from the outside of the wafer to its center is critically important. Sugimoto et al. 22 showed that slurry transport in grooved pads is important in reducing thermal gradients across a wafer. Since pol-ishing rates are temperature dependent, a reduction in thermal gradi-ents across the wafer is believed to reduce the within-wafer-nonuni-formity. Ali et al. 23 postulated that the degradation in the removal rates of pads without conditioning is due to the decrease in the pad's slurry holding capacity. Liang et al. 24 postulated that Cabot's new open cell pads do not need macroscopic surface topography because of the pad's efficiency in channeling the slurry. Despite the fact that slurry flow generally is considered to be an important factor in the CMP process, there has not been an experimental study of the slur-ry flow or a numerical simulation sophisticated enough to examine the slurry behavior under realistic conditions. Slurry transport and mixing could influence the polishing perfor-mance in two ways: (i) transport of polished material and (ii) non-uniform slurry transport. The first mechanism was postulated by Cook in his research on glass polishing. 4 Cook suggested that polishing removal rate is influenced by the transport of polished material away from the wafer's surface, and substrate material is removed only if it is transported away from the wafer or chemically bound in some man-ner. Therefore, slower slurry entrainment rates should yield slower polishing rates. The same physical argument applies to slurry mixing. A reduction in slurry mixing should increase the polishing rate (pro-vided that the slurry is efficiently transported to and from the wafer surface) because polished material is not mixing with the new slurry, resulting in more new slurry delivery to the wafer surface. Nonuniform slurry transport also may influence the polishing un-iformity. Uneven or asymmetrical mixing across the wafer's surface may cause some areas of the wafer to be constantly exposed to a dif-ferent slurry environment than other areas of the wafer. We have noted that our polishing rates correlate with the nonlinear slurry transport across the wafer's surface. That is, the center of the wafer entrains new slurry more slowly than the edges of the wafer and the polishing rates are faster on the edge of the glass compared with the center of the glass. This paper focuses on factors that influence slurry transport beneath a wafer while polishing. Experimental We have created a functional small scale version of an industrial silicon dioxide (oxide) chemical mechanical polishing platform in order to develop a fundamental understanding of the CMP process. Although there are a variety of other polishing configurations in-cluding linear and orbital polishers, the rotary polisher is the least complicated and therefore was adopted as a starting point. Likewise, our setup was built to mimic oxide substrates because it is the most common type of polishing and the simplest in terms of slurry chem-istry and consumable sets. Figure 1 shows an example of typical pol-ishing rates obtained on the tabletop polisher. The polishing sub-strate was BK7 optical glass polished on an IC1000 pad with Cab-O-Sperse SC1 slurry. Note that the polishing rates are approximate-ly 2000 Å/min. Our system consistently has higher removal rates on the edges of the wafer compared to the center of the wafer (edge-fast polishing). Experiments showed that the polishing rates are not affected by the presence of fluorescent dyes. In order to create a 1:2 scale model of an industrial polisher, there were a number of parameters that had to be scaled. These parame-ters include platen diameter, wafer diameter, slurry flow rate, platen speed, head speed, and wafer load. Since the IPEC 472 industrial