Novel Pad Conditioning and Slurry Dispense Methods in Chemical Mechanical Planarization

Abstract

The first part of this study investigates the pad surface generated by conditioning with three different CVD-coated diamond discs and the corollary effect on polishing performance in copper CMP. The discs that were used had significantly different micro-structures with varying degrees of aggressiveness. Confocal microscopy was used to study the pad surface after the polishing experiments had been performed, where the contact area, contact density and surface topography were analyzed. The most aggressive disc generated a pad surface with the most contact area, contact density and the tallest asperities. These parameters decreased as the aggressiveness of the disc decreased. Thermal, tribological, and kinetic aspects of copper polishing were also investigated. The pad surface generated by the most aggressive disc produced the highest material removal rates. However, the pad surface generated by the least aggressive disc produced a slightly elevated coefficient of friction and mean pad temperature when compared to the other pad surfaces, most likely due to fluid suction caused by the glazed pad surface. Analysis of the chemical and mechanical rate constants indicated that this process was chemically limited for all P × V investigated. The second part of this study analyzed the thermal, tribological and kinetic aspects of the new and developing area of cobalt “buff step” CMP. A process-specific combination of consumables and polishing settings were used to investigate the removal of silicon dioxide in order to better characterize the second step of cobalt polishing in middle of the line (MOL) applications, where the overburden of deposited cobalt had already been polished away, and residual cobalt, along with the liner, needed to be completely removed. This was realized by polishing some of the surrounding dielectric in the “buff step”. Our study showed that the removal rate of the oxide and the mean pad temperature increased with increasing P × V, while the coefficient of friction remained relatively constant, indicating a “boundary lubrication” tribological mechanism. A well-established modified two-step Langmuir-Hinshelwood model was used, for the first time for this set of consumables, to simulate the removal rate data, which yielded chemical and mechanical rate constants. These results indicated that the process was mechanically limited for almost all polishing parameters investigated, except for the most elevated P × V. The final part of this study continued work on the cobalt “buff step” for MOL applications by investigating the use of a novel slurry injection system (SIS) developed by our research group. The tests compared the effect of using the SIS versus the point application (PA) method for three different slurry flow rates at constant pressure and velocity. Higher silicon dioxide removal rates were realized by using the SIS for each flow rate, in comparison to those generated by the PA method. For both methods, the removal rate and coefficient of friction increased with increasing flow rate, while the mean pad temperature remained relatively constant. Similar removal rates were measured for the SIS versus PA at different flow rates, indicating that a 25 to 33% reduction in slurry consumption could be realized by implementing the SIS. A subtle yet critical change was made to the two-step Langmuir-Hinshelwood to account for the chemical effects of fresh slurry dilution by residual rinse water and spent slurry. A nearly three-fold reduction in the root mean squared error between the experimental and simulated removal rates was achieved by addressing these chemical effects, while leaving all other optimized parameters constant from the successful simulation of the removal rate data from part two of this study which used the same set of consumables but one constant flow rate and PA method.