Novel Studies in Silicon Dioxide, Copper and Tungsten Chemical Mechanical Planarization Processes Relating to Pad Conditioning and Micro-Texture, Slurry Nanoparticles, Tribology and Kinetics

Abstract

The first part of this study explored the effect of conditioning downforce during pad break-in, and its impact on the evolution of pad surface micro-texture. Results showed that the pad-wafer contact area and contact density decreased with conditioning downforce. Break-in at the higher conditioning downforce helped in reaching faster stabilized values of the analyzed micro-texture parameters. The evolution of these parameters was different for the two different downforces, however as break-in time increased past 30 minutes, mean summit height and mean summit curvature began to approach approximately the same value. These results indicated that, for the particular disc used in this study, a change in the magnitude of downforce during pad break-in caused a change in break-in time and stable values for some micro-texture parameters. However, a change in the magnitude of downforce during break-in only resulted in a change in break-in time for other micro-texture parameters. In the second part of this study, the impact of conditioner types and downforces during pad break-in, and the resulting effects on the evolution of pad surface micro-texture was investigated. All experimental cases resulted in similar trends of mean summit height. More importantly, each case resulted in a different evolution of summit height distribution. Comparing the two discs used, one was observed to be more sensitive to changes in downforce compared to the other. The differences in the behavior of the two discs was explained by the differences in cutting mechanics, which was due to the different characteristics of the two discs. Both discs generated large amounts of pad fragments, which were shown to cause pore obscuration on the pad surface. In 4 out of 5 cases, the pad surface micro-texture stabilized within 30 minutes of break-in and all cases stabilized within 60 minutes. As the third part of this study, the effect of conditioner type and downforce on blanket SiO2 wafer polishing performance was investigated. The goal of this study was to polish wafers after break-in to see how the polishing process behaved immediately after break-in. One of the discs used in this study produced similar micro-texture results at both downforces, which echoed the results seen in the mini-marathon. When comparing the different polishing results obtained from breaking-in the pad with the different discs used in this study, the coefficient of friction (COF) and SiO2 removal rate (RR) were uncorrelated in all cases. However, the use of different discs resulted in different COF and RR trends. The uncorrelated COF and RR, as well as the differing trends, were explained by pad micro-texture results (i.e. the differing amount of fractured, poorly supported pad asperity summits). In the fourth part of this study, the relationship between directivity (Δ) and RR during copper CMP was investigated. The high-frequency shear and normal forces generated by stick-slip (which had been routinely used to explain micro- and nano-scale interactions that led to material removal) were measured. There was found to be a strong correlation between Δ (defined as the ratio of variances in shear force to those of normal force) and copper RR so long as the tribological mechanism remained constant. In cases where the tribological mechanism changed from “boundary lubrication” (BL) to “mixed lubrication” (ML), values of Δ were much lower, yet the straight-line relationship between Δ and RR was maintained, albeit it was shifted significantly. This was due to the ML regime consisting of hydrostatic or buoyant forces supporting the wafer, which led to less variability in frictional forces or less stick-slip events. Additionally, it was found that Δ and RR increased with sliding velocity while in BL due to an increase in stick-slip events. Conversely, Δ and RR decreased at lower sliding velocities while in ML due to an increase in hydrostatic or buoyant force supporting the wafer. In the fifth part of this study, the effect of slurry abrasive nanoparticle (NP) concentration on the tribological, thermal and kinetic attributes of tungsten CMP was investigated. Three tungsten slurries with different amounts of colloidal silica NPs, but without any detectable changes in formulation chemistry, yielded three different polishing outcomes. As slurry nanoparticle concentration [NP] increased, values of COF, temperature and removal rate decreased. This was also true for tungsten to silicon dioxide removal rate selectivity. For tungsten removal, trends in COF, temperature and removal rate were successfully simulated using a two-step modified Langmuir-Hinshelwood model which also yielded values for the chemical and mechanical rate constants. Simulation results indicated that the process was chemically-limited for most polishing conditions, and that the process became even more chemically limited as P × v increased. Moreover, the process became more chemically-limited as [NP] increased. During the extraction of an important parameter for simulation purposes (that being the apparent activation energy), a unique change in curvature of the Arrhenius plot was observed and attributed to the fact that, generally, CVD tungsten films differ in their density, chemical compositions and film morphology as a function of their thickness. The same explanation also accounted for the observed changes in the instantaneous shear forces vs. polish time. In the sixth and final part of this study, fluid film thicknesses were measured and general flow patterns were analyzed on a polishing pad during conditioning. A recently developed UV-enhanced fluorescence experimental technique was used to measure fluid film thicknesses and to analyze how conditioners with different working face designs and polisher kinematics (platen angular velocities) affected fluid flow characteristics on the pad surface. In general, fluid film thicknesses followed the same general trends across the pad surface for both disc designs and platen speeds. Regardless of the parameters used, the fluid film was the thickest in sections nearest to the wafer track and was significantly thinner near the center and edge of the pad. For both discs, the time for film thicknesses to reach steady-state increased with distance from the radius. In general, the full-face conditioner had a smaller maximum attainable fluid thickness (MAFT) and time to reach steady-state (TTRSS) as it most effectively expelled (i.e. squeegeed) the fluid off the pad surface. In contrast, the partial-face conditioner had a larger MAFT and TTRSS as its more intricate design allowed for greater fluid retention and generated more back-flow.