Novel Studies in Silicon Dioxide, Tungsten and Shallow-Trench Isolation Chemical Mechanical Planarization Processes Relating to Pad Micro-Texture, Conditioning Disc Type and Age, Tribology, Fluid Dynamics and Kinetics
AuthorMariscal, Juan Cristobal
KeywordsElectronic Materials and Processing
MetadataShow full item record
PublisherThe University of Arizona.
RightsCopyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction, presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
AbstractThe first part of this study explored the effect of conditioner types and downforces during pad break-in on pad surface micro-texture evolution is investigated. In this study, two substantially different discs were employed (i.e. conventional versus CVD-coated), each at two different downforces. Pad samples were extracted throughout the break-in process and their surface micro-topography and pad-wafer contact characteristics were analyzed using confocal microscopy. The two conditioning discs resulted in different evolution paths during break-in. In general, the conventional disc produced more pad “fragments” that ended up getting counted as taller “artificial” asperities as compared to the CVD-coated disc. In contrast, the gentle shaving action of the CVD-coated promoted eventual flattening of the asperity tips. Regardless of the disc type, the mean summit heights decreased and reach stable values as break-in progressed. Compared to the CVD-coated disc, the conventional disc resulted in higher mean summit curvature indicating sharper asperities. In the second part of the study, the effect of different types of conditioners used during tungsten chemical mechanical planarization on frictional, thermal, and kinetic aspects of the process was investigated. Based on a previous work regarding the effect of conditioner type and downforce on the evolution of pad surface micro-texture during break-in, two significantly different discs were employed (i.e. conventional vs. CVD-coated). Mini-marathon style tungsten CMP runs were conducted with each disc type, followed by tungsten polishing at varying combinations of pressures and velocities. Pad samples were extracted before and after the mini-marathon polishing runs for confocal microscopy analyses of their surface micro-texture. Compared to the CVD-coated disc, the more aggressive nature of the conventional disc produced a greater mean summit height, sharper asperities, higher contact density, and more pad fragments (with the latter acting as “artificial pad asperities”). Consequently, the surface micro-texture generated by the conventional disc produced higher values of directivity and removal rates. We found that the mechanical effects were rate-limiting for tungsten removal for both discs. The conventional disc resulted in a removal rate constant that was 24 percent higher than its CVD counterpart owing to its more aggressive nature and the pad surface micro-texture that it caused. Within the third segment of this study, the effects of different types of conditioners (i.e. conventional vs. CVD-coated) on the evolution of frictional, thermal, and kinetic aspects of the tungsten chemical mechanical planarization were investigated. Two types of conditioning discs were used to conduct mini-marathons. Due to its more aggressive nature, the conventional disc was able to result in steady values of coefficient of friction (0.438) and blanket tungsten removal rate (2,530 Å/min) throughout the mini-marathon. In contrast, the CVD-coated disc resulted in a significant decay in coefficient of friction (from 0.440 to 0.373) as the mini-marathon progressed. At the same time, removal rates also dropped from 2,860 to 2,460 Å/min. The decays observed with CVD-coated disc were likely due to its gentle nature and thus in its inability to remove reaction by-products as they got generated during repeated polishing. This hypothesis was confirmed by performing a mini-marathon with a much less chemically active slurry which did not cause any decays in polish metrics. Since mechanical effects were previously found to be rate-limiting, Preston’s equation was able to adequately simulate the removal rates and their trends for each and every wafer polished during the mini-marathons. In the fourth part of this study, a brand-new conventional diamond disc was subjected to 32 hours of wear during which SEM images of certain active diamonds were taken and pad cut rates were measured. ILD wafers were polished before, midway through, and after the wear test (on a brand-new pad) at varying combinations of pressure and velocity. Tungsten wafers were also polished midway through, and after the 32-hour wear. Polishing was accompanied by pad surface topography analysis via confocal microscopy. The disc experienced significant diamond tip micro-wear along with dried slurry accumulation on its substrate causing pad cut rate to drop by a factor of 2. Despite this drastic change, over the duration of the wear test, there were no substantial changes in pad micro-texture, nor ILD or tungsten removal rates indicating the lack of any correlation between pad cut rate and film removal rate during the first 32 hours of wear. The fifth portion of this study explored a novel experimental technique utilizing UV-enhanced fluorescence which was developed and used to measure fluid film thicknesses and general flow patterns during conditioning on a polishing pad. The method was successfully applied to several case studies for analyses of how conditioners with different working face designs (i.e. complex vane, full-face and partial-face), in conjunction with different platen angular velocities, affected fluid transport. In general, for all discs types, fluid across the pad followed similar trends where films were thickest near the wafer track center and thinnest near the pad edge (measurements showed a thickness range of appx. 0.5 to 1.1 mm). For all discs, the time for the film thicknesses to reach steady-state increased in proportion to the distance away from the pad center (times ranged between 12 and 62 sec). The full-face conditioner consistently produced the thinnest films and reached steady-state the fastest. In contrast, the complex vane conditioner created the thickest films and took longest to reach steady-state. The sixth and final part of this study investigated the tribological, thermal and kinetic aspects of SiO2 and Si3N4 polishing on both blanket and patterned wafers for STI CMP. Results showed the absence of anomalous tribological vibrational behaviors thanks to synergies between the colloidal CeO2-based slurry and the application-specific conditioning disc. Pad micro-textural analysis revealed that the conditioner was able to maintain a high-quality pad surface without pore obscuration and with a sufficient number of tall asperities. In all cases, the dominant tribological mechanism was “mixed lubrication”. Removal rates for both SiO2 and Si3N4 processes showed non-Prestonian behavior as both mechanical and chemical factors were at work. However, the Si3N4 process was much more non-Prestonian than SiO2. As expected, Si3N4 polishing resulted in COF values that were approximately one-half of their SiO2 counterparts. This resulted in high values of SiO2:Si3N4 removal rate selectivity. A modified Langmuir-Hinshelwood model was used to simulate removal rates with remarkable accuracy allowing us to extract both chemical and mechanical rate constants, and conclude that the process was designed to be (due to the nature of the slurry used) mechanically-limited for SiO2 and highly chemically-limited for Si3N4. Time traces extracted from patterned wafer polishing showed that COF could indeed be utilized as a real-time indicator for end-point detection. Data on patterned wafers was consistent with the observed COF time traces in that after 6 minutes of polishing, we observed the total removal of SiO2 with a hard stop on Si3N4. End-points reached were also consistent with our blanket wafer polishing data. Regardless of pattern density and pitch, SiO2 removed was not proportional to polish time. This was a result of the low colloidal ceria nano-particle content in the slurry which was explained via a phenomenological model.
Degree ProgramGraduate College