In this research, we conducted a series of experiments to investigate the mechanisms of chemical mechanical polishing (CMP) of silicon. Experimental approaches include tribological tests of frictional and lubricating behavior, chemical analysis, and surface characterization. Specifically, the effects of pH in slurry, surface roughness of wafers, and nano-particle size on removal rate were studied. A transmission electron microscope (TEM), a scanning electron microscope (SEM), and x-ray characterization tools were used to study the change of surface structure and chemistry. Experimental results indicate that the removal rate and planarization are dominated by the surface chemistry.

We investigate the post-CMP cleaning process with a tribological approach. A cleaning process involves three components: brush, wafer/disk, fluid undergoing three-body sliding contact between the brush, wafer, and particles from slurry. Having this in mind, we investigated cleaning mechanisms through experimental measurement of friction force and analyzed the contact condition for particle removal. Our investigation leads to the conclusions that the cleaning process is a boundary to elastohydrodynamic lubricating process that involves a constant contact between a brush and the wafer or disk surface. The motion of the brush nodule is such that the surface forces between the brush and workpiece change from an initial adhesion to sliding abrasion. These analysis leads to insight of particle removal mechanisms.

Based on the experimental removal rate data of silicon chemical-mechanical polishing (CMP), we conducted mathematical analysis on effects of surface roughness, asperities, and different length scales on material removal rate. A preliminary physical model was developed enabling consideration of mechanical and chemical removal during the Si CMP process. This model is compared with experimental results obtained through a tabletop polisher. Research results showed that the removal rate decreased with time and eventually became constant. The model has a higher prediction power when the surface roughness is high. However, with a smooth surface, the asperity height does not take a major role in the model. In such, the chemical interactions become important.