The silicon wafers used as substrates for semiconductor devices are required to be of increasingly stringent quality because of the downscaling of these devices. Particularly, the wafer surface quality, including the flatness and the local roughness, is critical in avoiding defects in the final products. These necessary surface perfections are achieved via chemical mechanical planarization (CMP) in the wafer manufacturing process, and repeatedly during both FEOL and BEOL flows to maintain global planarity between unit processes. The nanoparticles of the CMP slurry are important for achieving the desired surface finish because the slurry provides both a chemical effect via surface oxidation and a mechanical effect to remove the surface oxide layer of the work piece. In this study, we focus on the silica particles in CMP slurry. Typically, particle sizes in liquid are measured by Dynamic Light Scattering (DLS). However, there are challenges in using DLS to measure particle size and particle size distributions in CMP slurries because the signal is affected by the concentration of solvent and the surface modification of these particles. In addition, the DLS cannot measure the shape of the particles. Although the SEM and TEM image are used to check the particle shape, it’s difficult to get a quantitative value, nor can they provide information about the particle agglomeration in the liquid. Thus, a combination of methods must be used and the results correlated to obtain a full picture of the size, shape, and distribution of slurry nanoparticles; most importantly these features were connected with the final result, i.e. the surface quality of silicon wafers polished with these slurries.
The SEM images were analyzed by Image J, with background modification and binarization by auto setting of local threshold using the Otsu method, and some overlapped edges were separated using Watershed method. Using this optimized filtering and detection process, it was possible to differentiate the edges of even the particles that were overlapped. Image analysis time was also decreased 1/5 by building a macro for these processes.
The wafer polishing quality (the removal rate and the surface roughness) were compared with key particle shape values (diameter, particle size distribution, area, circularity etc…). The removal rate decreased when the particle circularity approached that of a perfect Circle. It seems that the surface oxidation as a chemical effect by particles decreased because the surface area of particles is lowest for a perfect sphere. However, there weren’t any significant differences found relating the wafer quality and the other particle shape values. More subtle effects based on these parameters might be obtained using a more sophisticated analysis method, such as a multiple regression analysis.
In conclusion, by optimized the processing method using the Image J, the quantitative value of particle parameters could be calculated from SEM images, while also improving analysis efficiency by 1/5. This analysis showed that the polishing behavior was inversely related to circularity, with more circular particles yielding a lower material removal rate than imperfect particles.