4H-SiC wafers are more likely to sustain a lower material removal rate (MRR) and severe subsurface damage in conventional chemical mechanical polishing (CMP) methods. To overcome the material removal bottleneck imposed by aqueous chemistry, a high-efficiency polishing of 4H-SiC wafers method by applying reactive nonaqueous fluids to self-sharpening fixed abrasive pads has been proposed in our former research works. Furthermore, to improve the material removal rate and reduce the surface roughness Sa value of 4H-SiC substrates of the Si face, the effect of organic acid, H2O2, and Triton X-100 in nonaqueous slurry on 4H-SiC polishing was investigated. The MRR of 12.83 μm/h and the Sa of 1.45 nm can be obtained by the orthogonally optimized slurry consisting of 3 wt% H2O2, 0.5 wt% Triton X-100 at pH = 3. It is also found that the addition of different levels of oxidant H2O2 and surfactant Triton X-100 components not only increased the MRR of the 4H-SiC substrates of the Si face but also achieved a lower Sa value; in that, the polishing efficiency of the Si side of the 4H-SiC wafers and the surface quality of the 4H-SiC wafers could be effectively improved by the optimization of the polishing slurry.
Acoustic emission (AE) technology has been widely utilized to monitor the SiC wafer lapping process. The root-mean-square (RMS) of the time–domain eigenvalues of the AE signal has a linear relationship with the material removal rate (MRR). However, the existence of background noise severely reduces signal monitoring accuracy. Noise interference often leads to increased RMS deviation and signal distortion. In the study presented in this manuscript, a frequency threshold noise reduction approach was developed by combining and improving wavelet packet noise reduction and spectral subtraction noise reduction techniques. Three groups of SiC lapping experiments were conducted on a fixed abrasive pad, and the lapping acoustic signals were processed using three different noise reduction approaches: frequency threshold, wavelet packet, and spectral subtraction. The results show that the noise reduction method using the frequency threshold is the most effective, with the best coefficient of determination (R2) for the linear fit of the RMS to the MRR.