Degradation mechanisms and intelligent lifetime prediction of polishing slurry in single-crystal SiC chemical mechanical polishing

A 16 h CMP service test was established to evaluate slurry degradation during single-crystal SiC polishing. • Slurry reuse caused progressive failure, with reduced MRR and increased surface roughness. • Electrochemistry and XPS revealed weakened oxidation and reduced oxygen-containing reaction layers in aged slurry. • Particle size, zeta potential, and TEM confirmed abrasive refinement, blunting, and poorer dispersion stability. • A BP neural network predicted slurry lifetime, MRR, and surface roughness for slurry health evaluation. Chemical mechanical polishing (CMP) slurries for silicon carbide (SiC) are commonly recirculated to reduce cost and environmental burden due to its extremely low polishing efficiency, rising a challenge to the determination of their lifetime. To address this challenge, this study develops a proactive prediction model for slurry performance and lifetime by integrating failure mechanism analysis with machine learning. During slurry recirculation, the material removal rate (MRR) decreased from ∼1.8 × 10 ³ to ∼1.1 × 10 ³ nm·h⁻¹ , while the surface roughness (Sa) increased from ∼0.12 to ∼0.57 nm. Potentiodynamic polarization, X-ray photoelectron spectroscopy (XPS), and abrasive-state characterization indicated that slurry degradation originates from weakened oxidizing capability together with deteriorated abrasive dispersion and mechanical removal capacity. Considering that slurry lifetime is closely related to the process conditions, the model takes slurry recipe and process parameters as inputs to jointly predict MRR, Sa and lifetime. A back propagation (BP) neural network was employed to establish an eight-input, three-output prediction model because of its ability to capture complex nonlinear relationships among multiple inputs and outputs. The model achieved high predictive accuracy, with all R² values exceeding 0.96, and its robustness was further confirmed by external-condition validation. This work provides a quantitative and transferable framework for slurry lifetime prediction, health monitoring, and process optimization in sustainable SiC CMP. Download: Download high-res image (219KB) Download: Download full-size image