Fusion of Physical Mechanism and Data-Driven Methods for Online Thickness Measurement and Error Compensation in SiC CMP

The thickness of silicon carbide (SiC) wafers is a crucial parameter that significantly affects the performance of devices, and its high-precision online measurement during chemical mechanical polishing (CMP) faces challenges from complex process-induced errors. To address this issue, this study develops a non-contact online thickness measurement system based on oblique-incidence laser triangulation and proposes a hierarchical hybrid error compensation method. Deterministic systematic errors caused by optical interference from polishing slurry are first compensated by combining an optical propagation physical model with experimental calibration. Subsequently, a Long Short-Term Memory (LSTM) network model is introduced to compensate for nonlinear, time-series-related dynamic random errors, primarily induced by temperature drift and associated thermal effects. Experimental results indicate that, after applying the proposed compensation method, the root mean square error (RMSE) of the online thickness measurement is 0.47471 μm, and the mean absolute percentage error (MAPE) is 0.1102%. The deviation from reference thickness values is maintained within ±1 μm. The proposed method provides an effective solution for high-precision online thickness measurement and error compensation in the SiC CMP process.