Silicon carbide (SiC), a wide-bandgap semiconductor, is renowned for its exceptional performance in power electronics and extreme-temperature environments. However, precision low-loss laser slicing of SiC is impeded by energy divergence and crack delamination induced by refractive-index-mismatch interfacial aberrations. This study presents an integrated laser slicing system based on a liquid crystal on silicon spatial light modulator (LCOS-SLM) to address aberration-induced focal elongation and energy inhomogeneity. Through dynamic modulation of the laser wavefront via an inverse ray-tracing algorithm, the system corrects spherical aberrations from refractive index mismatch, thus achieving precise energy concentration at wanted depths. A laser power attenuation model based on interface reflection and the Lambert–Beer law is established to calculate the required laser power at varying processing depths. Experimental results demonstrate that aberration correction reduces focal depth to approximately one-third (from 45 μm to 15 μm) and enhances energy concentration, eliminating multi-layer damage and increasing crack propagation length. Post-correction critical power measurements across depths are consistent with model predictions, with maximum error decreasing from >50% to 8.4%. Verification on a 6-inch N-type SiC ingot shows 90 μm damage thickness, confirming system feasibility for SiC laser slicing. The integrated aberration-correction approach provides a novel solution for high-precision SiC substrate processing.

Accurate quantification of dislocations in heavily doped 4H-SiC substrates is crucial for power semiconductor quality control but remains challenging owing to the extremely low contrast and high background noise in photoluminescence imaging. This study presents a high-accuracy quantitative metrology framework specifically engineered for the precise identification of threading dislocations (TDs) and basal plane dislocations (BPDs) under low signal-to-noise ratio scenarios. Methodologically, we established a high-quality and high-balance benchmark dataset by employing an oriented bounding box-aware balancing strategy to capture the specific geometric morphology of dislocations. A K -fold Cross-Validation ( K -fold) protocol was introduced as a rigorous metrological standard to ensure statistical reproducibility and stability of the measurement results across different wafer regions. Furthermore, we developed a customized You Only Look Once-Dislocation Evaluation architecture that integrates dynamic upsampling and multiscale context modeling to effectively preserve the geometric continuity of curvilinear BPDs and the distinctness of point-like TDs. Experimental results demonstrate that the proposed framework achieves a mean average precision (mAP) @50 of 95.57%, representing a significant leap in the detection reliability for industrial deployment. Ultimately, this research provides a robust, high-fidelity tool for non-destructive defect characterization in Silicon carbide manufacturing, bridging the gap between deep learning applications and rigorous industrial metrology.