Real-time spectral signal processing for robust thickness monitoring during wafer thinning

Wafer thinning, as a key step in 3D integration using through-silicon via (TSV), highly relies on precise in-situ thickness monitoring to control accurate endpoint and avoid overgrinding or undergrinding. For ultra-thin wafers, conventional contact-based techniques become increasingly limited due to insufficient accuracy and surface damage. Spectral interferometry has been employed for in-situ thickness measurement. However, its application during wafer grinding is challenged by mechanical vibration, cooling flow, silicon debris, and variations in surface conditions, which may lead to incorrect endpoint detection, increasing the risk of wafer scrap. In this study, a compact water-guided optical probe is developed to maintain a stable optical path and reduce contamination from slurry and debris. Meanwhile, a Fourier transform thickness extraction algorithm is combined with a Z-score quality-factor criterion and a dynamic thickness tracking strategy to identify invalid measurements and suppress outliers during continuous thinning. This approach minimizes misjudgments while maintaining a high valid data rate during real-time acquisition. The proposed method was validated over a wide thickness range and achieved accurate endpoint control down to 3 μm. Static comparison with the F50 showed absolute deviations below 0.5 μm within the designed range of 3–400 μm, while valid data rates remained above 80 % across all thinning tasks. Reliable monitoring was further verified for ultra-thin wafers below 10 μm, demonstrating the robustness of the proposed method under practical grinding conditions.