Machine learning-based 3D TSV metrology for advanced packaging

We propose a non-destructive method for Through-Silicon Via (TSV) metrology by combining optical spectral measurements with machine learning. Two approaches are introduced for 3D TSV characterization: a model-based algorithm and a data-driven machine learning method. The model-based approach estimates unknown TSV parameters— such as critical dimension (CD), depth, and bottom profile—by comparing measured optical reflectance with simulated results from optical modeling. In contrast, the machine learning approach automatically learns the correlations between TSV parameters and optical responses, eliminating the need for precise structural modeling.