AI-enhanced optical critical dimension metrology for high aspect ratio structures in semiconductor advanced packaging

This paper introduces a new AI-enpowered method for accurately measuring submicron structures with high aspect ratios (HAR) in semiconductor packaging using spectral scatterometry across DUV, visible, and SWIR wavelengths. By optimizing polarization and spectrometer calibration, the method improves spectral signal contrast for precise critical dimension (CD) metrology. An Artificial Neural Network (ANN) tackles phase shift problems for trench spacings near light wavelengths, enabling precise CD measurement. Experiments demonstrate DUV light's proficiency in measuring small CD differences and VIS and SWIR's effectiveness for larger, deeper structures. The DUV system measures HARs up to 10:1 and apertures down to 0.46 μm with accuracy within 3% of Focused Ion Beam/Scanning Electron Microscope (FIB/SEM) comparisons.