Morphology reconstruction of buried structures in scanning electron microscopy using gray-level algorithms
Accurate reconstruction of the surface and subsurface morphology is essential for semiconductor metrology, yet conventional scanning electron microscopy (SEM) provides only two-dimensional contrast and relies heavily on expert interpretation. This study presents an integrated framework that combines optimized SEM imaging conditions, physics-based simulation, and gray-level-driven algorithms to enable quantitative reconstruction of both exposed and buried structures in silicon-based devices. Systematic backscattered electron (BSE) imaging experiments from 0.5 to 30 kV reveal how primary beam energy and detector geometry jointly determine spatial resolution, penetration depth, and structural visibility. Low-energy imaging enhances surface-edge definition, whereas higher energies provide deeper material sensitivity at the cost of reduced sharpness. To address the limitations of manual interpretation, a deep learning model based on a ResNet–UNet architecture was trained using synthetic datasets generated from Blender 3D models and Nebula Monte Carlo simulations. The model achieved a mean absolute depth error below 30 nm when benchmarked against white-light interferometry, demonstrating its capability for nondestructive and high-throughput surface profiling. For buried structures, a gray-level interpretation method incorporating atomic number weighting successfully correlated BSE intensity transitions with subsurface composition and depth layering, enabling the semiquantitative reconstruction of tungsten inclusions beneath titanium and silicon layers. An oblique-view stereo imaging strategy further improved depth discrimination, achieving 99% height accuracy for a 30 nm-deep, 20 nm-high tungsten inclusion, 88% for a 40 nm-deep, 45 nm-high inclusion, and 70% for a 70 nm-deep, 60 nm-high inclusion—with accuracy degrading systematically with burial depth due to increased BSE signal attenuation. Together, these approaches establish a scalable image-based pathway for three-dimensional SEM reconstruction, providing practical utility for semiconductor inspection, defect analysis, and advanced process control.