TITLE:
Research on Simulating TEM Images Based on Multi-View SEM Images for Semiconductor Materials
AUTHORS:
Lu Li, Liqiang Zhang, Chinhuan Lin
KEYWORDS:
Scanning Electron Microscopy (SEM), TEM Image Simulation, Semiconductor Copper Interconnect Process, Silicon Oxide Trench, Lightweight Deep Learning Model, Microstructure Characterization
JOURNAL NAME:
Journal of Materials Science and Chemical Engineering,
Vol.14 No.10,
October
27,
2026
ABSTRACT: Transmission electron microscopy (TEM) is an indispensable tool for high-precision microstructural characterization of semiconductor materials. Nevertheless, it suffers from complicated sample preparation, high cost, destructive testing that consumes wafers for TEM morphology and structural analysis, and low efficiency. Benefiting from continuous improvement in instrument precision, scanning electron microscopy (SEM) delivers high resolution, supports multi-view observation, and requires simple sample preparation. This paper proposes a TEM-simulation method based on multi-view SEM images (combination of SE 0˚, SE 10˚ and BSE 0˚). Taking silicon oxide trenches etched in the semiconductor copper-interconnect process as the research object, SEM images at different tilt angles and viewing perspectives are acquired. After image preprocessing, a lightweight miniature U-Net supervised mapping AI model is adopted to realize TEM image simulation. The proposed method enables pre-observation of the overall spatial profile of trenches, including internal morphology, dimensions and defects, so as to provide data support for subsequent copper-filling process optimization. Experimental results reveal high consistency between simulated TEM images and actual TEM images in key features, with an average structural similarity index measure (SSIM) of 0.93. This work opens up a new pathway for efficient quality control of semiconductor copper-interconnect processes.