Deep learning-based defect detection using large FOV SEM for 28 nm pitch BEOL layer patterned with 0.33NA single exposure EUV

Abstract

As we strive toward smaller and smaller pitches and new 3D constructs to enable device scaling, thorough defect characterization at wafer scale is essential during the early phases of process optimization. Often CD (critical dimension) variations and roughness lead to high SEM (scanning electron microscope) inspection noise. It is important to suppress this noise and increase DOI (defect of interest) SNR (signal-to-noise ratio) for better detection efficiency while maintaining high speed for meaningful wafer coverage. In this work, we describe experiments and show characterization results for capturing EUV (extreme ultraviolet) stochastic defects across various test structures of 28nm pitch devices that have been patterned using single exposure 0.33NA EUV lithography. We have used KLA eSL10 (TM) for SEM inspection and analysis. The tool can also be used for high resolution and high-speed metrology, providing quick feedback on observed defect signatures and further root cause analysis.