High-throughput, nondestructive assessment of defects in patterned epitaxial films on silicon by machine learning-enabled broadband plasma optical measurements

Historically, haze metrology on KLA-Tencor Surfscan^® unpatterned wafer inspection systems is the preferred inline non-destructive method for ascertaining crystal quality of epitaxial deposited films. However, this metrology is limited to unpatterned blanket wafers. This paper describes a non- destructive inline optical methodology for measuring epitaxial quality of both blanket and patterned wafers using a novel fast turnaround machine learning method that can be applied to patterned and unpatterned substrates by utilizing the background noise obtained during broadband plasma optical defect inspection. This machine learning method is an innovative nuisance filtering algorithm used in inline defect inspection tools, named iDO™ 2.0 (inLine Defect Organizer™). The study showed a promising machine learning approach that repeatably measures low and high defect densities which are consistent with Secco etch data.