Development of SiGe Indentation Process Control for Gate-All-Around FET Technology Enablement

TL;DR

This paper presents non-destructive, in-line metrology techniques, including scatterometry and x-ray fluorescence combined with machine learning, for accurately characterizing SiGe nanosheet indentation in Gate-All-Around FETs, enabling improved process control.

Contribution

The study introduces a novel combination of optical and X-ray metrology with machine learning for fast, accurate, and non-destructive SiGe indentation measurement in nanosheet transistors.

Findings

Scatterometry and x-ray fluorescence effectively quantify average SiGe indentation.

Machine learning accelerates measurement without full optical modeling.

Spectral interferometry matches well with electron microscopy reference data.

Abstract

Methodologies for characterization of the lateral indentation of silicon-germanium (SiGe) nanosheets using different non-destructive and in-line compatible metrology techniques are presented and discussed. Gate-all-around nanosheet device structures with a total of three sacrificial SiGe sheets were fabricated and different etch process conditions used to induce indent depth variations. Scatterometry with spectral interferometry and x-ray fluorescence in conjunction with advanced interpretation and machine learning algorithms were used to quantify the SiGe indentation. Solutions for two approaches, average indent (represented by a single parameter) as well as sheet-specific indent, are presented. Both scatterometry with spectral interferometry as well as x-ray fluorescence measurements are suitable techniques to quantify the average indent through a single parameter. Furthermore, machine learning algorithms enable a fast solution path by combining x-ray fluorescence difference data with scatterometry spectra, therefore avoiding the need for a full optical model solution. A similar machine learning model approach can be employed for sheet-specific indent monitoring; however, reference data from cross-section transmission electron microscopy image analyses are required for training. It was found that scatterometry with spectral interferometry spectra and a traditional optical model in combination with advanced algorithms can achieve a very good match to sheet-specific reference data.