Optical deep learning nano-profilometry

TL;DR

This paper introduces an optical nano-profilometry method using convolutional neural networks to accurately measure nanostructure dimensions beyond the diffraction limit, validated on various nanostructures.

Contribution

It presents a novel deep learning framework for optical nano-metrology that overcomes diffraction limitations and accurately retrieves nanostructure profiles.

Findings

Validated on three nanostructures with high accuracy

Demonstrated efficiency and generality of the method

Potential to advance optics-based nano-metrology

Abstract

Determining the dimensions of nanostructures is critical to ensuring the maximum performance of many geometry-sensitive nanoscale functional devices. However, accurate metrology at the nanoscale is difficult using optics-based methods due to the diffraction limit. In this article, we propose an optical nano-profilometry framework with convolutional neural networks, which can retrieve deep sub-wavelength geometrical profiles of nanostructures from their optical images or scattering spectra. The generality, efficiency, and accuracy of the proposed framework are validated by performing two different measurements on three distinct nanostructures. We believe this work may catalyze more explorations of optics-based nano-metrology with deep learning.