https://doi.org/10.1051/epjap/2016160157
Regular Article
Analysis and detection of an incorrect profile shape in a classical scatterometric process
1
Univ. Lyon, UJM-Saint Etienne, LaHC, 42023 Saint-Etienne, France
2
Univ. Chad, LARTIC, INSTA, BP 6077, N’djamena, Chad
3
Univ. Grenoble, LTM, UMR CNRS 5129, 38054 Grenoble, France
a e-mail: zakisabit@gmail.com
Received:
26
April
2016
Revised:
6
November
2016
Accepted:
18
November
2016
Published online: 9 January 2017
Scatterometry has become an efficient alternative method for subwavelength diffraction grating characterisation in semiconductor industries. It is based on the reconstruction of the periodic surface from its optical response. Ellipsometry seems to be a more powerful technique for optical measurement and neural networks has proved its effectiveness in the inverse scattering problem. However, in all cases, inverse characterisation processing needs a previously defined geometrical model. The aim of this works to study the impact of an incorrect profile shape in the characterization process and to measure the ability to detect it. Two type of neural network will be treated based respectively on a fixed trapezoidal profile and a generic one involving four different shapes. Theoretical results are presented for different simulated samples including several profile shapes. Experimental results are performed on a photoresist grating with a period of 140 nm on silicon substrate.
© EDP Sciences, 2016