Ellipsometry is a surface, film, and interface analytical technique based on measuring the change in the polarization state of incident light interacting with a sample. The measured change contains information about the sample. Ellipsometry advantages include being fast, nondestructive, and performable in air. A very useful technique is to solve by numerical iteration using spectral measurements and fitting spectral functions to ɛ (or n) to achieve computational data sufficiency. The artificial intelligence (AI) method presented here achieves data sufficiency by using measurements at two different thicknesses. For either method, there must be a measurable intensity of reflection or transmission, sufficient sensitivity of the desired parameter to the measurements, and mathematical independence expressed as the condition of the equations comprising the applied mathematical model. The focus here is to further develop AI methods of function approximation to perform data processing with the goal of greatly increasing the speed and ease of data processing. Application of rapidly evolving AI methods to ellipsometry is showing a speedup of a factor of nearly 1000 and simplifies the process by eliminating initial estimates and stopping conditions. This qualitative speedup factor differs for sample choices as well as user solving choices and is based on experience and so will vary. This does not obviate the restrictions presented by the physics of the light–material interaction and the associated mathematics. In this work, AI methods are applied to extract the desired reflecting surface parameters from the measurements on homogeneous isotropic materials.
Urban et al. (Tue,) studied this question.