Affiliation:
1. Electrical and Computer Engineering, Florida International University, University Park Campus , Miami, Florida 33199
Abstract
Ellipsometry is an optical analytical method in which desired reflecting or transmitting surface parameters are related to measurements by mathematical models. Recent work has shown that using artificial intelligence (AI), methods can result in predicting reflecting surface parameters faster and more easily than by using iterative methods. This prior AI work used artificial neural networks applied to a growing absorbing film on a known substrate. Each different substrate required a set of separately trained networks across the wavelength spectrum, thus necessitating training a new set of networks for each new substrate. The work presented here does not require substrate optical property data. Thus, one set of spectroscopic networks can serve a large number of different substrates. This becomes possible by increasing the number of measurements per wavelength from two to three. For now, we consider transparent substrates for which the extinction coefficient of the substrate (k2) is zero or near zero. As before, the noniterative, stable, and fast performance lends itself to real-time, in situ monitoring of thin film growth. Examples for such growth of an absorbing metal film, chromium, will be given using two different substrates. The multilayer perceptron configuration consists of six input and six output neurons with two hidden layers of 80 neurons each. Solutions are performed at each wavelength independently and do not rely on fitting functions for optical properties.