Optical interference coating characterization using neural networks

In contrast to 'conventional' algorithms for determining the otpical and geometrical characteristics of interference coatings, we present an alternative approach using artificial intelligent systems for reverse search tasks. The goal is to develop a neural network which is able to distinguish characteristics spectral features of optical thin films, such as specific interference pattern as well as absorption lines or edges. We demonstrate the application of neural networks to determine film thickness, refractive index and surface roughness of a thin film from the interference pattern of the specular reflectance spectrum in the near IR. For simplicity, in this case both film and substrate materials were assumed to be free of absorption losses. Such tasks may be solved using considerably simple neural networks containing up to 52 neurons for 128 spectral points. Current activities include the extension of the method to absorbing thin film system. We particularly emphasize the significance of a mathematical pre-processing of the spectral data in order to keep the network dimensionality as low as possible. Basing on the first result we strongly suggest that neural networks may be successfully applied for fast estimation of thin film thickness, roughness and optical constants and may consequently supply reliable initial values for any subsequent local minimum search procedure.