DEVELOPMENT OF ARTIFICIAL NEURAL NETWORKS FOR IN-SITU ELLIPSOMETRY OF FILMS GROWING ON UNKNOWN SUBSTRATES

An in situ ellipsometry method has recently been developed by the authors in which the refractive index and thickness of a growing film can be computed without knowledge of the underlying interface layers and the substrate. The method makes use of five in situ measurements through windows in a deposition system. A ''virtual interface'' is defined by the first measurement made after the growth has settled down following the initial interfacial microstructure such as island formation and interface reactions. A variably damped least squares algorithm computes the optical properties of the growing overlayer film, four thicknesses (of growth past the virtual interface), and the complex reflection coefficients of the virtual interface. However, this algorithm alone is slow and requires initial estimates within a few percent of the solution. Artificial neural networks have been developed to provide very quickly (sub millisecond) initial estimates of sufficient accuracy to improve overall solution speed with the goal of real time operation. Results using several different multilayer perceptron architectures will be presented. Performance using existing in situ data recorded for Ni and Au grown on BK7 substrates are presented.