Numerical ellipsometry: Artificial intelligence based real-time, in situ process control for absorbing films depositing on virtual substrates including multiple unknown layers

Ellipsometry can be used to determine the optical properties and thicknesses of thin films depositing on virtual substrates by analyzing the change in polarization state of light reflecting from or transmitting through a sample. For growth in chambers, this approach has the advantage of being able to be used in situ using commercially available equipment. Serious challenges arise if the material structure underlying the growing film is composed of multiple layers. Another challenge is that the computation normally takes significant time using conventional iterative solution methods. The work here demonstrates the use of a very fast, sub 4 ms, Artificial Intelligence (AI) method, which is suitable for real-time growth monitoring in which the underlying structure may be unknown and complicated and is treated as a virtual substrate underlying a virtual interface. The method uses a single artificial neural network (ANN) for each wavelength directly based upon previous developments. The method advances what is rapidly being learned about applying AI methods to ellipsometry and may well be leading to an ever-increasing set of methods to compliment and possibly replace some of those currently in use, which depend on numerical techniques. It uses four separate measurements to obtain the film parameters. Examples are given for absorbing films (up to 33 nm) using multilayer perceptron ANN configurations. Predictions are performed at each wavelength independently and do not rely on fitting functions. The design, training, and use of a number of neural networks are presented.