Design, realization and test of InAs nanowire chessboards for Photonic Learning Machines

Machine Learning is a sub-branch of artificial intelligence with the main aim to infer plausible models to relate a set of data and exploit them to make predictions. Complex systems, from which analytical models are hard to derive or numerical procedures are time consuming, are particularly suitable to be optimized or to be described exploiting machine learning. In particular, Artificial neural networks are machines designed to model the way in which human brain perform a particular function of interest, allowing to achieve interesting applications in many fields such as, to cite a few, in biometrics, speech recognition or pattern recognition. Despite the great research efforts and the results obtained in this field, an optical implementation of artificial neural networks for machine learning still represents a promising idea for its parallel, efficient and fast computing capability. Modern machine learning paradigms, like the Extreme Learning Machine, make possible to implement disordered media to realize optical neural networks, exploiting the effect of the multiple scattering at the media surface and modelling it as a single-hidden-layer feed-forward network. The aim of this work is the design,realization and test of semiconductor nanowire metasurfaces which are implemented in a Photonic extreme learning machine for the realization of flat, trainable optical components. In particular, the possibility to realize with this system a flat optical lens able to mantain the polarization state of light is investigated during the work, together with the influence on the device performance of the metasurface features wide parameter space