Metasurfaces for Diffractive Deep Neural Networks (MetaNN)

The aim of Dr. Saravi's research group is to investigate fundamental and technological aspects of the use of nanostructured metasurfaces as a platform for optical diffractive neural networks in order to transfer them into applications. Such neural networks realize an optical computer by means of successive layers of linear and nonlinear metasurfaces with fixed parameters. This optical system will have an architecture similar to a deep neural network, with its design parameters found by machine learning on a conventional computer.

In the long term, these purely optical neural networks can be used as a replacement for certain computer-based implementations of artificial neural networks, both to increase their computing power and to reduce their power consumption. These are both important prerequisites for the further development of artificial intelligence systems.