Université de Montréal

Published purpose

This Project will focus on adapting a general learning framework to improvements related to nanophotonic component design. Novel optimization techniques such as inverse design are promising tools to significantly reduce the size of nanophotonic components while maintaining their functionality and performance. Although published results demonstrate various proof of concept miniaturized devices with pre-determined size and aspect ratio, their performances as of now are inferior to the state of the art classical devices which are however much larger, with designs that are highly non-interpretable. In this respect, recent developments in learning generative models through the use of generative-adversarial architecture is a promising technological breakthrough. Specifically, this type of training provides a framework to build neural-network based models capable of generating a diverse set of useful data instances in high-dimensional spaces, which are well-performing designs in this context