UCLA

A central challenge in contemporary materials and photonics research is understanding how intrinsic materials properties can be optimally combined with nano- or micro-scale structuring to deliver a target functionality. By leveraging subwavelength nanostructures and the intrinsic dispersion of constituent materials, tailored changes in the amplitude and phase of incident wavefronts can be precisely engineered, along with desired spectral characteristics. However, our ability to meet increasing demands in the performance of photonic structures faces roadblocks due to the complexity of the materials and structural design spaces that are currently accessible. Conventional optimization methods, which rely on numerical simulations that solve Maxwell’s equations, have shown remarkable capabilities in designing nanophotonic structures and are now commonly used. However, they can be computationally costly and are often intractable for large-scale designs or high-dimensional design spaces. As a result, data-driven approaches based on machine learning (ML) have been extensively explored in order to tackle challenging photonics design problems. To this end, this work explores the application of various advance deep learning methods for the design and characterization of nanophotonic materials and structures.