July 19, 2022, 3:30 p.m.
Photonic components design: AI opens new avenues
The demand for photonic component miniaturization, increased efficiency and complex functionality requires innovative design approaches. Advances in modeling and computational tools allow physicists and engineers to rely more on numerical methods for design and less on simplified analytical models that are often approximate and limiting. This creates a space for AI and machine learning techniques to assist in the design process by collecting and analyzing data from such models, offering wide ranging benefits.
Several example applications of machine learning to surrogate modeling and optimization developed by different groups will be briefly presented. This will be followed by a more in-depth look at dimensionality reduction methods in design, an approach developed and tested by our group and collaborators. Dimensionality reduction facilitates an efficient exploration of a high-dimensional design space, identifying different performance or merit tradeoffs and the overall limits of the design space. A number of use cases of linear dimensionality reduction in the context of integrated photonics will be discussed, as well as our recent work on leveraging neural-network-based methods demonstrating further improvement.
Yuri Grinberg is an Associate Research Officer in Digital Technologies Research Center, National Research Council of Canada (NRC). He obtained his PhD degree in Computer Science from McGill University in 2014, focusing on theoretical and applied artificial intelligence and machine learning, and held an NSERC Postdoctoral Fellow position in Ottawa Hospital Research Institute until 2015.
His current research interests are development and use of machine learning methods to address problems and advance state-of-the-art in physics and engineering. In particular, in the last four years he has been primarily involved in the use of AI in design of photonic components. He has been the scientific lead of AI-Assisted Photonic Components Design theme within the NRC-developed AI for Design Challenge Program since its inception in 2019. He has coauthored over 40 peer reviewed articles and given numerous invited talks.