The advent of 5G cellular networks and the Internet of Things (IoT) technology has significantly upped the demand for high-performance networks. 5G promises greater bandwidth and higher download speeds, up to 10 gigabits per second. IoT is well known as the backbone of what we call the "smart home" devices and appliances including lights, thermostats and home security systems that can be controlled remotely via devices such as smartphones and smart speakers.
The large increase in connectivity demanded by these technologies requires advances in performance from the nanophotonic components that produce and detect the information through communication networks. These advances can only be achieved through innovations in components design. Researchers from the Advanced Electronics and Photonics and Digital Technologies research centres at the National Research Council of Canada (NRC) had their work in this area published in Nature Communications. The article entitled "Mapping the global design space of nanophotonic components using machine learning pattern recognition" demonstrates a machine-learning-based approach to efficiently map and characterize the complex and high-dimensional design space of nanophotonic components.
This new research by the NRC represents a major step forward by providing methods to enable multi- objective design optimization, as well as insights into how designs influence performance.
Together, the two NRC research centres, through the High-throughput and Secure Networks and Artificial Intelligence for Design Challenge programs, developed a novel design approach based on machine learning (ML) dimensionality reduction to solve this problem. Through this process, all favourable design solutions can be characterized by a small set of parameters which are identified using a sparse initial collection of different designs. This compact representation highlights device performance as well as structural differences and limitations. Even when many initial design parameters are involved, this approach allows the designer to obtain insights on the behaviour of the device and a clear understanding of the design space, making the discovery of superior designs based on the relative priorities for a particular application possible.
The research on ML-assisted photonic design was conceived approximately 2 years ago when the Advanced Electronics and Photonics Research Centre's Dr. Dan-Xia Xu explored how ML could be used to help design better photonic components. After many rounds of discussions with her colleagues, including Digital Technologies Research Centre's Dr. Yuri Grinberg, and collaborators from the University of Málaga in Spain, they embarked on a new research journey. Combining expertise in photonic design and artificial intelligence, they determined that the challenges of finding good designs in high-dimensional design spaces, a feature deeply imbedded in the conventional approach to designing components, is exactly suited to the strength of ML.
According to Dr. Daniele Melati, one of the lead authors of the Nature Communications paper, when a new device in photonics is created, "we rely on our theoretical knowledge and intuition to identify the potential structure and design parameter ranges. We then exploit analytical and numerical tools to fix some of the parameters to reasonable values, sweep the remaining ones, simulate the device response, and iterate until we have a good understanding of the general behaviour and the effect that each parameter has on the device. As the complexity of photonics grows, this classical approach becomes impractical. "
This global perspective on high-dimensional design problems represents a major shift in modern nanophotonic design and provides a powerful tool to explore complexity in next-generation devices.