National Center for Scientific Research

Published purpose

Integrated photonics, and in particular silicon-based photonics, is a rapidly developing technology showcasing immense potential in fields such as communication, sensing, and ultrafast analog computing. However, the quest for innovative devices that are more compact, efficient, and multifunctional presents new challenges. The high refractive index of silicon facilitates strong light confinement, which leads to compact devices. Yet, this same property means even tiny structural changes, on the order of a few nanometers, can significantly impact device performance. Such nanometer scale variations are strongly influenced by factors such as the feature size, pattern density, and process uniformity across the die or wafer. Controlling the device fabrication to such a degree of precision is pushing the limit of current fabrication technologies. In silicon photonics, this is the dominant cause for device performance degradation. Notably, the performance of fabricated on-chip integrated silicon photonic components and devices deviates from the theory due to imperfect fabrication and other external factors. Using a variety of machine learning approaches that include surrogate models, inverse models and transfer learning, the project will develop methods to correlate the actual shape of the fabricated devices to the optical measurement data. The devices in questions are individual grating couplers and coupler-based optical phased arrays. For the former, the results can be used by various fabrication facilities to perform higher quality process calibrations than what is currently available. For the latter, the resulting models will allow for a significantly faster calibration of a fabricated OPA.