A design method of micro-ring resonators accelerated by machine learning for self-injection locked lasers

To accelerate the design of on-chip self-injection locked (SIL) lasers, we propose a fast and accurate design method based on the extended supermode analysis, by developing a dynamic physical model that highlights the critical role of coupling efficiency in micro-ring resonators (MRRs). This method efficiently computes the coupling efficiency, thereby overcoming the computational burden associated with traditional 3D numerical simulations. Further integrating machine learning (ML), we accelerate the process of eigenmode analysis from hours to seconds, achieving over 99 % accuracy compared to analytic results. Our approach demonstrates excellent agreement with 3D finite element method (FEM) across a wide radius range (10 μm–250 μm), covering key dimensions for high-quality factor (high-Q) SIL applications. And we achieve the SIL experimentally. Collectively, this work establishes a new design paradigm that advances the development of high-performance SIL lasers.