A Modeling Framework and LSTM Nonlinear Dynamics Predictor for Accelerating Data Generation in Ultrafast Optics and Laser System Design

Ultrafast optics and high-power laser systems are the backbone of nearly every major industry from semiconductor manufacturing and telecommunications to advanced medical procedures and next-generation energy and defense solutions. With the increasing integration of machine learning (ML) into laser system design, there is a growing demand for efficient data generation to aid experiment design, perform in-situ optimization, and improve the efficacy of system-wide digital twins. Toward this end, we present a novel start-to-end (S2E) modeling framework for complex laser systems that can be tailored to specific applications for large data generation. The simulation output can then be used in a wide variety of ML tasks from predicting pulse propagation behavior and laser system controls optimization to diagnostic characteristics extraction. However, the models, by necessity, can involve solving several coupled partial and ordinary differential equations including the Nonlinear Schrodinger Equation (NLSE), gain dynamics, and Maxwell’s equations, depending on the framework configuration. These complex cascaded nonlinear systems of equations become a significant time bottleneck in generating large quantities of data. To demonstrate a broad impact application of ML enhancing ultrafast optics simulations, we aim our studies on using long short-term memory (LSTM) networks to replace solving the NLSE for sum-frequency generation, a nonlinear optical process involving the interaction of three fields. We show how these models can provide significant speed-up for large data generation and can ultimately enable an S2E framework to be applied broadly across applications in the ultrafast optics field. Finally, we discuss a metric for assessing the performance of these networks in the context of the optics domain.