Two-step Machine Learning Assisted Extraction of VCSEL parameters

We propose a Machine Learning (ML) assisted procedure to extract Vertical Cavity Surface Emitting Lasers (VCSELs) parameters from Light-Current (L-I) and S21 curves using a two-step algorithm to ensure high accuracy of the prediction. In the first step, temperature effects are not included and a Deep Neural Network (DNN) is trained on a dataset of 10000 mean-field VCSEL simulations, obtained changing nine temperature-independent parameters. The agent is used to retrieve those parameters from experimental results at a fixed temperature. Secondly, additional nine temperature-dependent parameters are analyzed while keeping as constant the extracted ones and changing the operation temperature. In this way a second dataset of 10000 simulations is created and a new agent in trained to extract those parameters from temperature-dependent L-I and S21 curves.