Determination of transport characteristics of high-temperature THz quantum cascade lasers: numerical simulations and machine learning

We present our recent numerical study, in which the transport simulations are performed using the density matrix model presented in [1]. These results served as a basis for applying different machine learning models to determine the material gain, emission frequency, and current density of a two-well THz quantum cascade laser (QCL) designed for hightemperature operation [2]. In particular, we examine Random Forest (RF) [3], Extreme Gradient Boosting (XGBoost) [4], and Artificial Neural Network (ANN) [5]. RF reduces the variance through ensemble averaging, XGBoost improves predictive performance through gradient-boosted regularization, while ANN architectures employ neural representation learning to capture complex nonlinear relationships. The models were trained by varying the layer widths and external electric bias as the input variables, enabling the prediction of the desired outputs. Combining numerical simulations with machine learning models enables rapid and accurate prediction of key device characteristics, bridging detailed physics-based modelling with efficient data-driven approaches.