Dynamic Response Prediction of Bi-State Emission of Quantum Dot Lasers Based on Machine Learning

Dual-state emission is a common and important phenomenon which takes place in semiconductor Quantum Dot Lasers at different temperature and operating conditions usually investigated from microscopic carrier interaction modeling or even rate-equations based approaches. In this study, we revisit the topic, but the investigation is here performed from a system identification perspective; we built black-box models based on artificial neural networks approach, using the Multilayer Perceptron, the Extreme Learning Machine and a hybrid Echo State Network - Extreme Learning Machine. As a case study, we focused on switch-on transient and its prediction. The study revealed the model was able to separate and to predict, from the solely total power, without using any QDL design parameters, the optical power around the ground state and first excited state lasing lines of InAs/InGaAs quantum dot laser. The error performance was low as a RMSE of 2.81 μW and MAPE of 0.50% with processing time (training and testing time) of 15.27 s, enabling the alternative model to be used in optical filtering instrumentation as low-resolution and low-cost filters for applications in which it is not economically viable to use a spectrum analyzer, which can be replaced by a simple optical power meter.