Optothermal Properties of Van der Waals Heterostructures Supported by Machine Learning

Van der Waals (vdW) heterostructures of two-dimensional (2D) transition metal dichalcogenide (TMD) semiconductors show promise for quantum information science including phenomena such as Bose-Einstein Condensation, and quantum light emission. There are limited studies investigating the thermal and opto-thermal properties of semiconducting vdW heterostructures. Deep learning has proven to be useful in spectral analysis, particularly for predicting Raman spectra of TMDs. We present a deep learning methodology for hetero- and homo-structure of TMDs to investigate the thermal properties of moirė superlattices. This work aims to model the optothermal properties of MoS2 and WS2 homostructures and heterostructures by the prediction of single flake Raman spectrums to their superstructure. These spectra are a function of global temperature and laser power, which we can build a prediction of thermal properties such as thermal conductivity and the thermal expansion coefficient. A recurrent convolutional neural network was implemented and paired with metadata knowledge injection techniques to synthesize the spectra. This model accurately predicts Raman shifts in the temperature range of 90-300K, simultaneously enhancing the design selection of moirė superlattices.