Using hybrid deep learning to predict spectral responses of quantum dot-embedded nanoporous thin-film solar cells

Abstract

In this study, we propose an innovative design for nanoporous Si thin film (NPTF) solar cell, seamlessly integrated with semiconducting (CdSe)ZnS Quantum Dots (QDs), without the need for additional metal-dielectric interfaces to attain plasmonic like effects. The intricate network of randomized nano-scaled pores within thin film creates similar enhancement, complemented by QDs inducing excitonic resonances, and amplifying localized electromagnetic field density. To evaluate the spectral responses of the structure we use a supervised trained surrogate model. To train the model, we generate ground truth datasets by solving Maxwell's equations in the design domain and, subsequently, applying charge carrier dynamics model to evaluate the external quantum efficiency (EQE). To predict the spectral response for this stochastic design with randomized pore and QD positions, we feed the ground truth data to a customized Hybrid Deep Learning (HDL) model through in-vitro geometric features fused with dynamic features of QDs. The dynamic features are extracted using an electron dynamics (ED) study. We then evaluate the prediction accuracy of our HDL model. Results show that our designed model can predict absorptivity with an accuracy of R² >  0.96, and EQE with an accuracy of R² >  0.98. This investigation highlights the potential of coupling nanoporous thin film solar cells with QDs, an observed localized enhancement phenomenon, and HDL model to achieve high-performance thin-film solar cells, characterized by improved external quantum efficiency without using metal-dielectric interfaces.