Surrogate-assisted machine learning framework for performance enhancement of Rb2CuSbF6 perovskite LEDs with MoO3/WO3 charge transport layers

Machine learning has become an impressive method to speed up the design and performance prediction of optoelectronic devices, by making the exploration of efficient high-dimensional and highly nonlinear parameter spaces. In this paper, we propose a surrogate-assisted machine learning model to train a performance enhancement of lead-free Rb 2 CuSbF 6 perovskite light-emitting diodes (PeLEDs) using MoO 3 and WO 3 as hole- and electron-transport layer, respectively. This is done by physics-based simulations of devices to undertake a systematic parametric sweep of fourteen important material and device parameters producing over 11,000 samples and provides the variation in four important performance measures indicating electrical and optical efficiency. Several machine learning regression models are developed and tested to build a good surrogate model that is sufficiently accurate to predict the nonlinear interactions among all input parameters and the performance of the device. The trained surrogate model is subsequently used for rapid prediction and design-space screening across the multidimensional parameter space, enabling identification of high-performing combinations of transport-layer properties at substantially lower computational cost than repeated physics-based simulations. The suggested framework has a strong predictive potential and rapid prediction action plan in the design and development of perovskite LEDs. This paper shows how surrogate-assisted machine learning solutions can be an efficient and scaled route to creating high-efficiency, environmentally common-place lead-free perovskite light-emitting devices.