Machine learning enabled development of unexplored perovskite solar cells with high efficiency

Machine learning (ML) is emerging to accelerate the exploration and development of new perovskite solar cells (PSCs). Herein, we use the ML with advanced algorithms to predict five unexplored (FAPbI(3))(x)(MAPbBr(2.8)Cl(0.2))(1-x) perovskites with low bandgaps for experimental guidance. The short circuit current density (J(sc)) and open circuit voltage (V-oc) are also predicated for the five PSCs. Experimentally, the highest power conversion efficiency of 22.5% is achieved for the planar (FAPbI(3))(0.95)(MAPbBr(2.8)Cl(0.2))(0.05) PSCs, where the Jsc, V-oc, and fill factor are 24.6 mA/cm(2), 1.11 V, and 82.4%, respectively. An agreement between the measured and predicated bandgaps is demonstrated with the relative error less than 2%. In addition, the measured J(sc) and Voc values show a consistency with the ML predictions, where the J(sc) value is also independently verified from the optical modelling and simulation. The photocurrent density and the efficiency are further enhanced via the light management strategy by adopting antireflective PDMS nanocone arrays on the top of the planar cell. It is found that the efficiency can be boosted to 23.6% due to the enhanced J(sc) value of 1.2 mA/cm(2). The stability measurements show significantly improved device stability of the present (FAPbI(3))(0.95)(MAPbBr(2.8) Cl-0.2)(0.05) PSCs than the (FAPbI(3))(0.95)(-MAPbBr(3))(0.05) PSCs over 600 h without encapsulation.