Noor Titan Putri Hartono1,Hans Köbler1,Paolo Graniero1,2,Mark Khenkin1,Antonio Abate1
Helmholtz-Zentrum Berlin1,Free Universität Berlin2
Noor Titan Putri Hartono1,Hans Köbler1,Paolo Graniero1,2,Mark Khenkin1,Antonio Abate1
Helmholtz-Zentrum Berlin1,Free Universität Berlin2
Despite the high efficiency, addressing the perovskite solar cells (PSCs) degradation issue remains critical for pushing their levelized cost of energy down and making the technology economically feasible to enter the manufacturing stage. Recent studies point out that the most common lifetime metric for solar cells, the T80 lifetime (the time required to reach 80% of initial performance), in PSCs vary due to the difference in degradation curve shape, due to an initial burn-in loss, an initial sharp efficiency increase, and a combination of both. The exact cause of such variations is unknown, which hinders the mitigation efforts for reducing the degradation in PSCs.<br/><br/>In this study, we utilize machine-learning-based clustering algorithms, including dynamic time warping (DTW) and self-organizing map (SOM), to categorize ~1,400 degradation curves of devices degraded under various environmental stressors with different electron- and hole-transport layer, perovskite composition, and contact materials. We also investigate how the various layers affect the shape of the degradation curves, pointing out the layers responsible for specific degradation curve shape. This will be an important step in finding the optimum layer combination, and finally, mitigating the degradation issue in PSCs.