posted on 2024-01-30, 19:33authored byZiyi Liu, Jiyun Zhang, Gaofeng Rao, Zijian Peng, Yixing Huang, Simon Arnold, Bowen Liu, Can Deng, Chen Li, Heng Li, Hanxiang Zhi, Zhi Zhang, Wenke Zhou, Jens Hauch, Chaoyi Yan, Christoph J. Brabec, Yicheng Zhao
Obtaining highly stable metal-halide perovskites is crucial
for
the commercialization of perovskite solar cells. However, current
methods for evaluating perovskite stability mainly rely on a time-consuming
and resource-intensive aging process. Here, we demonstrate a spectral
learning-based methodology that enables the prediction of perovskite
stability by leveraging the features in photoluminescence and absorption
spectra of fresh perovskite films. This methodology circumvents the
long-term aging process by combining a custom-developed spectral feature
extraction algorithm and an integrated voting machine learning model.
By integration of the early diagnosis program with high-throughput
experiments, the prediction accuracy for stable perovskites exceeds
86% in 160 fresh samples. The universality is further examined by
another batch of 224 fresh samples fabricated through different processing
conditions. Finally, the early diagnosis of perovskite films is successfully
translated to enhanced stability in perovskite solar cells. Our work
provides a new pathway to accelerate the discovery and development
of stable perovskite films.