TY - JOUR
T1 - Hierarchical optimization of perovskite solar cell fabrication via step-by-step machine learning
AU - Pu, Yang
AU - Du, Yujiang
AU - Song, Qizhen
AU - Jia, Ning
AU - Dong, Yingchu
AU - Yang, Mingye
AU - Ma, Luyao
AU - Chen, Ruihao
AU - Wang, Hongqiang
AU - Chen, Qi
AU - Bai, Yang
AU - Liu, Zhe
N1 - Publisher Copyright:
© 2025 The Author(s).
PY - 2026/1/21
Y1 - 2026/1/21
N2 - Scalable manufacturing of perovskite solar cells (PSCs) remains challenging due to the need for efficient co-optimization of multiple interdependent process parameters across stepwise fabrication stages. Here, we propose a machine-learning-driven multi-target Bayesian optimization (MTBO) framework that integrates adaptive constraints with hierarchical objective optimization to guide full process optimization. Applied to electrospray deposition, MTBO simultaneously optimized six processing parameters and five target variables related to perovskite absorbers. By incorporating the characterization results of precursor solubility, perovskite crystallinity, and film quality as adaptive constraints, MTBO ensures that optimization stays within feasible manufacturing conditions while hierarchically prioritizing optimization targets of carrier lifetime and film homogeneity aligned with fabrication workflow. Under MTBO-optimized conditions, a champion PSC based on a 1.63 eV FA0.82Cs0.18Pb(I0.86Br0.11Cl0.03)3 perovskite achieved a power conversion efficiency (PCE) of 21.95%. These results demonstrate that MTBO enables autonomous and intelligent optimization across a multi-stage, multi-target fabrication workflow, paving the way toward self-driving laboratories for scalable PSC manufacturing.
AB - Scalable manufacturing of perovskite solar cells (PSCs) remains challenging due to the need for efficient co-optimization of multiple interdependent process parameters across stepwise fabrication stages. Here, we propose a machine-learning-driven multi-target Bayesian optimization (MTBO) framework that integrates adaptive constraints with hierarchical objective optimization to guide full process optimization. Applied to electrospray deposition, MTBO simultaneously optimized six processing parameters and five target variables related to perovskite absorbers. By incorporating the characterization results of precursor solubility, perovskite crystallinity, and film quality as adaptive constraints, MTBO ensures that optimization stays within feasible manufacturing conditions while hierarchically prioritizing optimization targets of carrier lifetime and film homogeneity aligned with fabrication workflow. Under MTBO-optimized conditions, a champion PSC based on a 1.63 eV FA0.82Cs0.18Pb(I0.86Br0.11Cl0.03)3 perovskite achieved a power conversion efficiency (PCE) of 21.95%. These results demonstrate that MTBO enables autonomous and intelligent optimization across a multi-stage, multi-target fabrication workflow, paving the way toward self-driving laboratories for scalable PSC manufacturing.
KW - electrospray deposition
KW - machine learning
KW - multi-target Bayesian optimization
KW - perovskite solar cells
UR - https://www.scopus.com/pages/publications/105026405502
U2 - 10.1016/j.xcrp.2025.103043
DO - 10.1016/j.xcrp.2025.103043
M3 - 文章
AN - SCOPUS:105026405502
SN - 2666-3864
VL - 7
JO - Cell Reports Physical Science
JF - Cell Reports Physical Science
IS - 1
M1 - 103043
ER -