Related Experiment Video
Updated: Apr 1, 2026
11:38
Influence of Hybrid Perovskite Fabrication Methods on Film Formation, Electronic Structure, and Solar Cell Performance
Published on: February 27, 2017
19.2K
Machine Learning-Guided Discovery of High-Performance Perovskite Solar Cells via Cluster Analysis and Experimental
Wajeeha Rahman1, Chengquan Zhong2, Jingzi Zhang1
1School of Materials Science and Engineering, Harbin Institute of Technology, Shenzhen, Guangdong 518055, China.
ACS Applied Materials & Interfaces
|March 31, 2026
Summary
We developed an AI framework using machine learning (ML) and scanning electron microscopy (SEM) to discover high-efficiency perovskite solar cells (PSCs). This approach accelerates material discovery by identifying optimal compositions and reducing experimental searches.
Area of Science:
- Materials Science
- Renewable Energy
- Artificial Intelligence
Background:
- Optimizing perovskite solar cells (PSCs) is complex due to high-dimensional composition-property relationships in mixed-cation/halide systems.
- Machine learning (ML) can predict PSC performance, but extracting validated design rules autonomously is challenging.
Purpose of the Study:
- To develop an integrated unsupervised-supervised ML framework for extracting microstructural features from SEM images.
- To accelerate the discovery of high-efficiency PSCs by identifying precise design rules.
Main Methods:
- Utilized Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to link compositional data with morphological descriptors.
- Employed a predictive model achieving 97% accuracy in distinguishing performance clusters.
- Screened compositional spaces guided by the ML platform to identify high-performance regions.
Main Results:
- Identified a high-performance cluster (Cluster 0) with formamidinium-dominant compositions (>85%) and low bromine content (<3%).
- Synthesized a top-ranked AI-generated composition, FA$_{0.94}$Cs$_{0.03}$MA$_{0.03}$Pb(I$_{0.96}$Br$_{0.04}$)$_{3}$, achieving a 22.06% champion efficiency.
- Reduced the experimental search space by three orders of magnitude.
Conclusions:
- The proposed ML framework effectively bridges computational discovery and experimental validation for complex functional materials.
- Composition (formamidinium, methylammonium, cesium) was confirmed as the primary driver of PSC performance.
- This work offers a generalizable tool for data-driven materials design, significantly advancing PSC research.
