Quantitative Predictions of Moisture-Driven Photoemission Dynamics in Metal Halide Perovskites via Machine Learning

Area of Science:

• Materials Science
• Renewable Energy
• Photovoltaics

Background:

• Metal halide perovskites (MHPs) show promise as alternatives to silicon photovoltaics.
• Long-term stability remains a critical barrier to the commercial adoption of MHP devices.
• Operational stressors like moisture significantly impact MHP optoelectronic properties.

Purpose of the Study:

• To investigate the photophysical processes and light emission dynamics of MHPs under varying humidity.
• To develop quantitative models for predicting MHP performance degradation.
• To establish a framework for designing more stable perovskite solar cells.

Main Methods:

• In situ relative humidity-photoluminescence (rH-PL) measurements were conducted on methylammonium lead tribromide and triiodide thin films.
• Recurrent neural networks (RNNs) were trained using humidity and photoluminescence time series data.
• The predictive accuracy of the machine learning models was evaluated over a 4-hour period.

Main Results:

• The study successfully resolved moisture-driven light emission dynamics in MHP thin films.
• Trained RNNs demonstrated the ability to quantitatively predict future light emission trends with an 18% error over 4 hours.
• A correlation between relative humidity and photoluminescence decay was established.

Conclusions:

• The developed machine learning models provide a quantitative framework for predicting MHP stability under moisture stress.
• In situ rH-PL measurements combined with AI forecasting accelerate the understanding of degradation pathways.
• This research facilitates the rational design of stable perovskite solar cells, paving the way for commercialization.