In a recent study published in the journal npj Computational Materials, researchers investigated the application of molecular engineering and machine learning (ML) to enhance the stability and efficacy of halide perovskite materials. Their objective was to introduce a comprehensive methodology for designing and assessing molecule-modified perovskite materials capable of withstanding aqueous environments and producing high photocurrents.
a Summary plot of the feature based on the extra trees model using SHAP. b Feature importance ranking histogram based on the extra-trees model using SHAP. Image Credit: https://www.nature.com/articles/s41524-024-01297-4
Background
Halide perovskites, with the general formula ABX3 where X is a halogen atom, exhibit remarkable optoelectronic characteristics, including high light absorption, adjustable band gap, long carrier lifetime, and minimal defect density. These properties make them attractive for diverse energy conversion and storage applications, such as solar cells, light-emitting diodes, water-splitting systems, photocatalysts, and environmental sensors.
However, their susceptibility to environmental factors, particularly water and oxygen, poses a significant challenge as it can compromise their structure and functionality. Therefore, developing strategies to enhance the stability of halide perovskites is a key challenge for their practical implementation.
About the Research
In this article, the authors aimed to optimize and understand the molecular influence on the aqueous photoelectrochemical stability of methylammonium lead iodide (CH3NH3PbI3), also known as MAPbI3 perovskite films. They employed a combination of experimental techniques and data-driven approaches to screen and analyze various molecules capable of modifying the perovskite surface and interface.
These molecules encompassed solvents, additives, and post-treatment dyes, affecting the crystallization, morphology, and passivation of perovskite films. Photocurrent measurements of the molecule-modified perovskite films in aqueous solution under light illumination served as indicators of optoelectronic stability.
The study utilized two ML methods to decouple molecular contributions and provide a scientific interpretation of photoelectrochemical data. The first method employed the extremely randomized trees (extra-trees) algorithm, an ensemble learning technique aggregating results from multiple decision trees. The second method relied on the genetic programming algorithm, an evolutionary technique searching for optimal mathematical expressions.
Additionally, the researchers employed the SHapley additive explanations (SHAP) method, which is based on cooperative game theory, to determine feature importance and explain predictions of the extra-trees model. They also utilized the genetic model to unveil mathematical relationships among molecular and experimental features and photoelectrochemical stability.
Moreover, density functional theory (DFT) calculations were conducted to explore atomic and electronic structures of molecule-modified perovskite systems. Employing the Perdew-Burke-Ernzerhof (PBE) and Heyd-Scuseria-Ernzerhof (HSE06) functionals, the study calculated geometries, band structures, density of states, work functions, and absorption spectra of the systems, incorporating van der Waals interactions via the Tkachenko-Scheffler scheme. Comparative analysis of DFT calculations with experimental and ML results provided deeper insights into molecular interactions and optoelectronic properties of the materials.
Research Findings
The outcomes demonstrated that molecular engineering of the perovskite surface and interface significantly enhanced the aqueous photoelectrochemical stability and performance of the materials. The authors identified a leading system, "DMSO+PbBr2+calcein," delivering a large aqueous photocurrent of 10-5 A/cm2 and retaining 92.5% of its performance within 200 s. They attributed this system's superior performance to the synergistic effects of compatible molecules and their hydrophilicity/lipophilicity, charge state, and molecular topology.
Furthermore, the ML model effectively predicted and interpreted photoelectrochemical data. The extra-trees model achieved high area-under-curve (AUC) values of 0.86 for the test dataset and 0.94 for the training dataset. SHAP feature analysis underscored the importance of molecular features, such as atomic charge distribution, functional groups, surface area, and hydrophilicity/lipophilicity, for perovskite stability. The genetic model provided a mathematical expression of stability based on molecular and experimental features, revealing a negative correlation between hydrophilicity and the electro-topology of molecules with stability.
Additionally, DFT calculations revealed detailed intermolecular interactions and optoelectronic properties of molecule-modified perovskite systems. They revealed the formation of multiple bonds, including hydrogen bonds, halogen bonds, and anion-pi interactions, with the perovskite surface and between molecules, forming a self-assembled multilayer protecting the perovskite from water damage.
The calculations also indicated improvements in light-harvesting properties, reduction of work function and band gap of perovskite, enhancing energy conversion efficiency and conductivity of the materials.
Applications
The paper demonstrated the feasibility and effectiveness of utilizing molecular engineering and ML to design and evaluate halide perovskite materials for aqueous optoelectronic applications. It presented a comprehensive approach that merged experimental, computational, and data-driven methods to optimize and comprehend the molecular influence on perovskite stability and performance.
Moreover, it identified a promising molecule-modified perovskite system, showcasing high photocurrent and stability in water, with potential utility in solar cells, water-splitting systems, photocatalysts, and environmental sensors.
Conclusion
In summary, the novel approach effectively addressed the aqueous instability issue of halide perovskites, a significant challenge for their practical utilization. It holds the potential to substantially enhance their stability and performance in water while providing scientific interpretation and understanding of molecular contributions.
The researchers suggested that their approach could serve as a powerful tool for designing and evaluating these materials for various optoelectronic applications. Moving forward, future work should focus on optimizing the molecular design process using more advanced ML models and exploring a broader range of diverse and compatible molecules for halide perovskites, including organic cations, anions, and ligands.
Machine Learning Models Predict Arsenic Contamination