Answers are grounded in this paper and cite their sources.
Tin-based lead-free perovskite solar cells (PSCs), namely \(MASnI_3\) and \(FASnI_3\) , are a promising pathway to PV devices with improved efficiency, but optimizing the multi-dimensional parameters of these cells is difficult. This work proposes an advanced machine learning approach based on a Physics-Informed Feature Tokenizer Transformer (FT-Transformer) to forecast device performance. In order to maintain scientific integrity, a comprehensive 5-fold cross validation procedure was used with various seeds, avoiding the partitioning bias. The results showed that the FT-Transformer had good architecture stability by obtaining consistent cross-validated \(R^2\) values of \(0.8024 ± 0.0135\) for the \(MASnI_3\) and \(0.9574 ± 0.0024\) for the \(FASnI_3\) . Classical ensemble models such as Random Forest had better empirical \(R^2\) but did not maintain physical consistency. On the other hand, the monostaticity constraints we add to our Transformer are custom and enforce the physical properties of semiconductors, such as the fact that the higher the defect density, the lower the efficiency. Lastly, gradient-based inverse design and SHAP explainability analyses reveal that \(FASnI_3\) has stronger sensitivity to densities of defects at the interfaces than \(MASnI_3\) . Cross-validated benchmarking framework provides a comprehensive and physics-based suite of tools to accelerate perovskite technology development.
No takes yet. Share an insight, caveat, or question.
Abdullah et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: