Perovskite Solar Cell: Chemical Composition and Bandgap Energy via Machine Learning

Résumé

The exponential growth in publications and applications of perovskite photovoltaic cells highlights their significance in energy conversion and carbon emissions mitigation. From 2009 to 2023, the efficiency of these cells has significantly increased from 3.9% to 25.7%. The adaptive capacity of perovskite structures for solar spectrum absorption and current displacement is strongly influenced by the bandgap energy, ideally situated between 1.3 and 1.7 eV. Although various perovskite compositions can potentially attain this energy range, the synthesis methodologies remain empirically driven, presenting challenges to experimental viability. In this context, leveraging experimental databases provided by global researchers emerges as an effective approach to expedite and enable research on perovskite structures for photovoltaic cells. This study utilized the comprehensive MaterialsZone database to feed machine learning algorithms, focusing on Support Vector Machine (SVM) and Random Forest (RF) methodologies to predict the bandgap energy in a targeted perovskite composition. By conducting synthesis experiments towards specific compositions guided by model predictions, it becomes feasible to efficiently achieve the desired bandgap energy. Such a strategy not only accelerates research progress but also serves to curtail costs associated with the synthesis of perovskite materials. The RF model exhibited an average percentage error of 5.13%, a standard deviation of the percentage error of 6.99%, and a Root Mean Square Error (RMSE) of 0.119. In contrast, the SVM model recorded an average percentage error of 4.05%, a standard deviation of the percentage error of 6.45%, and RMSE of 0.881. These developed models not only demonstrate high predictive capacity but also contribute substantively to the comprehension of the intricate relationship between the chemical composition and bandgap energy values of perovskites. By deploying machine learning algorithms, this work paves the way for targeted optimizations and considerable strides in the manufacturing of perovskite-based photovoltaic cells.