Predicting high-performance perovskite solar cells using AI-based machine learning models

Abstract

Perovskite solar cells (PSCs) have garnered significant attention in the photovoltaic field due to their remarkable power conversion efficiencies (PCEs), with the certified PCE reaching 27.0% for single-junction cells and 30.1% for tandem perovskite/perovskite multijunction cells over the past decade. However, challenges remain including material instability, compositional inconsistency, and limited long-term performance. Machine learning (ML) has emerged as a transformative tool to address these challenges by accelerating material discovery, optimizing device design, and enabling data-driven insights from large and complex datasets. This review presents a comprehensive analysis of how ML is being applied to advance PSCs technologies. It begins with an overview of perovskite structures, device architectures, and performance parameters relevant to ML modelling. A structured ML workflow is introduced, covering data acquisition, feature selection, model development, performance evaluation, and model interpretability through explainable AI (XAI) techniques. Recent studies are examined across two major domains: material discovery and device performance optimization. Unlike previous reviews, this work emphasis on quantitative comparisons of ML algorithms by systematically assessing models reported in recent literature to identify the most effective predictors across various tasks. Furthermore, it discusses the strengths and limitations of current datasets and modelling strategies. The review concludes with insights into existing challenges and outlines future directions to support the efficient, interpretable, and scalable application of ML in PSCs research.