Researchers from Xi’an University of Architecture and Technology, Xidian University, Xi’an Jiaotong University, University of New South Wales, Beijing University of Posts and Telecommunications and Ningbo Weiyuan Optoelectronic Research Institute have developed a machine learning-assisted screening strategy to identify lead-free double perovskite materials tailored for agrivoltaic applications, where photovoltaic performance must be balanced with crop light requirements.
Agrivoltaic systems place unique constraints on solar absorbers: in addition to high power conversion efficiency, materials must allow selective transmission of photosynthetically useful light. While silicon solar cells can reach efficiencies of 29.4%, their opacity limits plant growth. Semitransparent technologies such as perovskite solar cells (PSCs), with certified efficiencies up to 27.3%, offer a more suitable platform due to their tunable absorption spectra. However, conventional lead-based PSCs face stability and toxicity concerns, motivating the search for lead-free alternatives.
In this work, the team focused on halide double perovskites with the general formula A₂BⅠBⅢX₆, where Pb²⁺ is replaced by a combination of monovalent and trivalent cations. These materials retain the structural stability of traditional perovskites while enabling flexible bandgap engineering and improved environmental compatibility.
To accelerate materials discovery, the researchers constructed an XGBoost regression model using optimized descriptors to predict bandgaps across a dataset of 1,189 inorganic perovskites. The model achieved a high predictive accuracy with 𝑅2=0.95, enabling efficient pre-screening prior to first-principles calculations. By combining bandgap predictions with stability and agrivoltaic suitability criteria, four promising candidates were identified: Cs₂AgSbCl₆, Cs₂InBiCl₆, Cs₂AgBiCl₆, and Cs₂TlSbCl₆.
Detailed electronic structure analysis revealed that the B-site cation configuration plays a decisive role in determining whether these materials exhibit direct or indirect bandgaps, which in turn affects optical absorption and carrier dynamics. Optical simulations showed that all four compounds exhibit distinct spectral responses spanning the blue-to-red wavelength range. These spectral features align with key plant photoreceptor pathways, including blue-light regulation via CRY-HY5-CHS/ELIP and red/far-red responses via PHY-PIF-HY5/FHY1, indicating the potential to simultaneously support photosynthesis and electricity generation.
Device-level performance was evaluated using SCAPS simulations. Among the candidates, Cs₂AgSbCl₆ demonstrated the most balanced performance, achieving a power conversion efficiency of 23.10% alongside an average visible transmittance of 24.27%. This combination resulted in the highest light-use efficiency (LUE) of 5.61%, a key metric for agrivoltaic systems that captures the trade-off between energy production and crop illumination. For example, a greenhouse integrating such a material could maintain sufficient blue and red light for plant growth while converting less useful spectral regions into electrical power.
In addition to optoelectronic properties, mechanical behavior was also considered. Cs₂TlSbCl₆ and Cs₂InBiCl₆ exhibited favorable flexibility, suggesting compatibility with bendable or roll-to-roll processed photovoltaic modules, which are particularly attractive for plastic-film greenhouse applications.
Overall, the study establishes a combined ML-DFT framework for screening agrivoltaic materials and highlights lead-free double perovskites as promising candidates. By linking composition, spectral selectivity, device performance, and plant photobiology, it provides a pathway toward spectrally optimized, high-efficiency photovoltaic systems for sustainable agriculture.
