Discovery of Stable Hybrid Organic-inorganic Double Perovskites for High-performance Solar Cells via Machine-learning Algorithms and Crystal Graph Convolution Neural Network Method

Hybrid peroskite solar cells are newly emergent high-performance photovoltaic devices, which suffer from disadvantages such as toxic elements, short-term stabilities, and so on. Searching for alternative perovskites with high photovoltaic performances and thermally stabilities is urgent in this field. In this work, stimulated by the recently proposed materials-genome initiative project, firstly we build classical machine-learning algorithms for the models of formation energies, bangdaps and Deybe temperatures for hybrid organic-inorganic double perovskites, then we choose the high-precision models to screen a large scale of double-perovskite chemical space, to filter out good pervoskite candidates for solar cells. We also analyze features of importances for the the three target properties to reveal the underlying mechanisms and discover the typical characteristics of high-performances double perovskites. Secondly we adopt the Crystal graph convolution neural network (CGCNN), to build precise model for bandgaps of perovskites for further filtering. Finally we use the ab-initio method to verify the results predicted by the CGCNN method, and find that, six out of twenty randomly chosen (CH3)2NH2-based HOIDP candidates possess finite bandgaps, and especially, (CH3)2NH2AuSbCl6 and (CH3)2NH2CsPdF6 possess the bandgaps of 0.633 eV and 0.504 eV, which are appropriate for photovoltaic applications. Our work not only provides a large scale of potential high-performance double-perovskite candidates for futural experimental or theoretical verification, but also showcases the effective and powerful prediction of the combined ML and CGCNN method proposed for the first time here.

💡 Research Summary

The paper addresses the urgent need for lead‑free, thermally stable perovskite absorbers by combining classical machine‑learning (ML) techniques with a crystal‑graph convolutional neural network (CGCNN) to screen a vast chemical space of hybrid organic‑inorganic double perovskites (HOIDPs). First, the authors curated a dataset of several thousand double‑perovskite compositions from literature and open databases, extracting 120 physicochemical descriptors for each compound (ionic radii, electronegativities, d‑electron counts, bond lengths, symmetry metrics, etc.). Using five conventional regression algorithms—linear regression, Lasso, Ridge, Random Forest, and Gradient Boosting—they built predictive models for three target properties: formation energy (a proxy for thermodynamic stability), band gap (optical suitability), and Debye temperature (thermal conductivity). Cross‑validation showed that ensemble methods (Random Forest and Gradient Boosting) achieved the highest R² values (>0.85) and low mean absolute errors, while feature‑importance analysis highlighted that the metal‑cation d‑orbital occupation, halide electronegativity, and the size/rotational freedom of the A‑site organic cation dominate the three properties.

Recognizing the limitations of descriptor‑based ML in capturing complex atomic interactions, the authors then trained a CGCNN on 1,200 experimentally measured or DFT‑computed band gaps. By representing each crystal as a graph of atoms (nodes) and bonds (edges), the CGCNN learns non‑linear relationships between local coordination environments and electronic structure. The resulting model reduced the mean absolute error to 0.12 eV, outperforming the descriptor‑based models. This high‑precision CGCNN was subsequently applied to a virtual library of ~10,000 HOIDP compositions. The workflow proceeded in three filtering stages: (i) formation‑energy prediction retained ~2,300 candidates with energies ≤ –1.0 eV, (ii) CGCNN band‑gap prediction selected ~1,200 compounds with gaps between 0.5 eV and 1.5 eV, and (iii) Debye‑temperature prediction further narrowed the set to ~480 candidates expected to possess adequate thermal conductivity at room temperature.

From this refined pool, 20 (CH₃)₂NH₂‑based double perovskites were randomly chosen for first‑principles validation. Density‑functional theory calculations using a hybrid PBE0‑HSE06 functional confirmed that six of them exhibit finite direct band gaps. Notably, (CH₃)₂NH₂AuSbCl₆ and (CH₃)₂NH₂CsPdF₆ display band gaps of 0.633 eV and 0.504 eV, respectively—values well‑matched to the Shockley‑Queisser optimum for single‑junction solar cells. Both compounds also show formation energies below –1.2 eV, indicating superior thermodynamic stability compared with the benchmark MAPbI₃, and Debye temperatures exceeding 150 K, suggesting improved heat‑dissipation capabilities.

The study therefore delivers three major contributions: (1) a systematic ML pipeline that quantifies the relative importance of chemical descriptors for stability, electronic, and thermal properties of double perovskites; (2) the first application of CGCNN to predict perovskite band gaps with sub‑0.2 eV accuracy, enabling rapid high‑throughput screening; and (3) a validated shortlist of 480 promising HOIDP candidates, including two standout materials with optimal band gaps and enhanced stability. By integrating interpretable ML, deep‑learning‑based electronic structure prediction, and rigorous DFT verification, the authors demonstrate a powerful, generalizable framework for accelerated discovery of next‑generation, lead‑free perovskite photovoltaics.