Within materials informatics, a rapidly developing subfield of materials research, past (curated) data are mined and learned for either discovering new materials or identifying new functionalities of known materials. This paper provides an example of this process. Starting from a recently developed (very diverse) dataset which includes 1346 hybrid organic-inorganic perovskites (HOIPs), we downselect a subset of 350 three dimensional HOIPs to a final set of four lead-free HOIPs, including CH3NH3SnI3, HC(NH2)2SnI3, NH2NH3SnI3, and NH2(CH2)3SnI3, in which the first two were experimentally synthesized and the others remain hypothetical. Using first-principles based computational methods, we show that these HOIPs have preferable electronic band structures and carrier effective mass, good optical properties, and high spectroscopic limited maximum efficiency. Compared to the experimental data, we find that state-of-the-art numerical methods can predict the electronic and optical properties fairly well, while the current model for the spectroscopic limited maximum efficiency is inadequate for capturing the power conversion efficiency of a solar absorber. We suggest that the HOIP dataset should be expanded to include larger structures of HOIPs, thereby being more useful for future data-mining and machine-learning approaches.

中文翻译：

---

用于太阳能电池的无铅杂化有机-无机钙钛矿。

在材料信息学中，对材料研究，发展迅速的子领域，过去的（经过整理的）数据进行挖掘和学习，以发现新材料或识别已知材料的新功能。本文提供了此过程的示例。从包括1346个杂化有机-无机钙钛矿（HOIPs）的最近开发的（种类繁多的）数据集开始，我们将350个三维HOIP的子集向下选择为最终的四个无铅HOIP，包括CH3NH3SnI3，HC（NH2）2SnI3 ，NH2NH3SnI3和NH2（CH2）3SnI3，其中前两个是通过实验合成的，其他两个仍然是假设的。使用基于第一原理的计算方法，我们证明了这些HOIP具有较好的电子带结构和载流子有效质量，良好的光学性能，高光谱限制了最大效率。与实验数据相比，我们发现最先进的数值方法可以很好地预测电子和光学特性，而当前用于光谱有限最大效率的模型不足以捕获太阳能吸收器的功率转换效率。我们建议应将HOIP数据集扩展为包含更大的HOIP结构，从而对将来的数据挖掘和机器学习方法更有用。