Vision-driven Autocharacterization of Perovskite Semiconductors

In materials research, the task of characterizing hundreds of different materials traditionally requires equally many human hours spent measuring samples one by one. We demonstrate that with the integration of computer vision into this material research workflow, many of these tasks can be automated, significantly accelerating the throughput of the workflow for scientists. We present a framework that uses vision to address specific pain points in the characterization of perovskite semiconductors, a group of materials with the potential to form new types of solar cells. With this approach, we automate the measurement and computation of chemical and optoelectronic properties of perovskites. Our framework proposes the following four key contributions: (i) a computer vision tool for scalable segmentation to arbitrarily many material samples, (ii) a tool to extract the chemical composition of all material samples, (iii) an algorithm capable of automatically computing band gap across arbitrarily many unique samples using vision-segmented hyperspectral reflectance data, and (iv) automating the stability measurement of multi-hour perovskite degradation experiments with vision for spatially non-uniform samples. We demonstrate the key contributions of the proposed framework on eighty samples of unique composition from the formamidinium-methylammonium lead tri-iodide perovskite system and validate the accuracy of each method using human evaluation and X-ray diffraction.