Optoelectronic perovskite film characterization via machine vision

We present our research for fast and reliable extraction of bandgap and absorption quality values for triple-cation perovskite thin films from sample scans. Our approach leverages machine learning methods, namely convolutional neural networks, to perform regression tasks aimed at predicting the properties of interest. To this end, thin film samples were synthesized via blade-coating and their photoluminescence and ultraviolet-visible spectra collected, along with the film thickness. We propose a method of computing a dimensionless figure of merit we called the Area Under Absorption Coefficient (AUAC), its purpose being to qualitatively evaluate the absorption quality of perovskite films for use in photovoltaic modules. This work demonstrates the usability of simple imaging techniques to analyze experimental samples while requiring only a feasibly acquirable initial amount of data. Our reported method can help speed up time consuming material optimizations by reducing lab time spent on recurrent characterization, nicely synergizes with high throughput production lines and could be adapted for quick extraction of other optoelectrical quantities.

Automated summary

We present a method for extracting bandgap and absorption quality values for triple-cation perovskite thin films. By using machine learning, such as convolutional neural networks, we can predict these properties using sample scans. We propose a dimensionless figure of merit called the Area Under Absorption Coefficient (AUAC) to evaluate the absorption quality of perovskite films for photovoltaic modules. This method can speed up material optimizations and be adapted for quick extraction of other optoelectrical quantities.