Proceedings of MATSUS Fall 2024 Conference (MATSUSFall24)
DOI: https://doi.org/10.29363/nanoge.matsusfall.2024.283
Publication date: 28th August 2024
The optimization of the perovskite solar cell has been considered one of the important technological challenges. Since the device can be developed through various combinations of materials and experimental conditions, efficient use of computational techniques for device optimization is highly demanded to overcome the problem of the so-called "combinatorial explosion" in experimental studies. In our laboratory, we are trying to collectively apply various computational tools in computer simulations and machine learning techniques. In this presentation, we will report on our recent trial to predict device characteristics of perovskite solar cells using machine learning and computer simulations. One challenge is predicting device performances such as photo conversion efficiency (PCE) from experimentally obtained cross-sectional scanning electron microscope (SEM) images. By applying convolutional neural network algorithms, we could have achieved moderate success in so-called image regression, in which SEM images are used as input to predict PCE. Another challenge is to discover an optimal material and device configuration through device simulation and machine learning (regression) modeling. In the presentation, we will discuss our computational method and tools with some predictive results toward optimizing the perovskite solar cell.
This presentation is based on results obtained from a project, JPNP21016, commissioned by the New Energy and Industrial Technology Development Organization (NEDO), Japan.