Generative Adversarial Networks for Data Augmentation and Inverse Design of Synthesis Conditions in Perovskite Solar Cells

Modeling the complex relationships among synthesis parameters, material compositions, and performance metrics is essential for accelerating the development of perovskite solar cells (PSCs). While common approaches utilize discriminative models, this study adopts Generative Adversarial Networks (GANs) for modeling the underlying data distribution. In this work, we evaluate this generative framework on two tasks. First, we utilize an unconditional GAN for data augmentation to densify the experimental manifold. Second, to enable targeted inverse design, we implement a Conditional GAN (cGAN) based on a Weighted AC-GAN architecture with an inverse frequency-based loss weighting strategy. Results show that, regarding data augmentation, our method reduces the root mean square error (RMSE) in predictive tasks by 7.1. Concerning inverse design, our proposed model enables the generation of synthesis recipes, even for high-efficiency targets, offering a new method to accelerate the discovery of perovskite-based photovoltaic devices.