Inverse Design of Microstructures via Generative Networks for Organic Solar Cells

Xian Yeow Lee, Aditya Balu, Balaji Sesha Sarath Pokuri, Adarsh Krishnamurthy, Soumik Sarkar, Baskar Ganapathysubramanian

Keywords: inverse design, generative adversarial networks, organic solar cells

TL;DR: conditional generative adversarial networks for generating microstructures for organic solar cells conditioned over current properties

Abstract: We consider the inverse problem of efficiently designing material microstructures that exhibit desired electrical properties in an organic solar cell design. We leverage data-driven generative models to learn the underlying data distribution and generate novel microstructures during test time. We focus on a recent framework, specifically generative invariance networks (InvNets), which simultaneously learns from a dataset of microstructures while constraining the output of the generative model to conform to constraints such as generating microstructure designs with a targeted short circuit current density, J values. While previous works in this area have focused on the model training and data efficiency aspects, the applicability and success of Generative Invariance Networks to different material systems (i.e., the donor material and the acceptor material chemistry) and device thickness remain unexplored. In this paper, we demonstrate that we can successfully adapt the same InvNet framework to different material systems and device thicknesses with minimal computational effort.