Abstract

The performance of an organic photovoltaic device is intricately connected toits active layer morphology. This connection between the active layer anddevice performance is very expensive to evaluate, either experimentally orcomputationally. Hence, designing morphologies to achieve higher performancesis non-trivial and often intractable. To solve this, we first introduce a deepconvolutional neural network (CNN) architecture that can serve as a fast androbust surrogate for the complex structure-property map. Several tests wereperformed to gain trust in this trained model. Then, we utilize this fastframework to perform robust microstructural design to enhance deviceperformance.

 

Quick Read (beta)