Identifying representative sub-domains in 3D microstructures for accelerated structure–property mapping in organic photovoltaic

The morphology of the active layer significantly influences the performance of organic photovoltaics (OPVs). Recent advances in experimental techniques have facilitated the visualization and digitization of complex microstructures within the OPV active layer, paving the way to construct quantitative structure–property relationships. Given a digitized morphology, detailed physics-based simulations can provide accurate characterization; however, performing these simulations on the full digitized morphology is computationally prohibitive, thus impeding high-throughput characterization and exploration. To address this issue, we propose a fast mathematical approach to select small sub-volumes of the full digitized morphology, which exhibit similar performance characteristics as the full morphology. The selection of such ‘representative volume elements (RVE)’ consists of two main steps: (a) Identification of a small set of morphology features (or ‘descriptors’) that are computationally light to calculate and that are correlated with the property of interest. These descriptors serve as a surrogate for the true property of interest. (b) Integrating this surrogate model with an adaptive Bayesian-based approach to efficiently identify sub-volumes of the morphology exhibiting nearly identical features to the full morphology. We illustrate the utility of this approach using large discretized morphologies containing greater than 64 million voxels, reconstructed using electron tomography. This approach produces a ∼400 fold reduction in computational effort with less than 1% reduction in accuracy. While our application focus is on OPVs, the methodology presented is generally applicable to problems where detailed physics simulations bottleneck quantitative structure–property relationships, including in fields like catalysis, membranes, and bioelectronics.