Machine learning-based analysis of finite element simulations for emulating the light-matter interactions in nanostructured, disordered photoelectrodes

Photoelectrochemical (PEC) solar energy conversion applications rely on chemical reactions driven by photogenerated minority carriers (electrons or holes) at a semiconductor-liquid junction. The optical, electronic, and chemical transport processes characteristic of these PEC reactions occur on independent and generally disparate length scales. Fabricating electrodes with hierarchical structure can optimize the performance for each of these processes simultaneously. Disordered materials with dielectric contrast on the length scales of the wavelength of light can trap light in localized modes. The simplicity of the fabrication alone makes this approach a particularly attractive one for engineering light trapping into scalable photoelectrode structures. One significant issue is that disordered materials can only be defined by ensemble or statistical parameters (pore diameter, scatterer diameter, relative volume fractions) rather than as precise structures, which results in a real, physical variance intrinsic to the ensemble structure. Simulations of the properties of a given ensemble (local light absorption, for example) require a large number of examples for generating statistically accurate measurements for those properties. In this talk, we will describe our recent efforts to use neural network emulation to explore light concentration in a simplified, disordered photonic glass. This system provides an apt model for developing mathematical representations for electrode structure (input) and finite element simulation data (output) to interface with machine learning algorithms. We will also discuss an approach to simplify representations for large-scale simulation data based on principal component analysis. We will outline the practical application of these models and show how algorithmic predictions can be used to identify the most efficient ensemble configuration for a nanostructured semiconductor photoelectrode based on self-organized colloidal composites.