Optimization of Silver Nanowire Spin Coating for Flexible Transparent Conducting Electrodes Guided by Machine Learning

Transparent conducting electrodes (TCEs) play a crucial role in various energy applications, including light-emitting diodes (LEDs), photovoltaics, and thin-film transistors. Achieving both high transparency and high conductivity in TCEs poses a significant challenge. When using metal grids, a higher filling factor produces a layer with a higher conductivity but lower transmittance. Similarly, indium tin oxide (ITO), the most commonly used transparent conducting oxide, exhibits reduced transmittance with increasing film thickness, even though its conductivity improves. To fabricate TCEs on polyethylene terephthalate (PET) substrates, we combine silver nanowires (AgNWs) with sol-gel indium zinc oxide (IZO). Doing so can enhance the conductivity of solution-processed IZO films to meet the sheet resistance requirements for TCEs used in LEDs and photovoltaics. [1]<br/> <br/>Our study primarily focuses on optimizing the spin-coating process of AgNWs, as they significantly influence the transmittance and sheet resistance of TCEs. Our goal is to achieve high transmittance and low sheet resistance simultaneously. The optimization problem involves mapping a three-dimensional input (AgNW solution concentration, spin speed, and dispense volume) to a two-objective output (transmittance and sheet resistance). Using data obtained by Latin hypercube sampling (LHS) of the input space, we construct two independent Gaussian process (GP) regression models: one for transmittance and one for sheet resistance, using cross-validation to tune their hyperparameters. These models have distinct kernel parameters due to the differing nature of the objectives. Having two models also enables us to conduct multi-objective analysis. We analyze the data and GP models in two ways. Firstly, we utilize a well-established criterion [2] to define a scalar figure-of-merit (FOM) and utilize the GP models to predict the maximum FOM and associated uncertainty. Secondly, we construct a Pareto frontier based on the GP models, which helps identify processing conditions that strike the best trade-off between higher transmittance and lower sheet resistance. It is important to note the difference between the two approaches. Using FOM cannot individually specify the objective values, so it proves useful when separate requirements for transmittance and sheet resistance are desired when optimizing FOM. Finally, we compare the predictions of models with tuned hyperparameters using different cross-validation methods. Subsequently, we conduct experiments using the identified input parameters, from both the predicted maximum FOM and the Pareto frontier, and compare the results with the model predictions.<br/> <br/>This work is supported by NSF CMMI-2109554.<br/> <br/>References:<br/>[1] R. T. Piper, W. Xu, and J. W. P. Hsu, “Silver Nanowire-Indium Zinc Oxide Composite Flexible Transparent Conducting Electrodes Made by Spin-coating and Photonic Curing,” MRS Adv. 8, 177-182 (2023)<br/>[2] M. Dressel and G. Gruner, “Electrodynamics of Solid”, Cambridge University Press, 2002