Accelerating evaluation of the mobility of ionic liquid-modulated PEDOT flexible electronics using machine learning

PEDOT has been widely used in advanced electronics, and one of the keys to determine the performance is hole mobility. PEDOT commonly shows amorphous morphology ascribed to the flexibility of its backbone, giving rise to a wide difference in mobility. To boost the mobility, one generally introduces an ionic liquid (IL) to modulate the morphology to be more ordered. To estimate the mobility, one needs to do molecular dynamics (MD) simulations to acquire the abundant conformers, then to investigate the transfer integral (V-ij) via quantum mechanics (QM) calculations theoretically or via quantum Hall effect measurements experimentally. Here, with the help of machine learning (ML) technology (involving supervised learning algorithms of linear regression (LR), artificial neural network (ANN), random forest (RF), and gradient boosting decision tree (GBDT)), we can predict V-ij accurately compared to the routine MD -> QM method (for ANN and RF, R-2 > 0.9 and MAE = 10(-3) eV), while shortening the prediction time by 6 orders of magnitude. Generalization verification on an additional five IL-PEDOT cases confirms the predictive ability of the model. Then, the predicted V-ij was used to estimate the mobility. Finally, representative IL [EMIM][TFSI]-regulated PEDOT aqueous solutions with different concentrations were experimentally characterized by AFM and conductivity measurements, the conductivity being in line with the change tendency of the estimated mobility. This alternative ML model opens up new perspectives for ultrafast prediction of the mobility of IL-PEDOT in any morphology and can be transferred to other analogs before real device construction.

Automated summary

Through the use of machine learning technology, it is possible to accurately predict the performance of PEDOT materials and reduce prediction time. Experimental validation confirms the predictive ability of the model.