Machine Learning-DrivenInk Concentration Optimizationin Solution-Processed Organic Light-Emitting Diodes for Enhanced CurrentEfficiency

Optimizing an organic light-emitting diode (OLED) structure for high efficiency typically requires extensive efforts in tuning layer thicknesses and conducting photometric analyses. While computational simulations can shorten the optimization process, applying them to solution-processed OLEDs (s-OLEDs) is challenging since ink concentration affects both the thickness and morphology of films, complicating the simulation substantially. To address this, we employed machine learning (ML), specifically Gaussian Process Regression (GPR), to optimize s-OLED ink concentrations without the need to calibrate material- or thickness-dependent properties typically required in conventional simulations. The GPR model efficiently suggested optimal ink concentrations for the hole transport layer (<i>c</i>_HTL) and electron transport layer (<i>c</i>_ETL) to maximize the current efficiency (CE). Using a data set of ink concentrations (2.5–17.5 g/L) as input and CE measured at 100, 1000, and 10,000 cd/m² as output, the model was trained to propose viable combinations of <i>c</i>_HTL and <i>c</i>_ETL. The GPR model successfully identified three distinct optimal ink concentration pairs within just three experimental iterations. By integrating GPR with spectral analysis, we also uncovered the interplay among ink concentration, recombination zone dynamics, and CE. Based on these findings, we proposed a mechanism explaining why distinct ink concentration pairs yield higher efficiency at different luminance levels. This study highlights the potential of ML techniques not only in streamlining s-OLED optimization but also in providing deeper insights into the relationships among the s-OLED structure, charge carrier dynamics, and performance.