Data-efficient prediction of OLED optical properties enabled by transfer learning.

Area of Science:

• Materials Science
• Optoelectronics
• Computational Physics

Background:

• Global structural optimization of organic light-emitting diodes (OLEDs) for maximal light extraction is a long-standing goal.
• Key challenges include time-consuming optical simulations and discrepancies between simulated and experimental results.

Purpose of the Study:

• To develop a fast and reliable method for predicting OLED optical properties.
• To improve data efficiency in surrogate modeling for OLED design.
• To bridge the gap between simulated and experimental OLED performance.

Main Methods:

• Leveraging transfer learning with artificial neural networks (ANNs).
• Training ANNs on simulated OLED data and transferring knowledge to new structures.
• Fine-tuning pre-trained ANNs with limited experimental data to correct systematic errors.

Main Results:

• Achieved significantly higher data efficiency compared to previous ANN-based surrogate solvers.
• Demonstrated accurate prediction of modified OLED structures with minimal additional training data.
• Enabled accurate prediction of experimental OLED measurements using ANNs trained on simulated data, correcting for experimental errors.

Conclusions:

• Transfer learning offers a practical approach for rapid and reliable prediction of OLED optical properties.
• This method facilitates the optimization of complex OLED structures with numerous design parameters.
• The approach enhances the design cycle for high-efficiency OLEDs by integrating simulation and experimental data effectively.