A machine learning study on superlattice electron blocking layer design for AlGaN deep ultraviolet light-emitting diodes using the stacked XGBoost/LightGBM algorithm

Abstract

Aluminium gallium nitride (AlGaN)-based deep ultraviolet (DUV) light-emitting diodes (LEDs) suffer from low internal quantum efficiency (IQE) and serious efficiency droop. One reason for this is the electron leakage and poor hole injection related to the band alignment of the heterojunctions, doping, polarization effect, and others. In the past, the AlGaN/AlGaN superlattice (SL) electron blocking layer (EBL) was proposed to optimize the carrier transport and improve the LED performance. However, the SL-EBL design is a trade-off of multiple physics mechanisms, and the LED efficiency deteriorates when the design is improper. We used extreme gradient boosting (XGBoost) and light gradient boosting machine (LightGBM) stacked machine learning (ML) models to predict various high-performance SL-EBLs considering different compositions, thicknesses, and band offset ratios. Based on the ML model, we propose an easier and experimentally achievable low Al-content SL-EBL (1 nm/5 nm Al_0.7Ga_0.3N/Al_0.58Ga_0.42N) that can significantly optimize carrier transport. The improvement in IQE and wall-plug efficiency could be as high as about 70% compared with those of the conventional bulk EBL. Moreover, we analyze the prediction data and reveal the influence of the composition and thickness on the IQE improvement. The composition difference should be enlarged at a higher band offset ratio, which may be explained by the electron potential and polarization modulation. The critical thickness of the optimized SL-EBL is investigated to guarantee effective electron blocking without destroying the material quality, doping modulation, and operating voltage. This work provides a systematic study of SL-EBLs and helps promote the development and application of SL-EBLs for high-efficiency DUV LEDs.