Wide Bandgap Semiconductor Device Design via Machine Learning

Abstract

Recently, there has been a surge in study of III-nitride wide-bandgap semiconductor devices like laser diodes (LDs), ultra-violet (UV) LEDs, and high electron mobility transistors (HEMTs). There are numerous opportunities for performance improvement in the wide bandgap semiconductor device structure, including material selection, compound compositions, polarization effects, and layer thicknesses. On the other hand, they can make optimization more challenging. It still takes a lot of resources to analyze and test complicated structures in a systematic manner. This dissertation creates a new path for device design by using TCAD and machine learning to deliver quick forecasts of the III-nitride semiconductor device. The dissertation includes three major components, which are as follows:

In Chapter 2, the TCAD assisted HEMT device design is discussed. Based on TCAD simulations, we demonstrated the performance improvement of using the new material BAlN as an interlayer in GaN/AlGaN HEMT devices and compared the various interlayer design alternatives for HEMTs. The outcomes demonstrate that by employing BAlN as the interlayer, the two-dimensional electron gas can be enhanced greatly, which improves the device's operational performance. Under V_GS-V_th = +3 V for the 1 nm interlayers, When compared to the traditional AlN interlayer, the saturation currents of the B_0.14Al_0.86N/AlN hybrid interlayer as well as the B_0.14Al_0.86N interlayer are, respectively, 5.8% and 2.2% greater.

In chapter 3, we propose asymmetrical p-AlGaN/i-InGaN/n-AlGaN tunnel junctions (TJs) by combining machine learning with TCAD calculations. LEDs and HEMTs have made extensive use of symmetrical p-AlGaN/i-InGaN/n-AlGaN TJs to enhance device performance. However, because of the intricacy of the design, asymmetric tunnel junctions were not carefully studied. In order to model p-AlGaN/i-InGaN/n-AlGaN TJs with varying Al or In compositions and InGaN layer thicknesses, TCAD software was used. We built a highly effective machine learning (ML) model for predicting TJ resistance after being trained by these data. The model, which develops a tool for real-time TJ resistance prediction, predicted the resistances for 22254 different TJ topologies. Our TJ predictions indicate that the asymmetric p-Al_0.7Ga_0.3N/i-In_0.2Ga_0.8N/n-Al_0.3Ga_0.7N TJ, which has a greater Al composition in the p-layer, has seven times less TJ resistance than the symmetric p-Al_0.3Ga_0.7N/i-In_0.2Ga_0.8N/n-Al_0.3Ga_0.7N TJ.

In Chapter 4, using the stacked XGBoost/LightGBM algorithm, we thoroughly examine the superlattice (SL) electron blocking layer (EBL) for AlGaN deep ultraviolet light-emitting diodes. Previously, the AlGaN/AlGaN SL EBL was proposed to optimize carrier transport and improve LED performance. However, the SL-EBL design is a compromise between a number of physical mechanisms, and when the design is flawed, the LED efficiency suffers. Here, we employed a stacked machine learning (ML) model that combines the light gradient boosting machine (LightGBM) and extreme gradient boosting (XGBoost) to forecast a variety of high-performance SL-EBLs while taking varying compositions, thicknesses, and band offset ratios into account. Based on the ML model, we suggest a low Al-content SL-EBL (1 nm/5 nm Al_0.7Ga_0.3N/Al_0.58Ga_0.42N) that is simpler and experimentally realizable and can greatly improve carrier transport. Unlike the conventional bulk EBL, the IQE and wall-plug efficiency could improve by as much as 70%. Additionally, we examine the prediction data and show how the composition and thickness affect the improvement of the IQE. With a higher band offset ratio, the compositional difference should be more pronounced, which the electron potential and polarization modulation may help to explain. The critical thickness of the optimized SL-EBL is investigated to guarantee effective electron blocking without destroying material quality, doping modulation, and operating voltage. This work contributes to the advancement and use of SL-EBLs for high-efficiency DUV LEDs by providing methodical research of SL-EBLs.

This dissertation presents novel approaches to the design of electrical and optical wide bandgap semiconductor devices, which opens new avenues for future research. It is possible that it might be used in a broad variety of sectors, including illumination, sensing, disinfection, and power devices.

Brief Biography

Rongyu Lin is a Ph.D. student at the Advanced Semiconductors Laboratory under the supervision of Prof. Xiaohang Li at King Abdullah University of Science and Technology (KAUST). He received his Bachelor of Science in Physics from Southern University of Science and Technology(SusTech). Rongyu's research interests include wide bandgap materials, optoelectronics and applied machine learning.