Publikationsserver
From First-Principles to Machine Learning: Advancing Bandgap Predictions in Strain-Engineered III-V Semiconductors
Semiconductor compounds composed of elements from groups 13 and 15 (main groups III and V) of the periodic table, commonly referred to as III-V semiconductors, are integral to modern (opto-)electronics. They play a critical role in applications such as solar cells, light-emitting diodes, optical tel...
保存先:
| 第一著者: | |
|---|---|
| その他の著者: | |
| フォーマット: | Dissertation |
| 言語: | 英語 |
| 出版事項: |
Philipps-Universität Marburg
2023
|
| 主題: | |
| オンライン･アクセス: | PDFフルテキスト |
| タグ: |
タグ追加
タグなし, このレコードへの初めてのタグを付けませんか!
|
| 要約: | Semiconductor compounds composed of elements from groups 13 and 15 (main groups III and V) of the periodic table, commonly referred to as III-V semiconductors, are integral to modern (opto-)electronics. They play a critical role in applications such as solar cells, light-emitting diodes, optical telecommunication, laser technology, photodetectors, and high-speed electronics. The performance and characteristics of these devices heavily rely on the bandgap value and its direct or indirect character. Consequently, tailoring bandgaps to specific applications is a major goal in semiconductor research field. This holds immense importance in advancing the capabilities and efficiency of semiconductor-based technologies.
This thesis focuses on two primary approaches for tuning bandgaps in III-V semiconductors: varying composition and applying strain to the materials. To identify tailored materials for specific applications, it is crucial to assess the dependence of bandgaps on composition and strain across a broad range of materials. However, experimental methods face limitations in exploring the vast chemical space of combinations of III- and V-elements with variations in composition and strain due to challenges in synthesizing new materials.
In this thesis, a density functional theory (DFT)-based first-principles approach is established to accurately predict bandgaps in strained III-V compound semiconductor materials. A robust scheme is developed within the DFT framework to accurately model the application of various types of strain on a material. The study reveals that not only do the bandgap values change under strain but also the nature of the bandgap can transition from direct to indirect or vice versa. The established DFT protocol enables a comprehensive mapping of bandgap properties with composition and strain in multinary III-V semiconductors, facilitating efficient screening of promising materials for device designs. The investigated materials span binary III-V systems such as GaAs, GaP, GaSb, InP, InAs, InSb, and Si, as well as various ternary materials including GaAsP, GaAsN, GaPSb, GaAsSb, GaPBi, and GaAsBi.
Furthermore, as the composition-strain space expands, standalone DFT approaches become computationally demanding for higher-order systems, such as quaternary and pentanary III-V semiconductor materials. The number of DFT calculations required increases significantly in those systems (~millions). To address this, a hybrid approach is developed by integrating a support vector machine-based supervised machine learning (ML) model with DFT. This hybrid DFT-ML approach reduces the number of DFT calculations required by a factor of 1000 while maintaining high prediction accuracy. The effectiveness of this approach is demonstrated through the mapping of bandgaps in the III-V quaternary compound GaAsPSb across its entire composition range and a wide range of strain values, which would otherwise be impractical with standalone DFT method. This hybrid approach enables computationally efficient bandgap predictions across a diverse range of materials and strains, offering a rapid virtual screening capability for the discovery of novel semiconductor materials in (opto-)electronic applications. |
|---|---|
| DOI: | 10.17192/z2023.0669 |