Scholarlink

<p>New advanced LEDs will have to be developed by developing completely new methods of modeling and optimizing semiconductor nanostructures, especially those with hybrid organic–inorganic interfaces. Traditional approaches, such as Density Functional Theory (DFT), cannot capture the richness of quantum interactions, or multi-objective design challenges taken on by such systems, which ultimately result in suboptimal device performance. In this work, we overcome these challenges by implementing a gamut of state-of-the-art computational methods suited to the complexities of hybrid nanostructures. For that purpose, Quantum-Inspired Tensor Networks are used, namely Matrix Product States and Tree Tensor Networks, to optimize the electronic and optical properties of nanostructures. These methods thus manage the high-dimensional quantum state space effectively and, with the resultant approach, enhance the accuracy of band structure predictions by 20–30%, while the efficiency of light emission is enhanced by about 15%. The next step in interface engineering was to set up SHapley Additive exPlanations (SHAP) for explainable AI such that a detailed understanding of several interface features' contributions to LED performance can be made. This has pointed out interface roughness and material composition as the most determining factors, hence guiding further optimization efforts. We use Proximal Policy Optimization (PPO)—a reinforcement learning algorithm—to improve fabrication processes. This leads to a 12% increase in the light emission intensity and a 20% reduction in the variability of the process. Finally, we employ Bayesian Optimization with Gaussian Processes (BO-GP) for effective exploration in the multi-objective design space that achieves an emission efficiency enhancement of about 18%, with material cost reduction up to 10%. Altogether, these techniques greatly advance the design, fabrication, and performance optimization of hybrid semiconductor nanostructures and represent an enabling step toward next-generation high-efficiency LEDs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The optimization of design and performance in hybrid organic/inorganic LEDs toward next-generation high-efficiency LEDs: application of multi-model hybrid machine learning approach

  • 期刊论文

  • 发表日期：

    2025年5月14日

查看全文
  • Ujwala S. Ghodeswar,
  • Rupali S. Balpande,
  • Vaishali P. Raut,
  • Manisha G. Waje,
  • Yoginee S. Pethe,
  • Nilesh Shelke,
  • Haytham F. Isleem,
  • Vikrant S. Vairagade

摘要

New advanced LEDs will have to be developed by developing completely new methods of modeling and optimizing semiconductor nanostructures, especially those with hybrid organic–inorganic interfaces. Traditional approaches, such as Density Functional Theory (DFT), cannot capture the richness of quantum interactions, or multi-objective design challenges taken on by such systems, which ultimately result in suboptimal device performance. In this work, we overcome these challenges by implementing a gamut of state-of-the-art computational methods suited to the complexities of hybrid nanostructures. For that purpose, Quantum-Inspired Tensor Networks are used, namely Matrix Product States and Tree Tensor Networks, to optimize the electronic and optical properties of nanostructures. These methods thus manage the high-dimensional quantum state space effectively and, with the resultant approach, enhance the accuracy of band structure predictions by 20–30%, while the efficiency of light emission is enhanced by about 15%. The next step in interface engineering was to set up SHapley Additive exPlanations (SHAP) for explainable AI such that a detailed understanding of several interface features' contributions to LED performance can be made. This has pointed out interface roughness and material composition as the most determining factors, hence guiding further optimization efforts. We use Proximal Policy Optimization (PPO)—a reinforcement learning algorithm—to improve fabrication processes. This leads to a 12% increase in the light emission intensity and a 20% reduction in the variability of the process. Finally, we employ Bayesian Optimization with Gaussian Processes (BO-GP) for effective exploration in the multi-objective design space that achieves an emission efficiency enhancement of about 18%, with material cost reduction up to 10%. Altogether, these techniques greatly advance the design, fabrication, and performance optimization of hybrid semiconductor nanostructures and represent an enabling step toward next-generation high-efficiency LEDs.