Scholarlink

<p>Molybdenum disulfide is an attractive material for photodetectors owing to its strong optical absorption and favourable charge-transport characteristics in the ultrathin limit. In this research work, we numerically investigated a vertically stacked FTO/MoS<sub>2</sub>/GQD/Au photodetector using the SCAPS-1D software. The study systematically examines the effects of key device parameters including the thickness of MoS<sub>2</sub> and graphene quantum dots, acceptor doping concentration, bulk defect density, series resistance, temperature, and back contact work function on essential photodetector performance metrics such as photocurrent, open-circuit voltage, responsivity, and detectivity. The optimized device configuration features a GQD thickness of 0.4&#xa0;μm and a MoS<sub>2</sub> thickness of 1&#xa0;μm, yielding a short-circuit current density of 34.90&#xa0;mA/cm<sup>2</sup> and an open-circuit voltage of 1.035&#xa0;V, the device also demonstrates strong responsivity of 0.3526&#xa0;A/W and a high detectivity of 19.72<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\:\times\:\)</EquationSource> </InlineEquation>10<sup>13</sup> Jones, under standard AM 1.5G illumination. The wavelength-based analysis shows a peak photocurrent density of 67.06&#xa0;mA cm<sup>− 2</sup> at 900&#xa0;nm (near-infrared), highlighting its suitability for NIR detection. To further analyse the influence of physical parameters on photocurrent, Linear Regression, Random Forest, and XGBoost models were applied. While tree-based models showed near-perfect performance under standard cross-validation, this was attributed to interpolation within the structured parameter grid. Under Leave-One-Level-Out (LOLO) validation, Random Forest and XGBoost failed to generalize to unseen parameter levels, yielding significantly degraded and negative R<sup>2</sup> values, particularly for MoS<sub>2</sub> thickness. In contrast, Linear Regression demonstrated stable performance, highlighting its superior extrapolation capability. These findings emphasize the importance of appropriate validation strategies and model selection for reliable prediction in physics-based simulation studies.</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

Design and optimization of hybrid photodetector for near-infrared detection: a SCAPS-1D and machine learning approach

  • 期刊论文

  • 发表日期：

    2026年5月30日

查看全文
  • Md Amanullah Saifee,
  • Mohd. Shahid Khan,
  • Javid Ali,
  • Anita Sharma

摘要

Molybdenum disulfide is an attractive material for photodetectors owing to its strong optical absorption and favourable charge-transport characteristics in the ultrathin limit. In this research work, we numerically investigated a vertically stacked FTO/MoS2/GQD/Au photodetector using the SCAPS-1D software. The study systematically examines the effects of key device parameters including the thickness of MoS2 and graphene quantum dots, acceptor doping concentration, bulk defect density, series resistance, temperature, and back contact work function on essential photodetector performance metrics such as photocurrent, open-circuit voltage, responsivity, and detectivity. The optimized device configuration features a GQD thickness of 0.4 μm and a MoS2 thickness of 1 μm, yielding a short-circuit current density of 34.90 mA/cm2 and an open-circuit voltage of 1.035 V, the device also demonstrates strong responsivity of 0.3526 A/W and a high detectivity of 19.72 \(\:\times\:\) 1013 Jones, under standard AM 1.5G illumination. The wavelength-based analysis shows a peak photocurrent density of 67.06 mA cm− 2 at 900 nm (near-infrared), highlighting its suitability for NIR detection. To further analyse the influence of physical parameters on photocurrent, Linear Regression, Random Forest, and XGBoost models were applied. While tree-based models showed near-perfect performance under standard cross-validation, this was attributed to interpolation within the structured parameter grid. Under Leave-One-Level-Out (LOLO) validation, Random Forest and XGBoost failed to generalize to unseen parameter levels, yielding significantly degraded and negative R2 values, particularly for MoS2 thickness. In contrast, Linear Regression demonstrated stable performance, highlighting its superior extrapolation capability. These findings emphasize the importance of appropriate validation strategies and model selection for reliable prediction in physics-based simulation studies.