Scholarlink

<p>In this work, we propose an AI-driven framework for the automated and intelligent design of photodetectors, which significantly enhances design efficiency. Specifically, by integrating a high-accuracy machine learning (ML) model with a genetic algorithm (GA) and a decision-making technique (TOPSIS), this intelligent and automated approach eliminates the need for labor-intensive manual analysis and extensive physical prototyping, while simultaneously accounting for multiple interdependent parameters. The ML model is implemented as a backpropagation-trained multilayer perceptron (BP-MLP) neural network, which effectively captures the complex and nonlinear relationships between device structure and performance, enabling rapid and accurate prediction of device characteristics across various structural configurations. Building on this, the GA can rapidly identify the optimal device structure for specific performance targets. To demonstrate its effectiveness, we design two modified uni-traveling carrier photodetectors (MUTC-PDs): a high-speed device with a 3-dB bandwidth of 246.1&#xa0;GHz and a responsivity of 0.12 A/W, and a high-power device achieving a 3-dB bandwidth of 46.6&#xa0;GHz and an RF output power of 31.82&#xa0;dBm at 30&#xa0;GHz.</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

An AI-driven framework for the intelligent design of high-performance photodetectors

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Jihong Ye,
  • Liwen Wang,
  • Shuhu Tan,
  • Xiaomin Ren,
  • Yongqing Huang

摘要

In this work, we propose an AI-driven framework for the automated and intelligent design of photodetectors, which significantly enhances design efficiency. Specifically, by integrating a high-accuracy machine learning (ML) model with a genetic algorithm (GA) and a decision-making technique (TOPSIS), this intelligent and automated approach eliminates the need for labor-intensive manual analysis and extensive physical prototyping, while simultaneously accounting for multiple interdependent parameters. The ML model is implemented as a backpropagation-trained multilayer perceptron (BP-MLP) neural network, which effectively captures the complex and nonlinear relationships between device structure and performance, enabling rapid and accurate prediction of device characteristics across various structural configurations. Building on this, the GA can rapidly identify the optimal device structure for specific performance targets. To demonstrate its effectiveness, we design two modified uni-traveling carrier photodetectors (MUTC-PDs): a high-speed device with a 3-dB bandwidth of 246.1 GHz and a responsivity of 0.12 A/W, and a high-power device achieving a 3-dB bandwidth of 46.6 GHz and an RF output power of 31.82 dBm at 30 GHz.