Scholarlink

<p>Self-heating effects (SHEs) pose significant challenges for advanced semiconductor devices, particularly gate-all-around field-effect transistors (GAAFETs). Although finite element simulation and thermal network modeling serve as robust electrothermal analysis tools, their reliance on intricate physical mechanisms and numerous parameters results in prohibitively high computational costs. By contrast, machine learning (ML) techniques provide an innovative paradigm for predicting device behaviors. In this work, a methodology for developing artificial neural network (ANN) models to rapidly integrate nanosheet geometry-dependent electrothermal properties of advanced GAAFETs into device and circuit simulations is proposed. To accurately replicate SHEs-induced I–V and T–V characteristics, the ANN model is optimized through output nonlinearity control and input range management via conversion functions and a logarithmic transformation-based preprocessing scheme. The trained model achieves test accuracy of ~ 98.5% for drain current and ~ 99% for temperature prediction across diverse geometries and supply voltages. These precise predictions demonstrate that ANN-based model effectively captures underlying thermal physics, exhibiting strong capability and generality for circuit-level performance evaluation.</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

Artificial neural network-assisted modeling of GAAFETs for geometry-dependent electrothermal co-simulation

  • 期刊论文

  • 发表日期：

    2026年6月28日

查看全文
  • Ziping Wang,
  • Linhui Lai,
  • Fei Li,
  • Yabin Sun,
  • Yanling Shi,
  • Bingyi Ye,
  • Yuhang Zhang,
  • Yang Shen,
  • Xiaojin Li

摘要

Self-heating effects (SHEs) pose significant challenges for advanced semiconductor devices, particularly gate-all-around field-effect transistors (GAAFETs). Although finite element simulation and thermal network modeling serve as robust electrothermal analysis tools, their reliance on intricate physical mechanisms and numerous parameters results in prohibitively high computational costs. By contrast, machine learning (ML) techniques provide an innovative paradigm for predicting device behaviors. In this work, a methodology for developing artificial neural network (ANN) models to rapidly integrate nanosheet geometry-dependent electrothermal properties of advanced GAAFETs into device and circuit simulations is proposed. To accurately replicate SHEs-induced I–V and T–V characteristics, the ANN model is optimized through output nonlinearity control and input range management via conversion functions and a logarithmic transformation-based preprocessing scheme. The trained model achieves test accuracy of ~ 98.5% for drain current and ~ 99% for temperature prediction across diverse geometries and supply voltages. These precise predictions demonstrate that ANN-based model effectively captures underlying thermal physics, exhibiting strong capability and generality for circuit-level performance evaluation.