Scholarlink

<p>This work presents the design and electromagnetic optimization of a silicon carbide (SiC)–based RF-MEMS switch targeting W-band applications in the 92–95&#xa0;GHz range, based on the ambitious energy-efficiency requirements and expanded frequency allocations envisioned for future 6G systems. To overcome the computational burden of full-wave electromagnetic simulations traditionally used for RF-MEMS design, artificial neural network (ANN) models are developed to perform fast and accurate prediction of the switch performance. The ANNs were trained using data obtained from finite-element simulations to learn the nonlinear mapping between geometrical parameters regarding the dimensioning of the coplanar waveguide recesses and key RF figures of merit. Once trained, such ANN models may provide near-instantaneous evaluations while preserving simulation-level accuracy, enabling a further efficient multi-parameter optimization of the SiC-based switch. To illustrate the benefits of including the developed model in the switch design process, appropriate examples of exploiting models are highlighted. It is shown how the developed models can be exploited to efficiently simulate the insertion and return losses behavior with changes of the recess dimensions, focusing on the sensitivity analysis with the deviations of the recess dimensions caused by the fabrication tolerances. In addition, it is illustrated how ANN simulations successfully facilitate expedient power analysis. The proposed combination of SiC technology and ANN-assisted modeling effectively supports the development of low-power, broadband, and reliable RF-MEMS switches for future 6G RF front ends.</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 networks for the future communication standards: case study of a RF-MEMS switch for W band applications

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Zlatica Marinković,
  • Biljana P. Stošić,
  • Marija Milijić,
  • Dušan Marković,
  • Girolamo Tagliapietra,
  • Jacopo Iannacci

摘要

This work presents the design and electromagnetic optimization of a silicon carbide (SiC)–based RF-MEMS switch targeting W-band applications in the 92–95 GHz range, based on the ambitious energy-efficiency requirements and expanded frequency allocations envisioned for future 6G systems. To overcome the computational burden of full-wave electromagnetic simulations traditionally used for RF-MEMS design, artificial neural network (ANN) models are developed to perform fast and accurate prediction of the switch performance. The ANNs were trained using data obtained from finite-element simulations to learn the nonlinear mapping between geometrical parameters regarding the dimensioning of the coplanar waveguide recesses and key RF figures of merit. Once trained, such ANN models may provide near-instantaneous evaluations while preserving simulation-level accuracy, enabling a further efficient multi-parameter optimization of the SiC-based switch. To illustrate the benefits of including the developed model in the switch design process, appropriate examples of exploiting models are highlighted. It is shown how the developed models can be exploited to efficiently simulate the insertion and return losses behavior with changes of the recess dimensions, focusing on the sensitivity analysis with the deviations of the recess dimensions caused by the fabrication tolerances. In addition, it is illustrated how ANN simulations successfully facilitate expedient power analysis. The proposed combination of SiC technology and ANN-assisted modeling effectively supports the development of low-power, broadband, and reliable RF-MEMS switches for future 6G RF front ends.