Scholarlink

<p>Crosstalk noise in sub-micron technology nodes significantly undermines the reliability of interconnects by inducing unwanted effects on neighboring interconnect lines and degrading signal integrity. Effective mitigation strategies are crucial to maintain the robustness and performance of these advanced systems. While crosstalk prediction aims to inform reliability mechanisms to advise avoidance methods efficiently, static models fail to adapt to post-layout, thermal, and temporal uncertainties. This paper introduces advanced crosstalk predictors designed to dynamically adapt to thermal and physical uncertainties in interconnect systems. These predictors feature adjustable parameters that can be precisely fine-tuned based on specific interconnect characteristics. When error predictions of crosstalk, the predictors enter a training phase guided by a learning algorithm and implemented in an on-chip manner, ensuring continuous improvement and accuracy in predicting crosstalk effects over time. Simulation results demonstrate that the predictor can adapt to electromigration and temperature variations with an average accuracy of 94.3% and 95.4 in the presence of 100% variations. The integration of the predictor and learning algorithm into the router of network-on-chips (NoCs) incurs an area overhead of 6.3%.</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

Uncertainty-Aware Crosstalk Predictors for Adaptive Reliability in Core Interconnects

  • 期刊论文

  • 发表日期：

    2025年5月21日

查看全文
  • Rezgar Sadeghi,
  • Amirhossein Ilkhani,
  • AmirHosein Yazdanpanah

摘要

Crosstalk noise in sub-micron technology nodes significantly undermines the reliability of interconnects by inducing unwanted effects on neighboring interconnect lines and degrading signal integrity. Effective mitigation strategies are crucial to maintain the robustness and performance of these advanced systems. While crosstalk prediction aims to inform reliability mechanisms to advise avoidance methods efficiently, static models fail to adapt to post-layout, thermal, and temporal uncertainties. This paper introduces advanced crosstalk predictors designed to dynamically adapt to thermal and physical uncertainties in interconnect systems. These predictors feature adjustable parameters that can be precisely fine-tuned based on specific interconnect characteristics. When error predictions of crosstalk, the predictors enter a training phase guided by a learning algorithm and implemented in an on-chip manner, ensuring continuous improvement and accuracy in predicting crosstalk effects over time. Simulation results demonstrate that the predictor can adapt to electromigration and temperature variations with an average accuracy of 94.3% and 95.4 in the presence of 100% variations. The integration of the predictor and learning algorithm into the router of network-on-chips (NoCs) incurs an area overhead of 6.3%.