Scholarlink

<p>Static and dynamic IR drop prediction is a challenging task in today’s nanometric design for power delivery network (PDN). An accurate prediction of static and dynamic IR drop is essential for ensuring that circuits meet their timing and performance constraints. By identifying potential voltage droops early, designers can mitigate timing violations and avoid performance degradation, thereby supporting the required throughput of the circuit. This work aims to predict static and dynamic IR drop using advanced Artificial Neural Networks (ANN) based models, that are optimized for accuracy and computational efficiency. For static IR drop prediction, an optimized ANN architecture utilizes features such as the number of sinks, half-perimeter wire length (HPWL), wire length, and resistance. Fine-tuned hyperparameters allow the model to achieve an average correlation coefficient (CC) of over 97% compared with the ground truth, significantly reducing simulation time compared to traditional methods. For dynamic IR drop prediction, a hybrid bagged ANN model is proposed that utilizes an ensemble of pre-trained base ANNs. The base ANN utilizes Swish activation, the Adam optimizer, and a selected set of features that represent power dissipation, switching activity, and impedance characteristics. The bagging framework improves robustness by decreasing variance and enhancing generalization. Experimental results show that the proposed models can provide fast and accurate predictions of IR drop. The bagged ANN model achieves a mean squared error (MSE) of 0.0018 and an <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10836_2025_6174_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="21" /> </InlineMediaObject> <EquationSource Format="TEX">\(R^2\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> score of 0.8119 for dynamic IR drop prediction, achieving a 26x speedup compared to commercial EDA tool, while the static IR drop model ensures high accuracy.</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 Based Prediction Model for IR Drop Measurement in a VLSI Power Delivery Network

  • 期刊论文

  • 发表日期：

    2025年5月30日

查看全文
  • Amit Mondal,
  • Tuhina Samanta

摘要

Static and dynamic IR drop prediction is a challenging task in today’s nanometric design for power delivery network (PDN). An accurate prediction of static and dynamic IR drop is essential for ensuring that circuits meet their timing and performance constraints. By identifying potential voltage droops early, designers can mitigate timing violations and avoid performance degradation, thereby supporting the required throughput of the circuit. This work aims to predict static and dynamic IR drop using advanced Artificial Neural Networks (ANN) based models, that are optimized for accuracy and computational efficiency. For static IR drop prediction, an optimized ANN architecture utilizes features such as the number of sinks, half-perimeter wire length (HPWL), wire length, and resistance. Fine-tuned hyperparameters allow the model to achieve an average correlation coefficient (CC) of over 97% compared with the ground truth, significantly reducing simulation time compared to traditional methods. For dynamic IR drop prediction, a hybrid bagged ANN model is proposed that utilizes an ensemble of pre-trained base ANNs. The base ANN utilizes Swish activation, the Adam optimizer, and a selected set of features that represent power dissipation, switching activity, and impedance characteristics. The bagging framework improves robustness by decreasing variance and enhancing generalization. Experimental results show that the proposed models can provide fast and accurate predictions of IR drop. The bagged ANN model achieves a mean squared error (MSE) of 0.0018 and an \(R^2\) R 2 score of 0.8119 for dynamic IR drop prediction, achieving a 26x speedup compared to commercial EDA tool, while the static IR drop model ensures high accuracy.