Scholarlink

<p>This study proposes a brand-new method for estimating power in Very-large-scale integration (VLSI) circuits: Pre- and Post-Synthesis Analysis with Interpretable Generalized Additive Neural Networks (PEVLSI-PPS-IGANN). The method starts with collecting data from the VLSI dataset, then handles missing values and normalizes the input data using the Cauchy Robust Correction-Sage Husa Extended Kalman Filter (CRCHEKF). The cleaned and preprocessed data are then fed into the IGANN, which is intended to predict power dissipation in integrated circuits during both the pre-layout and post-layout stages. The optimized IGANN model, assisted by PSOA, improves prediction accuracy and generalization while preserving interpretability. While IGANN provide interpretable predictions, it lacks the ability to automatically optimize its settings. To overcome this limitation, the PSOA is used to adjust IGANN’s mass parameters, hence improving prediction accuracy and performance. The main contributions of this study are threefold: development of a robust preprocessing framework using the CRCHEKF to handle noisy and incomplete data; design of an IGANN that explains the contribution of each feature to power dissipation; and integration of the PSOA for adaptive parameter tuning to enhance accuracy and generalization. Together, these innovations provide a unified and transparent power estimation framework for both pre- and post-synthesis analysis in VLSI circuits.</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

Power estimation in VLSI circuits: pre- and post-synthesis analysis with interpretable generalized additive neural networks

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • T. Thyagaraj,
  • N. Ramavenkateswaran,
  • Madhusudan Kulkarni,
  • Nagaraj Bhat

摘要

This study proposes a brand-new method for estimating power in Very-large-scale integration (VLSI) circuits: Pre- and Post-Synthesis Analysis with Interpretable Generalized Additive Neural Networks (PEVLSI-PPS-IGANN). The method starts with collecting data from the VLSI dataset, then handles missing values and normalizes the input data using the Cauchy Robust Correction-Sage Husa Extended Kalman Filter (CRCHEKF). The cleaned and preprocessed data are then fed into the IGANN, which is intended to predict power dissipation in integrated circuits during both the pre-layout and post-layout stages. The optimized IGANN model, assisted by PSOA, improves prediction accuracy and generalization while preserving interpretability. While IGANN provide interpretable predictions, it lacks the ability to automatically optimize its settings. To overcome this limitation, the PSOA is used to adjust IGANN’s mass parameters, hence improving prediction accuracy and performance. The main contributions of this study are threefold: development of a robust preprocessing framework using the CRCHEKF to handle noisy and incomplete data; design of an IGANN that explains the contribution of each feature to power dissipation; and integration of the PSOA for adaptive parameter tuning to enhance accuracy and generalization. Together, these innovations provide a unified and transparent power estimation framework for both pre- and post-synthesis analysis in VLSI circuits.