Scholarlink

<p>A machine learning-based approach to predict the variability of fully depleted silicon-on-insulator (FDSOI) MOSFETs is demonstrated in this paper. Here, mainly the on and off currents (<i>I</i><sub>ON</sub> and <i>I</i><sub>OFF</sub>), threshold voltage (<i>V</i><sub>TH</sub>), and subthreshold slope (SS) variability due to device geometries, such as channel length, body thickness, and oxide thickness, have been investigated in depth using an AI/ML framework. The random forest regression (RFR) model has demonstrated substantial effectiveness in evaluating the device characteristics. SILVACO ATLAS-based device simulation produces a dataset of 14,627 different readings. Out of different regression and classification methods, the RFR-ML model attains maximum accuracy. The efficacy of the ML frameworks has been measured using mean square error (RMSE) (~ 0.0001), <i>R</i><sup>2</sup> score (~ 0.9986), and accuracy (~ 99.60%) as figures of merit (FOMs), demonstrating their superiority over state-of-the-art literature. In addition to offering higher accuracy, the model can predict data within seconds, in contrast to traditional ways. This rapid and precise design approach is highly beneficial for the current VLSI industries in terms of accuracy and precision and achieves scalability and cost-effectiveness.</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

AI/ML-based model to investigate the variability of the fully depleted SOI MOSFET

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Manash Chanda,
  • Swarnil Roy,
  • Sudipta Ghosh,
  • Papiya Debnath

摘要

A machine learning-based approach to predict the variability of fully depleted silicon-on-insulator (FDSOI) MOSFETs is demonstrated in this paper. Here, mainly the on and off currents (ION and IOFF), threshold voltage (VTH), and subthreshold slope (SS) variability due to device geometries, such as channel length, body thickness, and oxide thickness, have been investigated in depth using an AI/ML framework. The random forest regression (RFR) model has demonstrated substantial effectiveness in evaluating the device characteristics. SILVACO ATLAS-based device simulation produces a dataset of 14,627 different readings. Out of different regression and classification methods, the RFR-ML model attains maximum accuracy. The efficacy of the ML frameworks has been measured using mean square error (RMSE) (~ 0.0001), R2 score (~ 0.9986), and accuracy (~ 99.60%) as figures of merit (FOMs), demonstrating their superiority over state-of-the-art literature. In addition to offering higher accuracy, the model can predict data within seconds, in contrast to traditional ways. This rapid and precise design approach is highly beneficial for the current VLSI industries in terms of accuracy and precision and achieves scalability and cost-effectiveness.