A Physical-Based Artificial Neural Networks Compact Modeling Framework for Emerging FETs

作     者：Yang, Ya-Shu Li, Yiming Kola, Sekhar Reddy Reddy 

作者机构：National Yang Ming Chiao Tung University Parallel and Scientific Computing Laboratory The Institute of Communications Engineering Hsinchu300093 Taiwan National Yang Ming Chiao Tung University Parallel and Scientific Computing Laboratory Institute of Communications Engineering Institute of Biomedical Engineering Department of Electronics and Electrical Engineering Hsinchu300093 Taiwan National Yang Ming Chiao Tung University Parallel and Scientific Computing Laboratory Electrical Engineering and Computer Science International Graduate Program Hsinchu300093 Taiwan 

出 版 物：《IEEE Transactions on Electron Devices》 (IEEE Trans. Electron Devices)

年 卷 期：2024年第71卷第1期

页      面：223-230页

核心收录：

学科分类：0808[工学-电气工程] 0809[工学-电子科学与技术（可授工学、理学学位）] 08[工学] 0702[理学-物理学] 

基　　金：This work was supported in part by the National Science and Technology Council (NSTC), Taiwan, under Grant MOST 111-2221-E-A49-181, Grant MOST 111-2634-F-A49-008, and Grant NSTC 111-2218-E-492-009-MBK in part by the "2022 Qualcomm Taiwan Research Program [National Yang Ming Chiao Tung University (NYCU)]" under Grant NAT-487835 SOW and in part by the "Center for mm-Wave Smart Radar Systems and Technologies" under the Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education in Taiwan 

主　　题：Neural networks 

摘      要：We report a compact modeling framework based on the Grove-Frohman (GF) model and artificial neural networks (ANNs) for emerging gate-all-around (GAA) MOSFETs. The framework consists of two ANNs;the first ANN constructed with the drain current model not only can capture the main trend of device I - V characteristics but also can predict its variation even when the amount of training data for the ANN is insufficient or outside the range of applied biases. The second one is then designed to improve the model accuracy by further minimizing the errors between the target and the model outputs. We implement the proposed framework to accurately model emerging GAA nanosheet (NS) MOSFETs and complementary FETs (CFETs) without suffering from divergent issues in circuit simulation. In addition, nonphysical behaviors, such as nonzero current at zero bias, do not occur in the modeling framework. Compared to recently reported machine-learning (ML) models, our approach can achieve a similar level of model accuracy with merely 20% amount of the training data. © 1963-2012 IEEE.