With the shrinking of technological nodes, analysis of nanosized-metal-grain pattern-dependent devices is becoming critical; various machine learning (ML) approaches have been utilized to study device characteristic and variability. The inevitable dataset, one of the requisite ML techniques, can be overcome by considering the transfer learning (TL) approach. In this work, an analysis of electrical characteristic affected by work function fluctuation (WKF) with a limited amount of dataset of gate-all-around (GAA) silicon (Si) nanofin (NF) field-effect transistors (FETs) is advanced along with the combination of collected data of GAA Si nanosheet (NS) FETs and TL models. Comparison of the baseline ML model and the proposed TL model shows significant improvement in terms of the values of root mean square error (RMSE) and R²-score. One of applications of this work is to estimate the WKF-induced variability without executing a huge amount of three-dimensional device simulation of GAA Si NF FETs.

随着技术节点的缩小，纳米金属晶粒图案相关器件的分析变得至关重要；各种机器学习 (ML) 方法已被用于研究设备特性和可变性。不可避免的数据集是必要的 ML 技术之一，可以通过考虑迁移学习 (TL) 方法来克服。在这项工作中，利用有限的环栅 (GAA) 硅 (Si) 纳米鳍 (NF) 场效应晶体管 (FET) 数据集分析受功函数波动 (WKF) 影响的电气特性。结合 GAA Si 纳米片 (NS) FET 和 TL 模型的收集数据。基线 ML 模型和建议的 TL 模型的比较表明，在均方根误差 (RMSE) 和 R ^{2的值方面有显着改进}-分数。这项工作的一个应用是在不执行大量 GAA Si NF FET 的三维器件模拟的情况下估计 WKF 引起的可变性。