Accuracy improvement with reliable statistical-based models for CNT-FET applications

作者：Pivezhandi Mohammad; Abedi Kambiz^*; Hassanzadeh Alireza

来源：Journal of Computational Electronics, 2017, 16(3): 610-619.

DOI：10.1007/s10825-017-1026-3

• Artificial neural networks
• Carbon nanotube
• Decision trees
• Nanoscale transistor designs
• Random forest
• Support vector regression

In this work, different statistical-based algorithms for modeling of carbon nanotube (CNT) field-effect transistors (FETs) are analyzed and implemented. CNT-FETs have complicated structures; thus simulating their models requires significant computational time. In previous studies, many researchers have concentrated on artificial neural networks (ANNs) to obtain accurate models that expedite simulation. However, due to the local optima problem, these designs require a trial-and-error approach and are less accurate than more recently developed algorithms. Therefore, we have utilized support vector regression to obviate the local optima problem, and have used decision trees (DTs) to ensure accuracy. The salient features of qualitative-based DT techniques include their accuracy and speed, as they propose a convex solution and do not require preliminary conditioning. Tenfold cross-validation is considered to yield infallible parameters for all of the analyzed techniques. Thus, the corresponding models are almost identical when compared with each other. Among the analyzed techniques, the random forest (RF) algorithm exhibits around 9% error for off-state current, i.e., eightfold greater accuracy than the previous compact model. The RF learning time is also more than ten times shorter than the fastest ANN package. All the plots and simulation results in this study are extracted from the R programming environment.

• 出版日期2017-9