Modeling and Optimization of GAAFET Transistor using Taguchi Approach and Artificial Neural Network (ANN)

In this work, a gate all around GAAFET transistor has been studied and numerically optimized. The finite element method was adopted for the discretization of the equations describing the electrical model. On the basis of a Taguchi
$\boldsymbol{L}_{\mathbf{27}}(\mathbf{3}^{\mathbf{7}})$
orthogonal array, the effects of control factors such as oxide thickness
$(\mathbf{t}_{\mathbf{ox}})$
, nanowire radius (R), gate length
$(\mathbf{L}_{\mathbf{g}})$
, trap density
$(\mathbf{N}_{\mathbf{T}})$
, ramp parameter
$(\boldsymbol{\alpha})$
, gate voltage
$(\mathbf{V}_{\mathbf{g}})$
and drain voltage
$(\mathbf{V}_{\mathbf{d}})$
were investigated. The optimization of the control factors was performed by calculating the signal to noise ratio (S/N). The development of numerical prediction models also included the use of multiple linear and nonlinear regressions as well as an artificial neural network. Numerical analysis revealed that the overall optimal drain current of the design parameters is found with
$\boldsymbol{t}_{\boldsymbol{ox}}= \mathbf{1}\mathbf{nm},\ \mathbf{R}=\mathbf{10}\mathbf{nm},\\boldsymbol{L}_{\boldsymbol{g}}=\mathbf{100}\boldsymbol{nm}, \boldsymbol{N}_{\boldsymbol{t}}= \mathbf{510}^{\mathbf{16}}/\boldsymbol{m}^{\mathbf{2}}\,\ \boldsymbol{\alpha}= \mathbf{1}, \boldsymbol{V}_{\boldsymbol{g}}=\mathbf{1}.\mathbf{5}\boldsymbol{V}$
and
$\boldsymbol{V}_{\boldsymbol{d}}=\mathbf{2}\boldsymbol{V}$
to reach the maximum drain current equal to 1.385mA. The analysis of variance (ANOVA) shows that the maximum contribution (39% for the drain current) is occupied by the ramp coefficient
$(\boldsymbol{\alpha})$
while the radius of the nanowire R, the gate length
$\boldsymbol{L}_{\boldsymbol{g}}$
, the density of
$\boldsymbol{N}_{\boldsymbol{t}}$
trap have the minimum contribution (less than 3%) to optimize the performance of the GAAFET transistor. Taking into account the statistical analysis, the developed ANN model is efficient for the prediction of the output characteristics of the GAAFET transistor.