6.3 Summarizing the Models

For discretization the one-dimensional projections of eqns. (5.38) and (5.40) onto a direction $ \ensuremath{\boldsymbol{\mathrm{e}}}_l$ has to be considered:

$\displaystyle J_{n, \, l}$ $\displaystyle = \mu_n \, \mathrm{k}_\mathrm{B}\, \Bigl( \ensuremath{\partial_{l...
...T_{ll} \bigr)}+ \frac{\mathrm{q}}{\mathrm{k}_\mathrm{B}} \, E_l \, n \Bigr) \ ,$ (6.16)
$\displaystyle S_{n, \, l}$ $\displaystyle = - \frac{5}{2} \, \frac{\mathrm{k}_\mathrm{B}^2}{\mathrm{q}} \, ...
...athrm{k}_\mathrm{B}} \, E_l \, n \, \frac{3 \, T_n + 2 \, T_{ll}}{5} \Bigr) \ .$ (6.17)

By assuming an isotropic MAXWELLIAN distribution, which results in $ T_{ll} = T_n$ and $ \beta_n = 1$, the conventional energy transport model is obtained.

The carrier temperature $ T_n$ defined by eqn. (2.94) is a measure of average carrier energy. The diagonal component of the temperature tensor is given by $ \mathrm{k}_\mathrm{B}\, T_{ll} =
\ensuremath{\langle v_l \, p_l \rangle}$. Off-diagonal components are neglected. The solution variable is still the carrier temperature $ T_n$, whereas the tensor components and the fourth order moment are modeled empirically as functions of $ T_n$ (enqs. (6.3) and (6.15)):

$\displaystyle \gamma_\nu$ $\displaystyle = \gamma_{0\nu} + \bigl( 1 - \gamma_{0\nu} \bigr) \, \exp \Bigl( ...
...}} \Bigr)^2 \Bigr) \ , \textcolor{lightgrey}{.......}\nu = \parallel, \perp \ ,$ (6.18)
$\displaystyle \beta_{n}$ $\displaystyle = \beta_0 + \bigl( 1 - \beta_0 \bigr) \, \exp \Bigl( - \Bigl( \frac{T_n - T_\mathrm{L}}{\mathrm{T_{ref, \beta}}} \Bigr)^2 \Bigr) \ .$ (6.19)

The empirical model for the temperature tensor distinguishes between directions parallel ($ \parallel$) and normal ($ \perp$) to the current density

$\displaystyle T_{\nu} = \gamma_\nu \, T_n \ , \textcolor{lightgrey}{.......}\nu = \parallel, \perp \ .$ (6.20)

The diagonal temperature for a generic direction $ \ensuremath{\boldsymbol{\mathrm{e}}}_l$ is obtained from the average $ \ensuremath{\langle \ensuremath{\boldsymbol{\mathrm{v}}} \ensuremath{\cdot}\en...
...{\mathrm{p}}} \ensuremath{\cdot}\ensuremath{\boldsymbol{\mathrm{e}}}_l \rangle}$ after neglecting the off-diagonal terms as

$\displaystyle T_{ll}$ $\displaystyle = T_\parallel \, \cos^2 \varphi + T_\perp \, \sin^2 \varphi \ ,$ (6.21)
$\displaystyle \varphi$ $\displaystyle = \arccos ( \ensuremath{\boldsymbol{\mathrm{e}}}_l \ensuremath{\cdot}\ensuremath{\boldsymbol{\mathrm{e}}}_J ) \ .$ (6.22)

The graphs of the functions of eqns. (6.18) and (6.19) are displayed in Fig. 6.7.

Figure 6.7: Shape of the functions used to model $ \gamma _\nu $ and $ \beta _n$. $ \gamma _0$ and $ \beta _0$ have been chosen to be $ 0.75$.
\includegraphics{gpfigure/Functional_Shape.color.eps}

Both functions assume unity for $ T_n=T_\mathrm{L}$ and an asymptotic value for large $ T_n$. The exact shape of the transition between these two regions is only of minor importance and mainly affects the numerical stability. Therefore the transition should not be too steep. $ \mathrm{T_{ref}} = 600 \, \mathrm{K}$ appeared to be an appropriate value. Parameter values for $ \gamma_{0\nu}$, $ \beta _0$, and $ \mathrm{T_{ref}}$ can be roughly estimated from Monte Carlo simulations of one-dimensional $ n^+$-$ n$-$ n^+$ test structures (Tbl. 6.1).


Table 6.1: Parameter values estimated from Monte Carlo simulations.
$ \gamma_{0\parallel}$ $ \gamma_{0\perp}$ $ \beta _0$ $ \mathrm{T_{ref, \gamma_\parallel}}$ $ \mathrm{T_{ref, \gamma_\perp}}$ $ \mathrm{T_{ref, \beta}}$
$ 1$ $ 0.75$ $ 0.75$ $ 600 \, \mathrm{K}$ $ 600 \, \mathrm{K}$ $ 600 \, \mathrm{K}$


Monte Carlo results for the anisotropic temperature in a MOSFET are shown in Fig. 6.8 and Fig. 6.9 in comparison with the analytical models. Fig. 6.8 indicates that values for the anisotropy parameter can be as low as $ \gamma _{0y} = 0.6$. Values close to $ \beta _0 = 0.75$ for the non-MAXWELLian parameter in the channel region can be estimated from Fig. 6.9. These parameters show only a weak dependence on doping and applied voltage.

Figure 6.8: Monte Carlo simulation of a $ 90 \hspace {.35ex} \textrm {nm}$ and a $ 180 \hspace {.35ex} \textrm {nm}$ MOSFET (Device 3 with different gate-lengths) showing the $ y$-component of the temperature tensor at the surface compared to the temperature $ T_{n, \textsf {MC}}$ from the mean energy. The analytical model for $ T_{yy}$ uses $ \gamma _{0y} = 0.6$.
\includegraphics{gpfigure/Tn_Tyy_MOS.color.eps}

Figure 6.9: Monte Carlo simulation of a $ 90 \hspace {.35ex} \textrm {nm}$ and a $ 180 \hspace {.35ex} \textrm {nm}$ MOSFET (Device 3 with different gate-lengths) showing the normalized moment of fourth order $ \beta _{n, \textsf {MC}}$ at the surface compared to the analytical model for $ \beta _n$ with $ \beta _0 = 0.75$.
\includegraphics{gpfigure/beta_MOS.color.eps}

M. Gritsch: Numerical Modeling of Silicon-on-Insulator MOSFETs PDF