next up previous contents
Next: 3. The Silicon/Silicon-Dioxide Interface Up: 2. Simulation of Semiconductor Previous: 2.3 Carrier Generation and

Subsections


2.4 Quantum Mechanical Effects

The classical semiconductor device equations from Section 2.1 imply that the mobile carriers, electrons and holes, behave like classical particles in the semiconductor. For large device dimensions this assumption gives very good results, but for small device geometries quantum mechanical effects like the quantum mechanical tunneling, described in Section 5.3, and the quantum mechanical confinement gain importance. The latter effect leads to a reduction of allowed states for electrons and holes near a \ensuremath {\textrm {Si/SiO$_2$}} interface. In classical device simulations using the drift-diffusion approximation the peak of the electron concentration in the channel of a turned on n-channel MOSFET is calculated to be directly at the \ensuremath {\textrm {Si/SiO$_2$}} interface. This is not correct as the number of allowed states is drastically reduced close to the interface and therefore the peak of the carrier concentration lies several angstroms away from the interface [9].


2.4.1 Quantum Confinement

For the modeling of NBTI the carrier concentration close to the \ensuremath {\textrm {Si/SiO$_2$}} interface plays an important role (Section 6.4.4). The use of quantum confinement models reduces this carrier concentration and might have significant influence on the NBTI model used.

In classical device simulators quantum confinement is often accounted for by using additional quantum correction models. These models locally change the carrier density-of-states [10,11] or they modify the conduction band edge close to the interface [12].

2.4.1.1 Density of States Correction

In classical device simulation the density-of-states (DOS) in homogeneous materials is modeled as a constant value throughout the device. In order to describe the quantum mechanical confinement a distance-dependent reduction of the DOS at the \ensuremath {\textrm {Si/SiO$_2$}} interface has been proposed in [10,11]
\begin{displaymath}
h_\mathrm{corr} = 1-\exp\left(-\frac{(z+z_0)^2}
{\zeta^2\lambda_\mathrm{TH}^2}\right)  ,
\end{displaymath} (2.106)

where $z$ is the distance to the \ensuremath {\textrm {Si/SiO$_2$}} interface and $z_0$ shifts the whole function relative to the interface. $\zeta$ is a newly introduced parameter which enables the variation of $\lambda_\mathrm{TH}$ for calibration purposes. The symbol $\lambda_\mathrm{TH}$ denotes the thermal wavelength which is given by
\begin{displaymath}
\lambda_\mathrm{TH} = \frac{\hbar}{\sqrt{2m^* \ensuremath{\textrm{k$_\textrm{B}$}}T}}   ,
\end{displaymath} (2.107)

where $\hbar$ is the reduced Planck constant, $m^*$ is the effective carrier mass, \ensuremath{\textrm{k$_\textrm{B}$}} the Boltzmann constant, and $T$ the temperature. The resulting DOS, $N_\mathrm{c}$, is then calculated from the classical DOS $N_\mathrm{c,0}$, which is normally modeled as a constant throughout the semiconductor, with the correction factor $h_\mathrm{corr}$ as
\begin{displaymath}
N_\mathrm{c} = N_\mathrm{c,0} h_\mathrm{corr}   .
\end{displaymath} (2.108)

Figure 2.5: Plot of the DOS correction factor $h_\textrm {corr}$ for $z_0=1 $nm and $\zeta =1$ at $T=300 $K.
\includegraphics[width=\figwidth]{figures/haenschcorr-new}

The interplay of the different parameters and the distance to the \ensuremath {\textrm {Si/SiO$_2$}} interface is schematically depicted in Figure 2.5. The parameter $z_0$ is important, because with $z_0=0$ the DOS at the interface becomes zero. This would cause numerical problems and reduce the convergence of the numerical solver. A positive number of $z_0$ shifts the correction function towards the dielectric, approximately considering wave function penetration. The value of $\zeta\lambda_\mathrm{TH}$ defines the effective depth of the correction. A high value, which can be achieved with $\zeta>1$, leads to a reduction of the DOS even deep in the substrate.

Note that the correction factor does not depend on the bias and the band edge energies are not influenced. Hence, the model can be evaluated in a preprocessing step and does not impose any additional computational burden during iteration steps.

2.4.1.2 Conduction Band Edge Correction

Figure 2.6: Band edge bending at the \ensuremath {\textrm {Si/SiO$_2$}} interface. The classical band edge is corrected by the factor $\Delta \ensuremath {E}_\textrm {g} F(z)$.
\includegraphics[width=10cm]{figures/vandort}
An alternative approach is based on $E_0$, the first eigenvalue of the triangular energy well, as seen in Figure 2.6. This model was proposed by van Dort [12]
\begin{displaymath}
\Delta E_\mathrm{g} = E_0 - E_\mathrm{c}(0) = \frac{13}{9} ...
...extrm{B}$}}T}\right)^{1/3}
\vert E_\mathrm{n}\vert^{2/3}   ,
\end{displaymath} (2.109)

where the proportionality factor $\beta = 4.1\times10^{-8} \mathrm{eV cm}$ is is found from the observed threshold voltage shift at high doping levels [13], $\varepsilon_\mathrm{si}$ is the permittivity of silicon, and $E_\mathrm{n}$ is the electric field at the \ensuremath {\textrm {Si/SiO$_2$}} interface perpendicular to the interface.

The value of $\Delta E_\mathrm{g}$ is multiplied with a distance-dependent weight function which has been introduced by Selberherr [14] for the modeling of surface roughness scattering in MOSFETs. The function is of the following form

\begin{displaymath}
F(z) = \frac{2\exp(-(z/z_\mathrm{ref})^2)}
{1+\exp(-2(z/z_\mathrm{ref})^2)}   ,
\end{displaymath} (2.110)

where $z_\mathrm{ref}$ is the scaling factor for the interface distance. Thus, the resulting band edge energy with van Dort's quantum correction of the classical band edge energy $E_\mathrm{class}$ reads as follows
\begin{displaymath}
E_\mathrm{c} = E_\mathrm{class} + F(z) \Delta E_\mathrm{g}   .
\end{displaymath} (2.111)

Figure 2.6 depicts the distance dependent weight function $F$ and the band edge energy for both, the classical approach and after quantum correction with van Dort's method.

2.4.1.3 Evaluation of Quantum Correction Models

For the evaluation of the quantum correction models a state-of-the-art three-dimensional n-channel FinFET device structure was chosen. The device geometry can be seen in Figure 2.7.
Figure 2.7: Device geometry of a triple-gate FinFET structure. Quantum confinement plays an important role in this device because of its small size and the formation of three, instead of one, channels.
\includegraphics[width=10cm]{figures/finfet}
The silicon fin has a cross section area of 6$\times$10nm$^2$. The gate length is 20nm with a gate oxide thickness of 1.5nm. The source and drain regions are heavily n-type doped whereas the channel itself remains undoped.

Figure 2.8: Electron concentration in a triple-gate FinFET for classical simulation (left), with the DOS correction model (middle), and the band edge energy correction model by Van Dort. The correction models force the peak of the carrier concentration away from the \ensuremath {\textrm {Si/SiO$_2$}} interface into the substrate.
\includegraphics[width=16cm]{figures/cut-econcentrations}

Figure 2.8 depicts the electron concentration in a two-dimensional cut through the silicon fin in the middle of the channels. The gates are biased at 0.9V with the source and drain contacts grounded. The classical simulation using the drift-diffusion approximation gives the highest magnitude of the electron concentration at the \ensuremath {\textrm {Si/SiO$_2$}} interfaces below the gate contacts. It can be seen that the peak electron concentration is found in the top corners, as two respective gates couple to the channel, each of them attracting carriers. With the quantum confinement correction models the maximum carrier concentration is moved to the inside of the fin by a distance depending on the chosen model and its calibration parameters.

Figure 2.9: Electron concentration across the fin using classical device simulation and the confinement correction models.
\includegraphics[width=\figwidth]{figures/econc_rot}

Figure 2.9 depicts a one-dimensional cut through the fin displaying the carrier concentrations for the different models for the same bias conditions. Comparing with Figure 2.10 reveals that qualitatively the DOS correction model delivers better results.

Figure 2.10: Comparison of classical and quantum mechanical carrier concentrations for different fin widths. The quantum mechanical calculations have been performed using a Schrödinger solver.
\includegraphics[width=\figwidth]{figures/qmclassic_rot}
A comparison of FinFETs with different fin widths can be seen in Figure 2.10. It shows the electron concentration across the fin simulated with both, the classical drift-diffusion model and a Schrödinger solver, respectively. The fin widths are 6, 12, and 18nm. At a fin width of 6nm the electron concentration has its maximum in the center of the fin. This shape of the carrier concentration cannot be reproduced by the quantum correction models. With larger widths the maximum moves to the interfaces enabling a better fit of the correction models.

Figure 2.11: Comparison of the output characteristics of double- and triple-gate FinFETs at a gate voltage of 0.9 V. Quantum confinement correction reduces the drain current as no carriers are allowed at the \ensuremath {\textrm {Si/SiO$_2$}} interface.
\includegraphics[width=\figwidth]{figures/outputchar-color}
The channels in the silicon fin are displaced from the surface to the inside of the silicon and thus the drive current is reduced. Figure 2.11 depicts the drain current for a gate voltage of 0.9V and different quantum correction mechanisms. Additionally to the triple-gate device the simulation has been performed with a double-gate structure, where the top gate from Figure 2.7 has been replaced with \ensuremath {\textrm {SiO$_2$}}. Simulation of the double-gate structure shows a reduced output current by a factor of approximately 20% due to the formation of only two channels.

Quantum correction leads to a considerable reduction of the saturation current. The DOS correction model yields reasonable results, but since it does not account for the band bending it must be calibrated for each bias point. Van Dort's model completely fails to reproduce the carrier concentration in the channel which may be due to the assumption of a triangular energy well. This assumption might be a too crude estimation for extremely thin channels. Therefore, these models can be very well used to describe the current reduction in very thin channel devices, but not when the shape of the carrier concentration is important. Here, the solution of Schrödinger's equation is necessary for accurate simulation of the carrier concentration.


next up previous contents
Next: 3. The Silicon/Silicon-Dioxide Interface Up: 2. Simulation of Semiconductor Previous: 2.3 Carrier Generation and

R. Entner: Modeling and Simulation of Negative Bias Temperature Instability