next up previous contents
Next: 4. Monte Carlo Technique Up: Dissertation Gerhard Karlowatz Previous: 2. Strained Band Structure

Subsections



3. The Semiclassical Transport Model

As computational power raises and more efficient new Monte Carlo algorithms are developed, the Monte Carlo approach to solve the Boltzmann equation for TCAD device simulation is getting more and more important. In this work the Boltzmann equation - originally developed to describe the flow of kinetic gases - is used with extensions to meet the properties of quantum mechanical transport occurring for electron or hole ensembles in crystals. These extensions include particle kinetics depending on a position-dependent band structure and on scattering events, which are calculated quantum-mechanically using Fermi's Golden Rule. The carrier motion consists of periods of collisionless acceleration caused by external forces, interrupted by instantaneous scattering events. The Monte Carlo approach solves the semiclassical Boltzmann transport equation [Kosina00].

During this work our in-house Monte Carlo tool named VMC [IuE06] was further developed in general and extended by a FBMC part. This chapter gives an overview of the used Monte Carlo models and algorithms with focus on the numerical methods used for CPU time efficient FBMC simulation.


3.1 The Equations of Motion

The semiclassical Hamilton function for an electron in a conduction band is given by

(3.1)

where is the conduction band edge, is the energy of the -th band relative to the conduction band edge, is the electrostatic potential, is the position, is the wave vector of the carrier, and e the elementary charge. The Hamilton function for holes reads

(3.2)

where is the valence band edge. The collisionless motion of the carriers is described by the equations of motions given by Newton's law

(3.3)

and the carrier group velocity

(3.4)

Here, is the reduced Planck constant, denotes the force and the group velocity. In the semiclassical Monte Carlo framework the velocity for a carrier in the band is the group velocity of the wave packet of the carrier and follows from (3.4)

(3.5)

The force denotes

$\displaystyle \mathbf{F}_n^\mathrm{e}(\mathbf{r},\mathbf{k},t) = \nabla_\mathbf...
...\mathbf{r})+\mathrm{e}\psi(\mathbf{r},t)-\varepsilon _n(\mathbf{r},\mathbf{k}))$ (3.6)

for electrons and

$\displaystyle \mathbf{F}_n^\mathrm{h}(\mathbf{r},\mathbf{k},t) = \nabla_\mathbf...
...\mathbf{r})-\mathrm{e}\psi(\mathbf{r},t)-\varepsilon _n(\mathbf{r},\mathbf{k}))$ (3.7)

for holes.


3.2 The Boltzmann Transport Equation

Because of the nature of scattering as a random process it is impossible to determine the path of a carrier exactly. Instead a stochastic approach is used where the carrier gas is described by a distribution function . This distribution function is related to the total number of particles in the system by

(3.8)

The integral in real space is hereby over the whole device domain whereas the integral in -space is over the first Brillouin zone (BZ), is the minimum phase-space volume of a particle and the factor of two originates from the two possible spin states of the carriers. The semiclassical Boltzmann equation reads then

$\displaystyle \left\{ \frac{\partial} {\partial t}+ \mathbf{F}_n^\mathrm{T}(\ma...
...right\} f_n(\mathbf{r},\mathbf{k},t) = S_\mathrm{SC}(\mathbf{r},\mathbf{k},t) .$ (3.9)

The left hand side is derived from the equations of motion (3.2) and (3.3). is the scattering integral given by

\begin{gather*}\begin{split}S_\mathrm{SC} = \frac{\Omega} {(2\pi)^3}\sum_{n'}\in...
...)(\mathbf{r},t) f_{n}(\mathbf{r},\mathbf{k}',t)d^3k'. & \end{split}\end{gather*} (3.10)

describes the transition from a state into a state and the inverse process. The rate for a transition from an initial state to a final state is proportional to the probabilities that the initial state is occupied, , and that the final state is not occupied, , and to the transition rate . The factor stems from the Pauli exclusion principle.

The Boltzmann equation in the form of (3.9) is non-linear, because the transition rate may depend on the carrier distribution and the scattering integral includes a product of the distribution function with itself. The latter can be avoided if the Pauli exclusion principle is neglected. If additionally it is assumed, that the transition rate does not depend on the distribution function, we achieve the linear Boltzmann equation

\begin{gather*}\begin{split}\left\{ \frac{\partial} {\partial t} + \mathbf{F}_n^...
...vert\mathbf{k}') f_{n'}(\mathbf{r},\mathbf{k}',t)d^3k'. \end{split}\end{gather*} (3.11)

The scattering rate is the rate at which carriers are scattered out of their initial state and is defined as

(3.12)

Equation (3.11) describes the kinetics of a carrier ensemble where the particles are considered independent and noninteracting.


3.3 Scattering Mechanisms

In this work scattering is treated on the basis of Fermi's Golden Rule [Landau81]

(3.13)

Here, is the transition probability from the initial state i to the final state f and is the energy of the final state. The matrix element is given by

(3.14)

where and are the wave functions of the initial and final state, respectively. is the perturbation potential.

The density of states per spin of a band is given by

(3.15)

The density of states integral can be evaluated numerically as for FBMC simulation or approximated by analytical expressions. Following the approach of Jacoboni [Jacoboni83] an analytical description for the conduction bands is derived by approximating the minima of the conduction bands - the so called valleys - by using the bandform function

\begin{gather*}\begin{split}\gamma^v(\varepsilon ) = \frac{\hbar^2}{2} \sum_{i,j...
...repsilon \left( 1 + \alpha^{v} \varepsilon \right)   , \end{split}\end{gather*} (3.16)

where denotes the valley index and is the energy relative to the valley energy offset . One can include strain effects by introducing for each valley strain-dependent effective mass tensors , nonparabolicity coefficients , and valley energy offsets. If is set to zero the shape of the band approximation simplifies from nonparabolic to parabolic. The density of states of a nonparabolic band can be written as

$\displaystyle g^{v}\left(\varepsilon \right)=\frac{1}{\sqrt{2}} \frac{\{m_\math...
...ar^{3}}\sqrt{\gamma^{v}\left(\varepsilon \right)}(1+2\alpha^{v}\varepsilon ) ,$ (3.17)

where is the density of states effective mass of the -th valley, which can be obtained from the effective mass tensor

$\displaystyle m^v_\mathrm{dos}= \sqrt[3]{ m_{11}^v m_{22}^v m_{33}^v }  .%\vphantom{\sum_i}
$ (3.18)

The approach to use valley-dependent scattering models can be adapted from Monte Carlo with analytical band structure models to fit into the framework of FBMC [Jungemann03]. The first Brillouin zone of the first conduction band of Si is divided into six volumes as defined in Table 3.1 where indices the -th valley in the -th band. The same approximation is also applied to the higher conduction bands. This approach is in the spirit of the analytical many valley model  [Jacoboni83]. In combination with constant matrix elements it gives a CPU efficient formulation of the scattering rates, because scattering rates are proportional to the density of states, which is calculated numerically from the full-band structure.


Table 3.1: The six volumes in the first Brillouin representing the six X-valleys of the first conduction band.
Volume Boundaries



3.3.1 Phonon Scattering

The transition rate from an initial state ( ) to a final state ( ) for phonon scattering in a non-polar semiconductor can be written as [Jacoboni83]

$\displaystyle \{\tiny S^{ \shortstack{abs  [-2pt] emi }} \} ^{v'v}(\mathbf{k}...
...repsilon ^{v'}(\mathbf{k}') - \varepsilon ^{v}(\mathbf{k}) \mp \hbar\omega_q] $ (3.19)

Here, the upper and lower signs denote phonon absorption and emission, respectively. The rate depends on the the phonon number , the momentum transfer $ \mathbf{q} = \mathbf{k} -
\mathbf{k}'$ , the deformation potential tensor , the mass density of the crystal , the overlap integral , the phonon angular frequency and its polarization .

The overlap integral

(3.20)

depends on the type of transition. For intravalley transitions of electrons it is common to set to unity, which is exact only for wave functions of pure -states or for exact plane waves [Jacoboni83]. The lowest conduction band of cubic semiconductors is a mixture of a and -type states and so overlap factors less than unity are obtained. Since for both intra- and intervalley transitions the overlap factors were found to be almost constant [Reggiani73] the values for can be included in the coupling constants.

The phonon number is given by the Bose-Einstein statistics

(3.21)

where denotes the lattice temperature and the Boltzmann constant.

Acoustic Intravalley Scattering at Room Temperature

Since at room temperature the acoustic phonon energy is very small compared to the thermal energy the expression for the transition rate for acoustic intravalley scattering can be simplified by using the elastic and equipartition approximation. Within the latter approximation the Bose-Einstein statistics (3.21) is replaced by its first order Taylor expansion which gives for the phonon population

(3.22)

The phonon dispersion relation for small is approximated as . Thus, (3.19) becomes

$\displaystyle \{ S_\mathrm{ac}^{\tiny\shortstack{abs  [-2pt] emi }} \} ^{v}(\...
...[\varepsilon ^{v}(\vect{k}') - \varepsilon ^{v}(\vect{k})\pm \hbar \omega_q] ,$ (3.23)

where denotes the average sound velocity and is the mass density of the crystal. is the acoustic deformation potential of the -th valley, which is derived by averaging the two non-zero elements of the deformation potential tensor over the polar angle [Jacoboni83].

In the elastic approximation the phonon energy is neglected, $ \hbar
\omega_q = u_\mathrm{s}q \to 0$ . Therefore, emission and absorption processes are equivalent and the transition probabilities can be added. This leads to a scattering rate for acoustic intraband scattering which is a function of energy only

$\displaystyle \{S_\mathrm{ac} \}^v\left(\varepsilon \right)=\frac{2\pi \ensurem...
...\{\Xi_\mathrm{adp}^{v}\}^2}{\hbar u_{\mathrm{s}}^{2}\rho}g^{v}(\varepsilon ) .$ (3.24)

In (3.24), denotes the density of states of valley . The values of the parameters used in (3.24) can be found in Table [*].

Table 3.2: Parameters for acoustic and optical intravalley phonon scattering in Si.
Symbol Value Units
g/cm
cm/sec
eV
meV
10 eV/cm

 [Jacoboni83],  [Jungemann03],  [Fischetti96b]


Acoustic Intravalley Scattering at Low Temperatures

When the lattice temperature is very low (3.22) does not hold anymore and the dependence on the acoustic phonon energy and momentum transfer have to be considered. Using a model in which the phonon dispersion has to be evaluated during simulation will demand long calculation times. Therefore it is useful to derive a temperature-dependent but otherwise constant mean phonon energy and formulate the acoustic intraband scattering rate as [R.-Bolívar05] [Bufler01]

$\displaystyle \{ S_\mathrm{ac}^{\tiny\shortstack{abs  [-2pt] emi }} \} ^v\lef...
...} \mp \frac{1}{2} \right ) g^{v}(\varepsilon \pm \hbar \omega_{\mathrm{ac}}) .$ (3.25)

To obtain the average momentum transfer a mean momentum transfer is calculated first by taking an average over the solid angle

$\displaystyle \overline{q}(\varepsilon )= \frac{1}{4 \pi} \oint qd\Omega = \fra...
...int k \sqrt{1-2\cos(\vartheta)} \sin (\vartheta) d \vartheta =\frac{4}{3}k   .$ (3.26)

This result is used to take a second average with the equilibrium distribution function, i.e., the Maxwell-Boltzmann distribution

$\displaystyle \left< q \right> = \frac{\displaystyle\int_0\nolimits^\infty \ove...
...{{\mathrm{k_B}}}\ensuremath {T_\mathrm{L}}} g(\varepsilon )d \varepsilon }   .$ (3.27)

With a parabolic band approximation evaluates to

(3.28)

where is the effective electron mass. Now, the average phonon energy is obtained by assuming a linear dispersion relation

(3.29)

Optical Intravalley Scattering

From the matrix element theorem one can derive that optical intravalley scattering occurs only in the conduction band valleys along the $ \langle
111\rangle$ directions [Harrison56]. Thus this type of scattering is important in Ge. In Si it contributes only at high electron energies.

By replacing in (3.19) with a squared optical coupling constant the scattering probability can be reformulated with a squared optical coupling constant  [Jacoboni83]. The optical phonon energies and the phonon number can be assumed to be constant since the dispersion curve is nearly flat for phonons involved in optical intraband transitions. If the overlap factor is lumped in the coupling constant the resulting transition rate can be written as

$\displaystyle \{ S_\mathrm{op}^{\tiny\shortstack{abs  [-2pt] emi }} \} ^{v}(\...
...on ^{v}(\vect{k}') - \varepsilon ^v(\vect{k}) \mp \hbar\omega_{\mathrm{op}}] .$ (3.30)

With the above formulation the scattering rate for optical phonons is a function of the final energy $ \varepsilon \pm\hbar
\omega_{\mathrm{op}}$

$\displaystyle \{ S_\mathrm{op}^{\tiny\shortstack{abs  [-2pt] emi }} \} ^v\lef...
...{1}{2} \right ) g^{v}\left(\varepsilon \pm \hbar \omega_{\mathrm{op}}\right) .$ (3.31)

The values used for optical intravalley scattering are given in Table [*].


Intervalley Phonon Scattering

Both acoustic and optical phonons can cause electron transitions between states in different conduction band valleys [Harrison56,Conwell67]. The scattering rate for intervalley scattering out of a valley for a phonon mode reads

$\displaystyle \{ S_\eta^{\tiny\shortstack{abs  [-2pt] emi }} \} ^{v}\left(\va...
...(\varepsilon ^{v'} \pm \hbar \omega_\eta - \Delta \varepsilon ^{v'v}\right)  .$ (3.32)

The possible final valleys are determined by two selection rules for the phonon mode : -type phonons induce transitions between opposite valleys on the same axis in space, and -type phonons induce transitions among orthogonal axes. The coupling constants depend on the phonon branch and the initial and final valley in a particular transition. denotes the number of possible final equivalent valleys in a transition, is the phonon number, and is the energy difference between the minima of the final and initial valley.

The numerical values for the bulk phonon scattering rates are summarized in Table 3.3.

Figure 3.1 depicts the low field electron mobility over temperature. This result is obtained by Monte Carlo simulation including only phonon scattering and agrees very well with simulation data from [Canali75]. Because of the influence of impurity scattering the mobility obtained from experimental data is lower in the low temperature regime.

Figure 3.1: Low field electron mobility in Si versus lattice temperature as a result of Monte Carlo simulation (VMC) compared to experimental and theoretical data [Canali75].
\includegraphics[width=3.6in]{xmgrace-files/MobTemp2.eps}


Table 3.3: Phonon modes, coupling constants, phonon energies, and selection rule for Si as used in the analytical intervalley phonon scattering model. Values are taken from [Jacoboni83].
Phonon Mode Selection Rule
Transversal acoustic 50 12.06 g
Longitudinal acoustic 80 18.53 g
Longitudinal optical 1100 62.04 g
Transversal acoustic 30 18.86 f
Longitudinal acoustic 200 47.39 f
Transversal optical 200 59.03 f


Full-Band Phonon Scattering

As explained in the introduction of Section 3.3, the many-valley approach for analytical band models is adapted for the full-band framework. The differences in the formulation of the scattering rates lie then in the analytical versus the numerical calculation of the density of states and the implementation of interband transitions.

Table 3.4: Phonon modes, coupling constants, phonon energies, and phonon branch of inelastic phonon scattering for Si [Dhar06] as used in FBMC simulation.
Phonon Mode Selection Rule
Transversal acoustic 47.2 12.1 g
Longitudinal acoustic 75.5 18.5 g
Longitudinal optical 1042 62.0 g
Transversal acoustic 34.8 19.0 f
Longitudinal acoustic 232 47.4 f
Transversal optical 232 58.6 f
Holes acoustic 991 63.3 a



Table 3.5: Phonon modes, coupling constants, phonon energies, and phonon branch of inelastic phonon scattering for Ge [Jungemann03] as used in FBMC simulation.
Phonon Mode Selection Rule
Transversal acoustic 47.9 5.6 g
Longitudinal acoustic 77.2 8.6 g
Longitudinal optical 928 37.0 g
Transversal acoustic 283 9.9 f
Longitudinal acoustic 1940 28.0 f
Transversal optical 1690 32.5 f
Holes acoustic 3500 37.0 a


The transition rate of full-band acoustic intravalley scattering is derived by applying the elastic approximation to (3.23)

$\displaystyle \{ S_\mathrm{ac} \}^{n'n}(\vect{k}'\vert\vect{k})= \frac{2\pi}{\h...
... \varepsilon _n^v(\vect{k}))\delta_{v_{n'}(\mathbf{k}'),v_{n}(\mathbf{k})}   .$ (3.33)

The deformation potentials are assumed to be 8.5 eV for electrons and 5.12 eV [Dhar06] for holes in Si [Dhar06] and 8.79 eV for electrons and 7.40 eV for holes in Ge [Jungemann03]. The Kronecker delta term on the right hand side defines that transitions are allowed only within a valley , but between different bands and . The probability to scatter to another band is determined by the contribution of the density of states in this band at the final energy.

For the simulation of SiGe alloys the parameter values for Si and Ge are linearly interpolated according to the material composition. Since only the -valleys are considered in the the implemented full-band scattering formalism, simulation of SiGe compounds is only valid as long as the -valleys are dominantly populated.

The coupling constants, phonon energies, and phonon modes , and the selection rule for inelastic full-band phonon scattering in Si are shown in Table 3.4. The coupling constants are taken from  [Jacoboni83] and [Jungemann03] and are fine-tuned to match the measured data for biaxial strained Si [Dhar06]. Table 3.5 shows the respective values for Ge, which are used to calculate the interpolated parameter values of SiGe alloys.

The phonon branches determine a set of selection rules labeled . In the full-band formulation these selection rules act on the density of states whereas the coupling constant is kept constant for all combination of valleys and bands. This leads to the expression

(3.34)

$\displaystyle = \frac{\pi \{D_tK_\eta\}^2}{ V \rho \omega_\eta} \left (N_{\eta}...
...bar \omega_\eta - \varepsilon _n(\vect{k}))r(\eta,v(\mathbf{k'}),v(\mathbf{k}))$    

for the transition rate due to intervalley phonon scattering. For holes there is no restriction in the selection of the final state by a selection rule and there is only one inelastic optical phonon mode.

Figure 3.2 shows the electron velocity for relaxed Si as a function of the electric field in direction. It can be observed that the data from VMC agrees well with measurements.

Figure 3.3 depicts the hole velocity as a function of the electric field and Figure 3.4 the energy as a function of the electric field in direction for relaxed Ge. These results are compared to values from literature [Fischetti96a][Yamada95][Ghosh06] and show good agreement.

Figure 3.2: Electron velocity versus field in [100] direction for relaxed Si. Monte Carlo results (VMC) are compared to measurement [Canali75].
Figure 3.3: Hole velocity versus field in [100] direction for relaxed Ge. Monte Carlo results (VMC) are compared to results from literature [Fischetti96a][Yamada95][Ghosh06].
\includegraphics[width=3.6in]{figures/fieldcmp.eps}


\includegraphics[width=3.74in]{ECSfigures/highVelocityComp2.eps}
Figure 3.4: Hole energy versus field in [100] direction for relaxed Ge. Monte Carlo results (VMC) are compared to results from literature [Fischetti96a][Yamada95][Ghosh06].
\includegraphics[width=3.7in]{ECSfigures/highEnergyComp2.eps}

3.3.2 Ionized Impurity Scattering

In this work the well known model of Brooks and Herring [Brooks51] is used in an extended form, where multi-potential scattering and dispersive screening is included. The Fourier transformed potential of a screened, ionized impurity is given by

(3.35)

Here, is the charge of the impurity center and is the inverse Thomas-Fermi screening length

$\displaystyle \beta^{2}_\mathrm{s}=\frac{\mathrm{e}^{2} n_\mathrm{I}}{\epsilon_...
...c{\mathcal{F}_{-1/2}\left(\eta\right)}{\mathcal{F}_{1/2}\left(\eta\right)}   ,$ (3.36)

where is the concentration of the impurity centers and denotes the Fermi integral of the -th order with the reduced Fermi energy as argument

(3.37)

Application of the Golden Rule (3.13) and the scattering potential (3.35) gives the transition rate

$\displaystyle \{ S_\mathrm{BH} \}^{n'n}(\vect{k}'\vert\vect{k})= \frac{2\pi}{\h...
...)^2} \delta(\varepsilon _{n'}^{v}(\vect{k}') - \varepsilon _n^v(\vect{k}))   .$ (3.38)

The ionized impurity scattering rate from the above potential can be formulated [Kosina98]

(3.39)

Here, and the prefactor is set

(3.40)

where denotes the magnitude of the group velocity. In the approximation based on non-parabolic analytic bands, the scattering rate for a valley evaluates to

$\displaystyle S_{\mathrm{BH}}^{NP}(\varepsilon ) = \frac{n_\mathrm{I} Z^2 \math...
...lpha^v \varepsilon }{1+\frac{4\gamma(\varepsilon )}{\varepsilon _{\beta,v}}} ,$ (3.41)

with

(3.42)

Multi-Potential Scattering and Dispersive Screening

The Brooks-Herring model significantly overestimates the mobility for higher impurity concentrations. To extend the validity of the model, multi-potential scattering and dispersive screening via Lindhard's dielectric function is included. Multi potential scattering to the first order considers only the Coulomb interaction of the carrier with pairs of impurities. The Fourier transform of the applied potential takes the form:

$\displaystyle V_{0}\left(\mathbf{q}\right)=\frac{Z \mathrm{e}}{4\pi \epsilon_{0...
...ert\mathbf{q}\vert^2} \left(1+\exp\left(-i\mathbf{q}\mathbf{R}\right)\right) ,$ (3.43)

where is the distance between the centers.

is the total potential which forms because of the response of the electrons to the applied potential . The total potential is related to the applied potential via the dielectric function

(3.44)

The frequency equals zero because the applied potential is time independent. When considering only low order screening effects, Lindhard's dielectric function is appropriate [Ferry91]:

(3.45)

Here, denotes the screening function for which an integral representation exists [Ferry91]

$\displaystyle G\left(\xi,\eta\right)=\frac{1}{\mathcal{F}_{-1/2}\left(\eta\righ...
...xp\left(x^{2}-\eta\right)}\ln\biggl\vert\frac{x+\xi}{x-\xi}\biggr\vert dx   ,$ (3.46)

with

(3.47)

The integral (3.46) cannot be evaluated analytically, but there are attempts in literature to approximate it with a sufficiently accurate rational expression [Kosina98]. After combining equations (3.43)(3.44) and (3.45) and averaging the term over the solid angle is found as

$\displaystyle \vert V_\mathrm{t}(q)\vert^2=\left(\frac{Z\mathrm{e}} {4\pi \epsi...
...thbf{s}}G(\xi,\eta)\right)^2} \left(1+\frac{\sin\left(qR\right)}{qR}\right)  .$ (3.48)

Figure 3.5: Electron low field mobility versus doping concentration for Si. Experimental data [Masetti83] are compared to Monte Carlo results. An ionized impurity model which includes a two-ion potential and dispersive screening is used for the Monte Carlo simulation.
\includegraphics[width=0.62\linewidth, clip]{figures/DopingMobSi.eps}

Equivalent Scattering Cross Section

The long range of the Coulomb force causes a large scattering cross section of a single ion. This makes Coulomb scattering occur very frequently and consume a high amount of computation time during a Monte Carlo simulation. The momentum transfer per scattering event is rather small. This leads to a very anisotropic scattering behavior with a high percentage of small-angle scattering events.

The number of scattering events can be significantly reduced by the introduction of an isotropic equivalent scattering cross section $ \tilde
\sigma$ , which has the same momentum relaxation time as the original cross section . These cross sections are related by [Kosina97]

(3.49)

where is the angle between and . The equivalent total scattering rate is

$\displaystyle \tilde S_\mathrm{BH}=2\pi n_{\mathrm{I}}v_\mathrm{g}(k)\int_{-1}^...
...ft(1-\cos\theta\right) \sigma\left(k,\cos\theta\right) \mathrm{d}\cos\theta .$ (3.50)

Using the potential (3.35) in Fermi's golden rule and integrating over the final states the equivalent total scattering rate is found to be [Kosina97]

(3.51)

 Figure 3.5 shows the low field electron mobility versus the doping concentration in Si. The Monte Carlo result is achieved by using a modified Brooks-Herring scattering model, which includes a two-ion potential and dispersive screening. The result shows fairly good agreement with experimental data [Masetti83]. Further improvements can be achieved by taking the second Born correction and plasmon scattering into account [Kosina98].

3.3.3 Impact Ionization

Impact ionization is modeled using a modified threshold expression [Cartier93]

$\displaystyle \renewedcommand{arraystretch}{0.0} S^\mathrm{II}= \left\{ \begin{...
...th} + \varepsilon _\mathrm{OS} < \varepsilon \end{array} \right. \vspace{0.1cm}$ (3.52)

where is the impact ionization scattering rate, and are threshold energies, and and are prefactors which determine the softness of the threshold. The value of the offset energy is chosen to render the scattering function continuous.

The parameters are tuned to reproduce measured electron velocity field characteristics [Canali75] and impact ionization coefficients [Slotboom87][Overstraeten70]
[Maes90] for relaxed Si: , $ \varepsilon ^2_{\mathrm{th}}=
\varepsilon _\mathrm{g}+444   \mathrm{meV}$ , , $ P_2 =
3.4\cdot10^{12}  {1}/{\mathrm{s}}$ , and $ \varepsilon _\mathrm{OS} = 622  
\mathrm{meV}$ . For strained Si the threshold values are adjusted in accordance with the bandgap change. After an impact ionization scattering event is evaluated the overall final energy is randomly distributed between the hole and the primary and secondary electrons.

Fig 3.6 depicts the impact ionization coefficient of electrons in Si as a function of the inverse electric field. The simulation result agrees very well with various measured values from literature.

Figure 3.6: Impact ionization coefficient versus the inverse electric field. Monte Carlo result (VMC) is compared to measurements [Overstraeten70] [Slotboom87][Maes90].
\includegraphics[width=3.7in]{figures/IICoeff.eps}


next up previous contents
Next: 4. Monte Carlo Technique Up: Dissertation Gerhard Karlowatz Previous: 2. Strained Band Structure

G. Karlowatz: Advanced Monte Carlo Simulation for Semiconductor Devices