2.2  The Boltzmann Transport Equation

The Boltzmann Transport equation (BTE) has been originally developed by Ludwig Boltzmann to statistically describe transport of atoms and molecules (particles) of an idealized diluted gas. The BTE describes the transport by two processes, namely free streaming and scattering. In order to use the BTE to describe transport of an electron gas in a semiconductor, the BTE for diluted gases needs to be modified. First the free streaming of the charge carriers in the lattice is described by the equations of motion (Newton’s law)

ℏ∂tk = F   and  ∂tx = v,
(2.13)

where the relation p = k and the band structure is used. Second, to describe the collisions between electrons in the electron gas, we will use quantum mechanical perturbation theory (Fermi’s Golden Rule). It has to be noted that in the classical framework of the BTE, where Heisenbergs uncertainty principle is being neglected, one could directly track position and momentum of each electron. However, this tracking of each single particle in a classical approach is currently infeasible.


PICT
Figure 2.4: Illustration of the movement of an electron through a 2D semiconductor. The BTE treats the free streaming (blue lines) classically and the scattering events (red crosses) quantum-mechanically.

Instead we will modify the BTE as stated above, by rederiving the BTE using Equation (2.4). Although the derivation of the BTE has been covered in many works before, its derivation will nevertheless be repeated from  [1312] for the sake of clarity. For this purpose we need to define a distribution function f(x,k,t) of the carriers in real- and reciprocal-space. The distribution function f(x,k,t) is defined such that

      --2--         3   3
dN  = (2π)3f (x, k,t)d xd  k
(2.14)

is the number of particles (electrons or holes) in the infinitesimal small volume d3xd3k in the six-dimensional phase-space. The distribution function is a semi-classical concept, since it assumes that both position and momentum of a particle can be measured to arbitrary accuracy at the same time. In order to introduce the semi-classical concept into the Hamiltonians of Equation (2.4), we treat the electrons as classical particles governed by classical mechanics in the time intervals they do not scatter with other particles. Thus the external force exerted on a single electron is

F(x, t) = ∂t◟ℏ◝k◜◞=  - ∥q∥∇x φout(x,t).
            =p

The group velocity vνg(k) can then be calculated  [12] as

  ν     ∇k
vg(k ) =-ℏ-E ν(k).

Starting from the definition of the density matrix

ρ(x1,x2, t) = ψ (x1, t)⋅ψ(x2,t)
(2.15)

we define the distribution function as the Fourier transformed density matrix

           ∫
                                   3
f(x, k,t) =   exp (- ik⋅x)ρ(x1,x2,t)d k.
(2.16)

This definition is compatible with the previous definition in Equation (2.14), since the density matrix describes a statistical ensemble of quantum states in real space. In order to transform the density matrix to the phase space, a Fourier transform needs to be carried out. To asses the time evolution of the density matrix we utilize the Liouville-Von-Neumann equation,

(H(x1) - H(x2))ρ(x1,x2,t) = tρ(x1,x2,t), (2.17)
itρ(x1,x2,t) + (Eν(-ix1) - Eν(-ix2) (2.18)
- φout(x1,t) + φout(x2,t))ρ(x1,x2,t) = 0, (2.19)
where the simplified and approximated Hamilton-Operators from Equation (2.4) have been used. Scattering will be neglected for the time being, since it will be treated as a perturbation later. After a transformation of coordinates,
    1-             ′
x = 2 (x1 + x2 ), x =  x1 - x2,
(2.20)

the operators can be linearized by assuming that each electron is sharply distributed in k-space and that the electrostatic potential changes over x are small compared to the changes of the electronic wavefunction over x

- φout(x1,t) + φout(x2,t) ≈∇xφout((x2 + x1)2,t)(x2 + x1) = -◟ ∇x-φo◝u◜t(x,t)◞=F(x)∕qx (2.21)
Eν(-ix1) - Eν(-ix2) ivν g(-ix)x (2.22)
Inserting the above approximations into Equation (2.17) one obtains:
          ′       ν                 ′     - 1      ′      ′
iℏ∂tρ(x,x ,t)+ iℏvg (- i∇x ′)∇x ρ(x,x ,t)+ ℏ   F(x)⋅x ⋅ρ (x, x ,t) = 0.
(2.23)

Finally, a Fourier transformation of the equation above is carried out and one obtains after a few algebraic rearrangements the well known homogeneous BTE for the νth band

    ν          ν          ν           -1           ν
∂tf  (x, k,t)+ vg(x,k-)⋅∇xf--(x,k,-t)-- ℏ--F-(x,t)⋅∇kf--(x,k,t) = 0,
              ◟                =L {fν◝(◜x,k,t)}                ◞
(2.24)

where L{fν(x,k,t)} is the free-streaming operator. In order to incorporate scattering processes a semi-classical perturbation term Q{fν(x,k,t)} on the right hand side of the BTE is required

∂tf ν(x, k,t)+ vν(x, k)⋅∇xf ν(x,k, t) - ℏ-1F (x,t)⋅∇kf ν(x,k, t) = Q {fν(x,k,t)}.
             ◟g--------------------◝◜--------------------◞
                              =L {fν(x,k,t)}
(2.25)

Q{fν(x,k,t)} is obtained by perturbation theory and modeled using a statistical description of each scattering process. In the BTE the operator L{fν(x,k,t)} describes the classical free-streaming of the carriers in between scattering events which are described by the scattering operator Q{fν(x,k,t)} (cf. Figure 2.4).

The force F(x,t) due to a gradient of the electrostatic potential or the band edges in the absence of magnetic fields is given by

F (x,t) = - q∇x φout(x, t) = - ∇x (ħEC ∕V ∓ ∥q∥ φ(x,t)(x )),
(2.26)

where qis the sign-less elementary charge q, φ(x,t) the electrostatic potential and ECV the band edge energy for either electrons or holes. Since the electrostatic potential is needed in order to compute the force exerted on each charge carrier, one needs to solve Poisson’s equation and the BTE for electrons and holes self-consistently. The full system of equations thus reads,

(ε(x )∇φ (x, t)) = q(n - p + C), tfν n +  n     ν    -1      ν
v◟-⋅∇xf-n-+◝ℏ◜--F-⋅∇kf-n◞=L{fνn} = Qn{fν n}- Γn{fν n,fν p }, tfν p +  p     ν    -1      ν
v◟-⋅∇xf-p--◝ℏ◜--F-⋅∇kf-n◞=L{fνp} = Qp{fν n}- Γp{fν n,fν p }, (2.27)
where additional scattering terms Γn∕p{fνn,fνp} have been added to account for recombination and generation of electrons and holes. Since each BTE, without further approximations, has to be solved in three spatial and three phase space dimensions in addition to the time dimension, a direct discretization of the full system would lead to prohibitive memory and computational time requirements for many applications. Thus further approximations or alternative discretization schemes have to be employed to solve the system of equations.

2.2.1  Scattering

In this section a short introduction into modelling the scattering term Q{fν(x,k,t)} on the right hand side of Equation (2.25) is given. While traveling through the lattice the electrons interact with the lattice or other electrons. Such interactions, where the electrons change their momentum, are termed scattering events. Considering an infinitesimal small volume d3kd3x in phase space (cf. Figure 2.5), the electrons can either scatter into or out of this volume. Thus the scattering operator is normally split into an in-scattering and out-scattering operator as follows:

    ν            in  ν            out  ν
Q{f  (x, k,t)} = Q  {f (x,k,t)}- Q   {f  (x,k,t)}.


PIC

Figure 2.5: Illustration of scattering events in phase space. Shown is a group of electrons moving from an infinitesimal volume d3kd3x at time t in phase space to d3kd3xat time t+, where the out-scattering of electrons is depicted using blue arrows, in-scattering is illustrated by red arrows and free streaming is indicated with a green arrow.

The transitions in and out of an infinitesimal small volume d3kd3x in phase space do not occur instantly. Nevertheless they are often modeled as an instant transition via a rate of expected scattering charge carriers per second. Additionally it is assumed that the scattering occurs at a certain point in real space and is a local event. For each flavor of scattering there is a rate of transitions Sν(k,k,t) from k to kin the phase space. An incomplete list of interactions considered in this work can be found in Table 2.1. Note that in a homogeneous semiconductor the scattering rate does not depend on the spatial location of the electron. This simplifies the evaluation of the scattering terms. In order for an electron to scatter, the initial state must be occupied by an electron and the final state must be empty by virtue of Pauli’s exclusion principle. With all of the above the scattering operator is often modelled as

Q{fν(x,k,t)} = --1--
(2 π)3 ν B   ν  ′     ν     ′       ν
S◟--(k-,k,t)f-(x,k◝,◜t)(1---f-(x,k,-t))◞Qin{fν(x,k,t)}
-  ν    ′   ν             ν    ′
S◟--(k,k-,t)f-(x,-k◝,◜t)(1---f-(x,k-,t))◞Qout{fν(x,k,t)}d3k, (2.28)
where the integral runs over the first Brillouin zone. In thermal equilibrium Equation (2.28) as scattering operator in the BTE yields the Fermi-Dirac distribution
                    1
fν(x,k,t) = -------(--------)-,
            1+ exp  E-ν(kkB)T-LEF
(2.29)

where EF is the Fermi-level and equals the chemical potential μ known from thermodynamics  [12]. The computational burden to evaluate the integral in Equation (2.28) can strongly depend on how the bandstructure is resolved and becomes easier to evaluate with the parabolic band approximation. Nevertheless, upon solving the BTE the term fν(x,k,t)(1 - fν(x,k,t)) in Equation (2.28) might prove to be too challenging for a certain numerical BTE solver. Thus the Pauli exclusion principle is often dropped and the scattering operator reduces to

Q{fν(x,k,t)} = --1--
(2π)3 ν B Sν(k,k,t)fν(x,k,t)
- Sν(k,k,t)fν(x,k,t)d3k (2.30)
and therefore gives the Maxwell-Boltzmann distribution as equilibrium distribution function
               (              )
fν(x,k,t) = exp  - Eν(k)---EF-  .
                      kBTL
(2.31)

The integral in Equation (2.30) is easier to evaluate than Equation (2.28), especially when using the parabolic band approximation and forms the basis of many important approximations, such as the relaxation time approximation (RTA)  [22]. Dropping Pauli’s exclusion principle is clearly justified if (1 - fν(x,k,t)) 1 and thus fν(x,k,t)) 1 , which is only true for low electron (hole) concentrations and non degenerate semiconductors. Thus dropping Pauli’s exclusion principle in Equation (2.28) is often termed low density approximation.


Interaction Elastic/Inelastic References



Acoustic Phonon Scattering Approx. Elastic  [232425]
Optical Phonon Scattering Inelastic  [232425]
Impurity Scattering Approx. Elastic  [232625]
Impact Ionization Inelastic  [2527]
Electron-Electron Scattering Inelastic  [28]

Table 2.1: A list of particle interactions considered in this thesis including references to the used models for the respective scattering rates.

2.2.2  Recombination and Generation

The scattering operator as described in Section 2.2.1 does not consider electron-hole recombination (cf. Equation (2.25)). In order to be able to consider recombination/generation processes the BTE needs to be split into a BTE for electrons and holes respectively. Since recombination/generation models can be quite complex  [29], we first restrict this section to the most simple two-state defect (cf. Figure 2.6), where only a single function over time, namely the trap occupancy fνT(t) and trap level ET are needed to describe the state of the defect. Additionally it will be assumed that there are enough electrons and holes that are either spin up or spin down, such that the trap occupancy is independent of the electron spin. In case the spin cannot be neglected, the trap occupancy needs to be split into its spin components, fνT(t) fT,upν + fT,downν. With these assumptions the macroscopic rate equation for the trap occupancy reads  [1230]

tfν T(t) = ν B(1 - fν T(t))   n    ν    p         ν
(◟R-ν(k)fn --G◝ν◜(k)(1--fp-))◞r21 (2.32)
+ fν T(t)  p    ν     n        ν
(◟Rν(k)fp---G◝◜ν(k-)(1---fn))◞r12d3k,
where Gνn(k) and Gνp(k) are the number of generated electrons and holes per second per d3k, Rνn(k) and Rνp(k) are the number of recombined electrons and holes per second per d3k.

PIC

Figure 2.6: Left: The trap occupancy is governed by a first order rate equation and two transition rates r12 and r21, which are obtained from the four recombination/generation rates shown in the band diagram. The defect can become charged (state 2) by either capturing an electron from the conduction band or by emitting a hole to the valence band and vice versa. Right: The two state model of a defect located within the band gap (EV < ET < EC). Only in state two does the defect carry a charge, which needs to be considered upon solving Poisson’s equation.

Putting all of the above together, the recombination operators Γp∕n{fνn,fνp} for electrons and holes, which have been introduced in Equation (2.27), read

Γn{fν n,fν p } =  N
---T3
(2π) ν B Gνn(k)fν T(t)(1 - fν n) - Rνn(k)(1 - fν T(t))fν nd3k, (2.33)
Γp{fν n,fν p } = -NT--
(2π)3 ν B Gνp(k)(1 - fν T(t))(1 - fν p ) - Rνp(k)fν T(t)fν p d3k, (2.34)
where NT is the trap concentration. Dropping the sum over all bands in the equations above, the recombination rates for electrons and holes are  [3130]
 n       n n              p       p p
Rν(k) = σ v  (k )  and   R ν(k) = σ v (k),
(2.35)

where σn and σp are experimentally determined capture cross sections and vn and vp reaction velocities. From detailed balance  [3112] the generation rates can be determined

                    (            )                     (            )
Gn (k ) = σnvn (k)exp ET----En(k-) ,Gp (k) = σpvp(k)exp  En-(k)---ET-  .
 ν                      kBTL         ν                      kBTL
(2.36)