PHYSICAL REVIEW B 91, 235123 (2015)

Ab initio electronic transport model with explicit solution to the linearized

Boltzmann transport equation

Alireza Faghaninia
Department of Energy, Environmental, and Chemical Engineering, Washington University, St. Louis, Missouri 63130, USA

Joel W. Ager III

Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA

Cynthia S. Lo*
Department of Energy, Environmental, and Chemical Engineering, Washington University, St. Louis, Missouri 63130, USA
(Received 21 January 2015; revised manuscript received 15 April 2015; published 12 June 2015)
Accurate models of carrier transport are essential for describing the electronic properties of semiconductor
materials. To the best of our knowledge, the current models following the framework of the Boltzmann transport
equation (BTE) either rely heavily on experimental data (i.e., semiempirical), or utilize simplifying assumptions,
such as the constant relaxation time approximation (BTE-cRTA). While these models offer valuable physical
insights and accurate calculations of transport properties in some cases, they often lack sufficient accuracy
particularly in capturing the correct trends with temperature and carrier concentration. We present here a transport
model for calculating low-field electrical drift mobility and Seebeck coefficient of n-type semiconductors, by
explicitly considering relevant physical phenomena (i.e., elastic and inelastic scattering mechanisms). We first
rewrite expressions for the rates of elastic scattering mechanisms, in terms of ab initio properties, such as the
band structure, density of states, and polar optical phonon frequency. We then solve the linear BTE to obtain
the perturbation to the electron distributionresulting from the dominant scattering mechanismsand use this
to calculate the overall mobility and Seebeck coefficient. Therefore, we have developed an ab initio model for
calculating mobility and Seebeck coefficient using the Boltzmann transport (aMoBT) equation. Using aMoBT,
we accurately calculate electrical transport properties of the compound n-type semiconductors, GaAs and InN,
over various ranges of temperature and carrier concentration. aMoBT is fully predictive and provides high
accuracy when compared to experimental measurements on both GaAs and InN, and vastly outperforms both
semiempirical models and the BTE-cRTA. Therefore, we assert that this approach represents a first step towards
a fully ab initio carrier transport model that is valid in all compound semiconductors.

DOI: 10.1103/PhysRevB.91.235123 PACS number(s): 72.20.i, 73.61.Ey, 31.15.A, 71.20.Nr

I. INTRODUCTION are governed by band conduction. The first category of BTE-

based models are commonly known as single parabolic band
Accurate models of carrier transport are essential for
models, even though the treatment of the conduction band may
describing the electronic properties of semiconductor ma-
not be explicitly parabolic. These models can be described as
terials, which are particularly important for clean energy
semi-empirical, since experimentally measured parameters,
applications ranging from photovoltaics to thermoelectrics
such as the electron or hole effective mass, band gap, dielectric
to photoelectrocatalysts. There has been an increased focus
constant, and polar optical (PO) phonon frequency, are used
on using compound semiconductors, including those that are
in closed-form expressions for the various scattering rates.
degenerately and heavily doped, for these applications. To
Note that the overall mobility due to elastic scattering is
better understand existing materials and discover new ones,
calculated by averaging, according to Matthiessens rule,
a fully predictive model that correlates electronic structure
the mobilities due to each scattering contribution. The main
to properties is essential. Unfortunately, to the best of our
adjustable parameter in these models is the effective mass,
knowledge, no model, based on ab initio calculations, currently
which can be varied to fit the calculated transport properties
exists to fully capture the elastic and inelastic scattering
to the experimental measurements. While such models often
effects of charge carriers; as a result, errors arise when
impressively capture the changes in properties over various
utilizing the current models. While an ab initio model will
ranges of temperature and carrier concentration, they are
certainly improve our understanding of the carrier transport
restricted to the materials for which experimental data are
mechanism(s) in existing semiconductors, it can also aid in
available; therefore, the predictability of such models is very
the search for high-performing materials by improving the
limited.
accuracy of high-throughput computations [1,2].
There are numerous examples of models in this category
There currently exist two main categories of models, based
[37], such as that by Ehrenreich [6], who modeled the GaAs
on the Boltzmann transport equation (BTE), for calculating the
band structure and PO-phonon scattering by reviewing the
conductivity and Seebeck coefficient of semiconductors that
experimental data [6], and that by Sankey et al. [5], who
considered the effects of resonance, ionized impurity, and
polar optical phonon scattering in GaAs. In these models, all
* of the scattering mechanisms are commonly treated using the
clo@wustl.edu

1098-0121/2015/91(23)/235123(14) 235123-1 2015 American Physical Society

FAGHANINIA, AGER III, AND LO PHYSICAL REVIEW B 91, 235123 (2015)

relaxation time approximation (RTA); here, the relaxation time properties, such as conductivity and Seebeck coefficient, as
is written as a power law function of energythus the details a function of temperature or carrier concentration. Validation
of elastic and inelastic scattering (e.g., PO phonon) captured of the theory against experimentally measured properties will
by the ab initio band structure are disregarded. Scattering thus give insight into which scattering effects are dominant.
rates, particularly inelastic ones, have already been shown to In this paper, we present a band transport model for
not follow such power law distributions [3,8], so the basic calculating low-field electrical drift mobility and Seebeck
assumptions fail. Even in cases where the BTE is explicitly coefficient of n-type semiconductors which we refer to as
solved for PO phonon and the perturbation to the electronic aMoBT: ab initio model for calculating mobility and Seebeck
distribution is obtained without the RTA assumption, the coefficient using the Boltzmann transport (aMoBT) equation.
results are still heavily dependent on available experimental We then validate aMoBT by calculating the properties of two
data. As an example, Miller et al. [8] used the latter III-V semiconductors, GaAs and InN, with different carrier
approach to calculate the mobility and Seebeck coefficient concentrations over various temperatures, and comparing them
of InN samples, which had been grown by molecular beam to experimental values as well as those calculated using the
epitaxy (MBE) and plasma assisted MBE so that all exhibited other transport models described above. We choose these
heteroepitaxial growth with linear charged dislocations; thus materials because the ab initio band structure of GaAs is
these dislocations were found to be the limiting scattering similar to those used in the earlier semiempirical models as
mechanism. it can be reasonably well described with a single band model,
The second category of BTE-based models relies on the whereas the ab initio band structure of InN and the limiting
ab initio band structure of the material, rather than specific scattering mechanisms are quite different; thus these two
experimentally measured parameters, but generally utilizes materials allow us to bracket the range of expected behavior
the relaxation time approximation to the BTE (BTE-RTA) as of our proposed model.
a simplification. Restrepo et al. [9] calculated the mobility
of n-doped silicon at different electron concentrations in II. CARRIER TRANSPORT MODEL
BTE-RTA and ab initio framework where electron-phonon
interactions are treated as elastic with the electron distribution A. Solution to the Boltzmann transport equation
unchanged from the equilibrium Fermi-Dirac. On the other In order to calculate the mobility and Seebeck coefficient,
hand, the constant relaxation time approximation (BTE-cRTA) we solve the Boltzmann transport equation (BTE) using Rodes
simplifies the equation even more, which enables closed form iterative method [3,8,1117] (Appendix A 2) to obtain the
expressions for both conductivity divided by relaxation time electron distribution in response to a small driving force (e.g.,
and Seebeck coefficient. The advantage of these models is a small electric field or a small temperature gradient). It is
the ability to calculate properties of new materials, for which important to note that we do not use the relaxation time
experimental data is unavailable. This type of model works approximation (RTA) in solving the BTE, so neither a variable
well for some materials for which the relaxation time is fairly nor a constant relaxation time appears in this expression. Due
constant, as evidenced by the work of Madsen and Singh to the assumption of a small driving force, we aim to calculate
[10]. However, inelastic scattering mechanisms change the only the linear response to the perturbation; thus the general
electron energy and directly affect the distribution. Lumping form of the electron distribution remains the equilibrium
all the elastic and inelastic scattering mechanisms into a Fermi-Dirac distribution. We can then write
single constant and assuming an equilibrium Fermi-Dirac
f (k) = f0 [ (k)] + xg(k), (1)
distribution in BTE-cRTA framework greatly damages the
predictive ability of such models; as an example, transport where f is the actual distribution of the electrons, including
properties in some cases are very far from experimental both elastic and inelastic scattering mechanisms, f0 is the
measurements. Furthermore, the relaxation time constant is equilibrium Fermi-Dirac distribution, x is the cosine of the
usually determined by fitting the calculated conductivity to angle between the small driving force and k, g (k) is the
experimental data. It should be noted that the calculation of perturbation to the distribution caused by the small driving
this constant is not necessary when calculating the Seebeck force, and finally k = |k|. For the sake of simplicity, we express
coefficient. This is due to the simplifying assumptions that the the conduction band as the average energy of the electrons as
relaxation time is both a constant and direction independent a function of distance, k, from the conduction band minimum
[10], which does not always hold. Therefore, BTE-cRTA (CBM) which is often at the center of the Brillouin zone (i.e.,
suffers not only from inaccuracy in predicting the changes of  point); furthermore, we assume that the small driving force
properties with temperature or carrier concentration in many is aligned with k (i.e., x = 1). Although this is similar in
materials, but also from lack of pure predictability since it spirit to the isotropic band assumption, we take the anisotropy
still relies on experimental data for the computation of the into account by averaging the energy values of the ab initio
relaxation time. calculated band structure, (k), as a function of k rather than
Instead, we propose that accurate calculations of electronic explicitly including k in every direction. Alternatively, if we
transport properties, within the Boltzmann transport frame- wish to consider the directional transport properties, we can
work, are possible by combining relevant treatment of the include the calculated band structure only in that specific
elastic and inelastic scattering mechanisms with ab initio direction. Here, we will focus on calculating and reporting
calculations of the electronic and phonon band structures. the overall average mobility and Seebeck coefficient.
Ultimately, an ab initio theory for carrier transport will need Our goal is to calculate the perturbation to the distribution
to qualitatively and quantitatively predict trends in material [3], g (k). In the reformulated Boltzmann transport equation

235123-2
AB INITIO ELECTRONIC TRANSPORT MODEL . . . PHYSICAL REVIEW B 91, 235123 (2015)

shown in Eq. (2), there are scattering-in, Si (g), and scattering- model. In other words, the main difference between our
out, So , terms for inelastic scattering mechanisms. However, proposed carrier transport model and previous semiempirical
these terms also depend, in turn, on the electronic distribution models [36,8,1114,1820] is the use of ab initio parameters
as well as elastic scattering rates, el . Therefore, the BTE must instead of experimentally measured electron effective mass,
be solved self-consistently to obtain g (k): band gap, etc., which eliminates the need for theories such
    as k p to describe the nonparabolicity or anisotropy of
Si [g(k)] v (k) f z
eE
f
k the conduction band. Instead, for calculating the overlap
g (k) = , (2) integral, we express the conduction band wave function as a
So (k) + el (k)
linear combination of s-type and p-type basis functions, with
where E is the low electric field and v (k) is the electron group coefficients of a and c, respectively [3]. These coefficients can
velocity. The derivation of the BTE in the form shown in Eq. (2) be directly calculated ab initio without the need to assume an
can be found in the literature [3]. The inelastic scattering s-like conduction band wave function (i.e., no assumption of
mechanism that tends to dominate at room temperature is polar a parabolic band).
optical (PO) phonon scattering, for which we have provided the The rates of the elastic scattering mechanisms, el , are
description of the Si (g) and So terms in Eqs. (A9) and (A10). calculated from the electron group velocities, v, and density
The influence of inelastic scattering mechanisms on g, and of states, DS ; thus the mobility may be calculated directly
therefore the overall mobility, are captured through the terms from the electronic band structure. The original form of these
Si (g) and So in Eq. (2), while elastic scattering mechanisms equations from k p theory, and the modified equations that
affect the overall mobility by the term el . This term is the we propose, are listed in Table I. Note that in every equation,
sum of all elastic scattering rates inside the material; it can be k
m d(k)
, which, in semiempirical models, is the group velocity
evaluated according to Matthiessens rule: fitted to experiment by the band gap and effective mass
el (k) = ii (k) + pe (k) + de (k) + dis (k), (3) of the semiconductor [included in d (k); see Table I], has
been replaced by its ab initio counterpart, or v (k), which is
where the subscripts el, ii, pe, de, and dis stand for elastic, calculated directly from the band structure.
ionized impurity, piezoelectric, deformation potential, and As an example, the DFT-calculated density of states
dislocation scattering rates, respectively. Therefore, the effect (DOS) can be plugged into Eq. (A7) to obtain the inverse
of relevant elastic and inelastic scattering mechanisms are charge screening length, , in ionized impurity scattering.
taken into account by explicitly solving the BTE [Eq. (2)] Furthermore, the numerator and denominator of the integrand
to obtain g (k). in Eq. (5) both contain the density of states of a free electron
When calculating various properties, several terms in gas, ( k )2 . Since this can also be calculated ab initio for the
Eq. (2) will be set to zero. For a Seebeck coefficient, S, specific system of interest, DS can instead be substituted in
calculation, the applied electric driving force, ( eE
)( f
k
), is the equation for calculating the mobility and reformulated in
f terms of the energies, :
set to zero. Only the thermal driving force, v( z ), in Eq. (2)
is taken into consideration when calculating the perturbation 
1 v () DS () g () d
to the electron distribution [3]. Assuming a uniform electron =  , (6)
concentration over the space at which a small temperature 3E DS () f () d
difference exists, the Seebeck coefficient is [3] where, again, v (k) is the electron group velocity and g is the
  2   
kB F k f (1 f ) kBT dk J perturbation to the electron distribution, which is calculated
S=  2 T

. (4) iteratively using Eq. (2), and can be expressed both as a
e kB T k f (1 f ) dk z function of k or (k) (i.e., the band structure).
For a mobility calculation, the applied thermal driving force Once the mobilities of the electrons and holes are known,
in Eq. (2) is set to zero, so that only the contribution of the the electrical conductivity can be readily calculated:
electric driving force is included. The mobility is = nee + peh , (7)
  2  g 
1 v (k) k dk where n and p are the concentration of electrons and holes,
=   k 2
E
. (5)
3 f dk respectively, e is the absolute value of the charge of an

electron, and e and h are the mobility of electrons and
Note that, in Eq. (5), the free electron density of states, ( k )2 , holes respectively.
has been used, which would limit its applicability in compound It should be noted that there are fundamental differences
semiconductors. Thus the replacement of this term by its between the model that we have presented here and those
ab initiocalculated counterpart would greatly improve the relying on the relaxation time approximation (RTA), and
accuracy of the resulting mobility. Furthermore, the scalar particularly, BTE-cRTA. Rather than simplification of the
group velocity, v (k), is used since the energy is averaged as collision term in the BTE [Eq. (A2)] through the RTA
a function of distance from the  point. In general, we use [Eq. (1)], we fully involve this term by considering both
the band structure, density of state, electron group velocity, elastic and inelastic scattering mechanisms. It is noteworthy
conduction band wave function admixture, and PO phonon that the BTE-cRTA formulation only implicitly takes into
frequency in calculating the mobility and Seebeck coefficient. account elastic and inelastic scattering mechanisms, by fitting
Therefore, all of the required inputs to Eq. (5) are calculated the overall relaxation time to experimental data with no
ab initio, which greatly enhances the predictability of the explicit consideration of changes in electron distribution from

235123-3
FAGHANINIA, AGER III, AND LO PHYSICAL REVIEW B 91, 235123 (2015)

TABLE I. Original equations [3,8], based on k p theory for elastic scattering rates and overall drift mobility, and proposed modifications,
based on ab initio parameters, introduced in this work.

k p theory with empirical parameters Ab initio

e4 Nm d(k) 4k 2 e4 N 4k 2
a
ii (k) = 8 02 3 k 3
[D(k) ln(1 + 2
) B(k)] ii (k) = 8 02 2 k 2 v(k)
[D(k) ln(1 + 2
) B(k)]
e2
 e2

2 = 0 kB T
( k )2 f (1 f )dk 2 = 0 kB T
DS ()f (1 f )d
2 2 e2 kB T P 2
pe (k) = e kB6T P3 mkd(k) [3 6c2 (k) + 4c4 (k)] pe (k) = 6 2 0 v(k)
[3 6c2 (k) + 4c4 (k)]
0
2 2
c (k) = 1 1+
b 2
2
, 2 (k) = 1 + 2mkg ( mm 1) c(k): obtained directly from wave functions
e2 k T E 2 mk d(k) e2 kB T ED2 k2
de (k) = B3 D3 c [3 8c2 (k) + 6c4 (k)] de (k) = 3 2 cel v(k)
[3 8c2 (k) + 6c4 (k)]
el

= 1 + m/m 1
4 e4 k
dis (k) = Ndis3e 2mc2d(k) 4k21 3/2 4 , d(k)
1
dis (k) = 2N dis 2 2
1
4k 2 )3/2 4
0 l (1+ ) 0 cl v(k) (1+
2 2
 3 
 k (g(k)/E d(k))dk v()DS ()g()d
overall = 3m

k 2 f dk
overall = 1
3E

DS ()f ()d
g(k) = f (k) f0 (k) g() = f () f0 ()
a
The subscripts stand for ii (ionized impurity), pe (piezoelectric acoustic phonon), de (deformation), and dis (charged dislocation scattering).
The parameters are m (electron mass), m (effective mass), 0 (low-frequency dielectric constant), g (band gap), v (k) (electron group velocity),
DS () (ab initio calculated density of states), c (k) (contribution of p-type orbitals to the conduction band), (inverse ionized impurity charge
screening length), ED (deformation potential), cl (lattice constant), E (small electric field), and cel (spherically averaged elastic constant). B (k)
and D (k) are just a collection of the parameters: c, k, and . Their purpose is to simplify the equation [3].
b
The c(k) parameter is the contribution of the p orbital to the wave function of the band. In the k p formulation, it has a closed-form expression
that includes the band gap and experimental effective mass. In the ab initio formulation, this wave function admixture can be calculated
by projecting the wave functions onto spherical harmonics that are nonzero within the sphere around each ion; this procedure is already
implemented in the Vienna ab initio Simulation Package (VASP) [2124].

each type of scattering mechanism. Furthermore, unlike the where the density of dislocations, Ndis , is only relevant for
semiempirical models that were described above, we use InN and is considered to be zero for GaAs. In both GaAs and
ab initio parameters; thus higher predictability and little to InN, temperatures lower than 20 K need not be considered due
no dependence on experimental data is achieved. to the deionization of shallow donors at lower temperatures,
as observed experimentally [26]. In the case of InN, electronic
B. Ab initio parameters scattering from existing linear charged dislocations thus
The main input that is needed for the transport model is the becomes important. The density of the dislocations, Ndis , can
crystal structure of the semiconductor material, from which be determined from TEM images, in the units of (cm2 ). We
ab initio parameters, such as the (optimized) lattice constant, can thus obtain the overall density in bulk, by assuming that
PO phonon frequency, dielectric and piezoelectric constants, these linear dislocations are uniformly developed along the c
deformation potential, and effective mass, can be computed. axis. This is reflected in dividing the dislocation density by the
We also need to know the Fermi level to compute scattering lattice constant, cl , in Eq. (10). By doing that, we are assuming
rates in Table I. In order to obtain the Fermi level, we that there is one unit of positive charge (donor) per unit cell. For
calculate the carrier concentration and match it to the given InN samples, according to Miller et al. [8], one can assume full
concentration (input), n, according to Eq. (8): ionization of the donors, and, therefore, a compensation level
of one (i.e., ND +N A
= 1 or ND  NA ). Also, the assumption
1 + n
of donor or acceptor charged dislocations yields similar results
n= g () f () d. (8)
V c [8]; therefore, we assume donor dislocations dominate here. It
Since both of the III-V semiconductors considered here are should be noted that we compare the calculated Ndis with the
n-type, the concentration of hole carriers is negligible. The corresponding experimental data if available; otherwise, the
concentration of ionized impurities, Nii (see Table I), is the limit for electronic properties at different values of Ndis can
sum of the concentration of all ionized centers regardless of be calculated without the need for experimental data.
the sign of their charge, since they are scatterer centers in both On the other hand, in a pure, epitaxially grown, high-
cases [25]: mobility GaAs sample with an electron concentration of n =
3.0 1013 , no dislocations exist (i.e., Ndis = 0). The concen-
Ndis
Nii = NA + ND + , (9) trations of donors and acceptors have been separately reported
cl [3,26], so this provides validation of the accuracy of aMoBT,
where ND and NA are concentration of donors and acceptors, without needing to solve for Nii . However, in the general case
respectively. Nii can then be calculated at a given electron where the electron concentration is unknown, we can plot the
concentration, n, by iteratively solving the charge balance mobility and Seebeck coefficient at different compensation
equation [8]: ratios to define the limit of the transport properties, as shown
Ndis in Fig. 5. Therefore, it is important to note that only when
n + NA = ND + , (10) comparing with experimental mobilities/Seebeck coefficients
cl

235123-4
AB INITIO ELECTRONIC TRANSPORT MODEL . . . PHYSICAL REVIEW B 91, 235123 (2015)

do we use experimentally measured electron concentrations; TABLE II. Structure of GaAs and InN calculated with DFT, using
otherwise, we may calculate ab initio mobility or Seebeck co- the GGA-PBE exchange-correlation functional. Changes in the lattice
efficient, for example, at various electron concentrations, with- constants compared to experimental values [42,43] upon optimization
out any reliance on experimental data (e.g., as shown in Fig. 5). are reported below.
We use Brooks-Herring theory for singly charged ionized
impurity scattering [25], as shown in Table I. This is supported Compound Space |a| (A) % change |c| (A) % change
by the fact that in GaAs, oxygen impurities, O+1 group in |a| in |c|
As , have been
confirmed to be dominant and singly charged [27], while in GaAs F -43m 5.75 2.17%
InN, nitrogen (donor) vacancies, V+1 N , are dominant and singly InN P 63 mc 3.533 0.56% 5.693 0.8%
charged [28]. It should be noted that the Brooks-Herring
formulation is more accurate at low carrier concentrations,
since at high concentrations, despite the inherent assumption
III. COMPUTATIONAL METHODOLOGY
of the theory, not all electrons are screened by the charge of
an ionized center. More information on the Brooks-Herring For each semiconductor material, the geometry of the unit
ionized impurity model is available in Appendix A 2. cell is optimized, and the density of states and band structure
In order to calculate the low- and high-frequency dielectric are calculated. In the case of zinc blende GaAs and wurtzite
constants, we use density functional perturbation theory InN, the unit cells are optimized using Kohn-Sham density
(DFPT), as implemented in VASP, to determine Born effective functional theory (KS-DFT) [36,37], as implemented in VASP.
charges, dielectric and piezoelectric tensors, including local The generalized gradient approximation of Perdew, Burke,
field effects in DFT, as well as the force-constant matrices and and Ernzerhof (GGA-PBE) [38,39] is used to express the
internal strain tensors. We then subtract the ionic contribution exchange-correlation potential, and projector augmented wave
to the static dielectric tensor to obtain the high-frequency di- (PAW) potentials [40,41] are used to represent the valence
electric constant [29,30]. Furthermore, the inelastic scattering wave functions. Information regarding the structure of these
effect is strongly dependent on the longitudinal polar optical two systems and their changes upon geometry optimization
phonon frequencies, po . These frequencies can be calculated have been summarized in Table II. The initial structures are
using the Phonopy code [31], where we identify the highest obtained from the literature [42,43].
energy peak in the optical phonon density of states. It should We then compute the electronic band structure of these
be noted that at and around the  point, the phonon frequency materials. The energy cutoff for the plane wave basis set
is almost constant (Fig. 8). is set to 500 eV. The band structure is computed in line
To calculate ab initio the deformation potential, ED , we mode along seven high-symmetry k-points in the IBZ, with 20
strain the system and calculate the energy of the conduction k-points between each pair of high-symmetry points. The self-
band of InN and GaAs unit cells at different volumes. Then, consistent density of states (DOS) calculation is performed
we approximate the deformation potential using the following using a 20 20 20 k-point mesh, for both GaAs and InN.
equation: The non-self-consistent energy calculations are performed in

  a special k-point mesh around the  point, at which the
ECBM  conduction band minimum (CBM) occurs in both direct band
ED = V  , (11)
V T at V =V0 gap GaAs and InN. This k-point mesh contains a total of 10 234
points in the irreducible Brillouin zone (IBZ), with mesh
where V is the volume, ECBM is the energy of the CBM, and spacing of 0.001, 0.01, or 0.1 fractional units, to completely
V0 is the volume of the relaxed structure (i.e., zero pressure) account for band anisotropy while remaining dense enough
[32,33]. It should be noted that since the absolute value of around the  point to obtain accurate group velocity and
ECBM is a function of the volume itself, we use the difference effective mass values. For BTE-cRTA calculations, we use
between the energy of the first conduction band and the first a 21 21 21 k-point mesh, equivalent to 4831 k-points in
valence (core) band. Furthermore, the elastic and piezoelectric the IBZ, to minimize the band crossing error [10]. We did
constants have been already calculated ab initio for GaAs and not observe any changes in the calculated properties in larger
InN, and are available in the literature. For GaAs, we use the k-point meshes. To determine the effect of presumably more
values calculated by Beya-Wakata et al. [34], and for InN we accurate band structure calculations on the band curvature,
use the values calculated by Sarasamak et al. [35], to obtain effective mass, and group velocity, we have also employed the
the piezoelectric coefficient and elastic constant used in the GW method. Only 941 k-points in the IBZ have been used for
equations for piezoelectric scattering in Table I. GW calculations since it is more computationally demanding.
As a comparison, the electrical conductivity and Seebeck Using more k-points does not change the calculated effective
coefficient are also computed using the widely used BTE- mass. The GW 0 band structure calculations are performed
cRTA formulation. We choose the BoltzTraP package [10], using the maximally localized Wannier functions (MLWFs)
which uses Fourier interpolation of the calculated bands, and interpolation, as implemented in VASP and the Wannier code
differentiate the band energies to find the group velocities [44]. It is very important to also show the feasibility of
of the electrons. Other than the need to fit the relaxation the ab initio model with band structures calculated using
time to experimental measurements of the conductivity, the DFT + U , since the GW and hybrid functional (e.g., HSE
BoltzTraP/BTE-cRTA implementation represents an other- [4547]) methods are computationally demanding for complex
wise parameter-free model that can be adapted to different materials with larger unit cells than GaAs. On the other hand,
semiconductor materials. for InN, all methods and functionals attempted, including

235123-5
FAGHANINIA, AGER III, AND LO PHYSICAL REVIEW B 91, 235123 (2015)

Energy (eV)
Energy (eV)

(a) GaAs (b) InN

FIG. 1. Band structure of cubic GaAs and wurtzite InN, normalized so that the Fermi level is set to zero at the conduction band minimum.

LDA, GGA, HSE, and GW 0, resulted in a band structure discontinuity where the polynomials meet. This results in very
with zero band gap and a falsely predicted p-like conduction smooth calculated group velocities, and, subsequently, other
band. Only GGA + U [48], with U values obtained from transport properties.
the literature [49,50] (Ud = 6 for In and Up = 1.5 for N),
produced a correct band structure and with a more s-like
conduction band particularly around the  point, which is IV. RESULTS AND DISCUSSIONS
consistent with the self-interaction corrected band structure
reported by Furthmuller et al. [51]. In the process of choosing A. Ab initio calculated parameter inputs to the transport model
the U value for GGA + U band structure calculations on The computed band structures of GaAs and InN are shown
GaAs, however, the values (U = 8 eV for both d orbitals of in Fig. 1. We have calculated a GW 0 band structure, which
Ga and As) recommended by Persson and Mirbt [52], with an starts from the wave functions previously computed using the
emphasis on correctly obtaining the band gap and effective GGA-PBE functional, as shown in Fig. 1(a).
mass values, result in GaAs falsely becoming an indirect The band structures used in previous semiempirical models
semiconductor, with the conduction band minima located at [3,8] express the energy of the conduction band as a function of
the L and X k-points rather than the  point [52]. Therefore, the distance from the  point. Instead, we calculate the ab initio
we have also employed effective U values of 7 eV (Ga) and band structure in a three-dimensional grid around the CBM,
6 eV (As), for which a direct band structure is obtained. We and then average the energy values of the k-points that share
have calculated mobilities obtained from both of these band the same distance from the  point (Fig. 2). For both GaAs and
structures and compared them with the ones obtained by the InN, the ab initio and k p band structures agree well at small
GW band structure. In order to calculate the group velocities, k-points; however, they diverge at larger k-points. This directly
v (k), and the overall average effective mass, we have fitted impacts the group velocity of the electrons and, ultimately,
a sixth degree polynomial to the calculated conduction band the transport propertiesparticularly at higher temperatures
[i.e., average energy as a function of distance from  point or where higher energy electrons have nonzero occupation.
(k)] with R 2 > 0.99: We have also calculated a GGA + U [48] band structure,
1 with U values taken from the published literature [49,50], as
v(k) = , (12) shown in Fig. 1. For InN, GGA + U correctly yields an s-like
 k
conduction band and a band gap of 0.5 eV, which is comparable

1  to the self-interaction corrected band gap of 0.58 eV reported
1 2 
m =  . (13) by Furthmuller et al. [51] and the experimental values of
2 k 2  0.6750.7 eV [55,57,61] (Table III). We include DFT + U
at k=0
calculations only to show the feasibility of these less-expensive
It should be noted that we do not use the value of effective methods, in the case of more complex semiconductor materials
mass in the proposed carrier transport model. Rather, we for which a GW calculations is too expensive. Also, DFT
calculate it solely to compare with experiment and evaluate usually suffers from vastly underestimating the effective
the effect of the shape of the conduction band (i.e., group mass[1,2], and the introduction of the fitting parameter U may
velocities) calculated by various methods, such as GGA, reduce the predictability of the ab initio model as a whole.
GGA + U , and GW . Fitting polynomials to the numerically Therefore, we stress that all reported transport properties are
calculated conduction band and density of states results in calculated here using the parameter-free GW band structures,
smooth plots of mobility and Seebeck coefficient, as presented unless otherwise stated.
here, while preserving the values that are calculated ab initio Although we do not directly use the value of the electron
with R 2 > 0.99 in all segments fitted. We fit these polynomials effective mass in the transport property expressions, we see that
at different segments of the band structure and carefully choose the calculated effective mass of 0.062 for InN is consistent
only the ones that result in the maximum R 2 and minimum with the previously calculated effective mass (0.066) using

235123-6
AB INITIO ELECTRONIC TRANSPORT MODEL . . . PHYSICAL REVIEW B 91, 235123 (2015)

0.25 0.25
kp kp
GW
GW
0.20 DFT+U(7,6)
0.20 DFT+U

Energy (eV) DFT+U(8,8)

Energy (eV)
0.15 0.15

0.10 0.10

0.05 0.05

0.00 0.00
0.0 0.2 0.4 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6
-1 -1
k (nm ) k (nm )

(a) GaAs (b) InN

FIG. 2. (Color online) Conduction bands expressed in terms of the average energy as a function of distance from the CBM (i.e., center
of Brillouin zone, or  point), as calculated from semiempirical expressions (in k p formulation) versus ab initio. The difference at higher
k values has a significant impact on transport properties, especially at high temperatures. The values of U for the d orbitals of gallium and
arsenic, respectively, are in parentheses, while those for InN are taken from the published literature [49,50].

an empirical pseudopotential [62], and well within the range characterized GaAs samples, as described by Stillman et al.
(0.050.08) measured experimentally [5560]. [26]. We also perform this analysis over a wide temperature
We also show the calculated phonon band structure and range for high purity GaAs samples with very low electron
density of states of these two compounds in Fig. 8. For GaAs, concentrations, as labeled as pure in Table IV.
the calculated PO-phonon frequency of 8.16 THz is shown in As shown in Fig. 3(a), the most accurate GW band structure
Fig. 8(a). For InN, the calculated optical phonon frequency of results in the best agreement with experimental data. The
17.83 THz is close to the 17.65 THz value reported by Bungaro DFT + U band structure, however, does provide us with limits
et al. [55,63]. We have listed all the parameters that are used in of the mobility over different temperatures. When calculating
our transport model in Table III. We have calculated all of these the mobility and Seebeck coefficient, we calculate the Fermi
parameters, as bolded in Table III, ab initio to demonstrate the level by first calculating the electron concentration through
feasibility of a fully predictive model for transport properties. Eq. (8), and then matching it to a given concentration. The
The only exceptions are the elastic and piezoelectric constants, calculated properties are very sensitive to the calculated Fermi
which are necessary to calculate the piezoelectric coefficient, level. Therefore, for comparison, we have included the results
P , in Table I. As described earlier, we have instead used the using both the ab initio DOS used in Eq. (8), and the free
previously calculated values from published DFT studies for electron DOS. As shown in Fig. 3(a), the ab initio model
these constants [34,35]. for DOS performs better for lower electron concentrations
and lower temperatures, while the free electron DOS is more
suitable for higher temperatures, and, particularly, at higher
B. Model validation on GaAs electron concentrations. We acknowledge that because of the
In order to evaluate the accuracy of aMoBT, we first log scale in Fig. 3(a), seeing the quantitative agreement is
calculate the mobility of three experimentally synthesized and difficult. Therefore, we report the calculated relative error

TABLE III. Inputs to the transport model, as calculated ab initio compared to experimentally measured values. The bolded numbers are
used in our transport property calculations; note that not all appear in the final expressions listed in Table I.

GaAs InN
a
Parameter Ab initio Expt. Ab initio Expt.
cl (nm) 0.562 0.575 [3] 0.565 0.569 [3]
po (THz) 8.16 8.73 [3] 17.83 17.65 [3]
0 12.18 12.91 [3] 11.42 10.3 [8]
10.32 10.91 [3] 6.24 6.7 [8]
ED (eV) 6.04 8.6 [3] 4.46 3.6 [3]
m 0.0530.066b 0.06360.082 [53,54] 0.062, 0.071 (GW ) 0.050.08 [5560]
g (eV) 0.96, 1.19 (GW ) 1.424 [6] 0.50 0.6750.7 [55,57,61]
a
The parameters are cl (lattice constant), po (PO phonon frequency), 0 (low-frequency dielectric constant), (high-frequency dielectric
constant), ED (deformation potential), m (effective mass), and g (band gap).
b
The GaAs effective masses are calculated as 0.053 (GGA + U , this work), 0.066 (GGA + U , with published U [52]), and 0.063 (GW 0).

235123-7
FAGHANINIA, AGER III, AND LO PHYSICAL REVIEW B 91, 235123 (2015)

TABLE IV. Carrier concentrations of various experimentally 1000

fabricated and characterized GaAs samples. For the pure sample, exp

Electrical Conductivity (S/m)

this work, GW
data is available roughly between 5 and 1000 K. For the real samples, BTE-cRTA matched Fermi
mobility data is also tabulated at different temperatures. 100 BTE-cRTA Fermi at 200K
BTE-cRTA Fermi at 300K
BTE-cRTA Fermi at 400K
Sample Concentration, n (cm3 ) Donor, ND Acceptor, NA Ref.
10
Pure 3 1013 5.2 1013 2.2 1013 [3]
a 2.7 1013 4.8 1013 2.1 1013 [26]
c 7.7 1014 1.1 1015 3.3 1014 [26] 1
e 3.1 1015 4.7 1015 1.6 1015 [26]

0.1
compared to the experiment for the best cases for each 200 400 600 800 1000
samplefrom the ab initio DOS for sample a and from the free Temperature (K)
electron DOS for samples c and e. The minimum, maximum,
and the relative error in calculating the mobility of sample a are FIG. 4. (Color online) Electrical conductivity of GaAs calculated
2.25% (at 195 K), 29.42% (at 29 K), and 13.33%, respectively. using aMoBT (solid line) and the BTE-cRTA framework, and
These numbers are 1.02% (at 167 K), 15.01% (at 49 K), and compared to experimental [3] data. The Fermi level is calculated by
7.97% for sample c and 0.22% (at 195 K), 7.90% (at 40 K), matching the calculated carrier concentration to n = 3 1013 . This
has been done either at the mentioned temperature and kept constant
and 4.04% for sample e. Overall, the agreement is poorer at
over the whole temperature range, or in the case of matched Fermi,
higher electron concentrations and lower temperatures; this
at each temperature, the Fermi level is adjusted to the given n. The
is attributed to the inaccuracy of the Brooks-Herring ionized
relaxation time, , is determined by fitting the calculated conductivity
impurity scattering model at high electron concentrations, as to the corresponding experimental value at 300 K. The calculated
briefly described in Sec. II. Furthermore, the model has also value for is 5 1013 s.
been validated with the data on crystalline samples with very
high purity. The calculated electron mobilities, assuming the
limit that only one scattering mechanism exists at a time,
along with the overall mobility, are shown in Fig. 3(b). The
reasonable agreement between the calculated and experimental
mobilities provides independent validation of the transport the correct changes with the temperature and the carrier
model. The minimum, maximum, and average relative error concentration.
of calculated mobility are 0.46% (at 394 K), 23.55% (at Once we have the calculated mobility, at a given electron
175 K), and 9.53%, respectively, for temperatures above 20 K. concentration, we can calculate the electrical conductivity
The mobility is mainly limited by ionized impurity scattering of GaAs by Eq. (7). For now, we assume that the carrier
at low temperatures, piezoelectric scattering at intermediate concentration remains constant with temperature over the
temperatures, and polar optical phonon scattering at higher range of interest. We then compare to the experimental
temperatures (>60 K); all of these are consistent with the conductivity and those values calculated using the BTE-cRTA
previous results shown by semiempirical models [3,4,6] yet framework, under the scenarios listed in Fig. 4. As shown,
no experimental parameter has been used here in predicting not only does BTE-cRTA fail to correctly predict the trend for

4
a DFT+U (7,6) overall
de
2 GW free-e DOS 6
10 ii
GW ab initio DOS pe
c
Mobility (cm2/ Vs)

po
Mobility (cm / Vs)

5
10 exp
8
2

6 5
10
4

2
e 10
4

4
10
8
6
3
10
50 100 150 200 250 300 100 200 300 400 500 600 700 800 900 1000
Temperature (K) Temperature (K)

(a) (b)

FIG. 3. (Color online) Calculated and experimental [3,64] mobility data for GaAs at various electron concentrations and temperatures.
More details on the experimental data, including donor and acceptor concentrations, are available in Table IV. (a) Comparison of model trends
with varying temperature and electron concentration to experimental data. The values of U used for the d-orbitals of Gallium and Arsenic,
respectively, are in parentheses. (b) Limitation of pure GaAs mobility from each scattering mechanism.

235123-8
AB INITIO ELECTRONIC TRANSPORT MODEL . . . PHYSICAL REVIEW B 91, 235123 (2015)

the Seebeck coefficient, assuming Pisarenko behavior, and

Seebeck Coefficient (V/K)
-1200 exp
compared it to aMoBT in Fig. 5. We use Eq. (14) with two
Nii = 1 x n
fitting parameters: effective mass, m and r. It should be noted
-1000 Nii = 10 x n
Pisarenko
that in the case where the best agreement with experiment,
-800 BTE-cRTA through Pisarenko behavior, is only achievable by choosing
either m = 0.11 or r = 0.35, both of these values are far from
-600 experimental measurements and thus lack physical meaning.
Furthermore, in Fig. 5, while BTE-cRTA correctly predicts
-400
the trend in Seebeck coefficient with carrier concentration,
-200 without the need to calculate the relaxation time constant,
the predicted values are far from the experimental results and
13 15 17 19
those calculated by aMoBT. This is attributed to the treatment
10 10 10 10 of () as a single constant, , which affects both conductivity
-3
Carrier Concentration (cm ) and Seebeck coefficient when integrated over energy,
 
FIG. 5. (Color online) Calculated (by aMoBT) and experimental kB 5 2 (2 m kB T )3/2
[4] GaAs Seebeck coefficient, at different ratios of the ionized S + r + ln . (14)
e 2 h3 n
impurity concentration, Nii , to the electron concentration, n. For the
Pisarenko plot, using Eq. (14), we have used values of r = 21 for
acoustic phonons, and m = 0.063. The Fermi levels for BTE-cRTA C. Model validation on InN
calculations were obtained by integrating the density of state to obtain
the corresponding carrier concentrations; thus the same band structure In order to further evaluate the accuracy of aMoBT and
and Fermi levels have been used to obtain the black and green lines. its applicability to more complicated semiconductors, we also
calculate the mobility and Seebeck coefficient (Fig. 6) of three
experimentally synthesized and characterized InN samples by
conductivity with temperature, but also quantitatively differs Miller et al. [8]. These calculations are more challenging due
from the experimental values. to the reported presence of linear charged dislocations in the
Finally, we calculate the Seebeck coefficients of the GaAs crystal structure [8,6467], due to the processing conditions
samples (assumed to be at 300 K), and compare them to the employed. For each sample at a given carrier concentration, as
values reported previously by Rode and Knight [4] (Fig. 5). shown in Table V, we change the concentration of dislocations,
Since the data are for various samples with different electron Ndis , until the calculated mobility values match the experi-
concentration and compensation ratios, we choose various mental measurements. The fitted Ndis (Table V) is within the
values of Nii /n = (ND + NA )/n. As shown, a range of See- range of measured concentrations from transmission electron
beck coefficients are calculated at each electron concentration, microscopy analysis (TEM) [8], which confirms that the lim-
which includes the experimentally measured points. It should iting mechanism is indeed scattering from dislocation lines.
be noted that not knowing beforehand the compensation and As shown in Figs. 6(a) and 6(b), while there is an
concentration of donors and acceptors, as well as their charge excellent agreement between the calculated and experimental
states, limits the overall predictability of aMoBT. However, mobility, the calculated Seebeck coefficients for samples B
even given these limitations, the close fit between ab initio and and C exhibit more pronounced changes with temperature
experimental properties provides independent validation of the than the experimental Seebeck coefficients. The mobility of
viability of aMoBT. For further evaluation, we have calculated the samples is found to be limited by charged dislocations,
Seebeck Coefficient (V/K)

1400
A
1200 -50
Mobility (cm /Vs)

-100
2

1000 semi-empirical
this work B
800 -150

600 C
-200
400 -250
50 100 150 200 250 300 150 200 250 300
Temperature (K) Temperature (K)
(a) Mobility (b) Seebeck coecient

FIG. 6. (Color online) Calculated (by aMoBT) and experimental transport properties of the three InN samples listed in Table V. The dashed
line is calculated by the semiempirical model used by Miller et al. [8], while the solid line is calculated by the proposed ab initio transport
model (aMoBT).

235123-9
FAGHANINIA, AGER III, AND LO PHYSICAL REVIEW B 91, 235123 (2015)

TABLE V. Measured [8] and calculated InN dislocation density, equilibrium Fermi-Dirac distribution; thus the applicability of
corresponding to the mobility and Seebeck coefficient reported in the current model for high-field transport or heavily doped and
Figs. 6(a) and 6(b). polar semiconductors where the linear BTE formulation fails
[68] is very limited. Furthermore, we have averaged the energy
Ndis (cm2 ) around the CBM and expressed the energy values in the band
Sample Expt. Semiempirical [8] This work structure as a function of the absolute value of k, or simply
the distance from  point in the reciprocal space. Therefore,
A 1 1011 1.5 1011 8.20 1010 the reported mobility values are averaged and the effect of
B (25) 1010 1.5 1010 1.18 1010 band structure anisotropy is not fully captured. It is possible,
C 1 109 5 1010 4.1 109 3.47 109 however, to include the band structure of the material only in
the specific orientation of interest to account for anisotropy.
Currently, the model is limited to a single conduction band.
particularly at low temperatures. The next limiting mechanism Although the single band ab initio model can be used for
is polar optical phonon scattering, which is more important prediction of many direct band gap semiconductors, it will
at higher temperatures while ionized impurity scattering is only result in an overestimation of transport properties of
more important at lower temperatures. This can be seen in semiconductors with more complex band structure. This is
Fig. 7, which shows the mobility of sample B if it were due to the fact that currently, interband scatterings between
limited by each type of scattering mechanism, as well as the several bands that are participating in transport are neglected.
overall mobility. These findings are in agreement with the Future work could include solution of the coupled-BTE and
semiempirical transport model [8], except that all parameters taking into account two and more participating bands, which
are calculated ab initio, which requires knowledge only of enables calculation of both electron and hole mobilities in
the crystal structure of the material. Comparing the transport more materials. Finally, although the usage of the Hubbard U
properties calculated using aMoBT with those calculated using parameter in the band structure calculation might limit the
semiempirical models [including experimentally measured predictability of the model in calculating overall transport
band gap and effective mass (see Table III under Expt.)] properties, this can be properly addressed by using more
in Fig. 6 shows that although quantitative agreement with accurate methods of band structure calculations as reported
experiment is slightly better with the semiempirical model, here. We include DFT + U calculations here only to show
Seebeck coefficient calculations on samples B and C, and the the feasibility of working with the model when GW or other
mobility of the sample at high temperature, show much better less commonly used methods are not technically or otherwise
accuracy with the ab initio model presented here. feasible.
Finally, we should once again acknowledge the assumptions
and limitations of the current model when applied to the other
V. CONCLUSIONS
types of semiconductors. Most importantly, the formulation
presented in this work is for low-field transport (particularly We have presented an ab initio transport model for
drift mobility and Seebeck coefficient), in which the changes to calculating the electrical mobility and Seebeck coefficient of n-
the electron distribution are merely a linear perturbation to the type semiconductors using the Boltzmann transport (aMoBT)
equation. By using the inputs from density functional theory
calculations, and considering all relevant physical phenomena
(i.e., elastic and inelastic scattering mechanisms), we have
6 overall
10 de successfully calculated highly accurate transport properties
ii of GaAs and InN over various ranges of temperature and
pe carrier concentration. aMoBT provides both qualitative and
po quantitative improvements in accuracy compared to the widely
Mobility (cm /Vs)

5
10 dis used semiempirical and constant relaxation time approxima-
2

exp tion model solutions to the Boltzmann transport equation.

Future work will focus on extending this model to p-type
4
semiconductors.
10

ACKNOWLEDGMENTS

3 This research is based upon work supported by the Solar

10 Energy Research Institute for India and the U.S. (SERIIUS)
funded jointly by the U.S. Department of Energy subcontract
50 100 150 200 250 300 DE AC36-08G028308 (Office of Science, Office of Basic
Temperature (K) Energy Sciences, and Energy Efficiency and Renewable
Energy, Solar Energy Technology Program, with support
FIG. 7. (Color online) Calculated and experimental [8] values for from the Office of International Affairs) and the Government
InN mobility at n = 9 1017 cm3 (sample B in Table V). Each of India subcontract IUSSTF/JCERDC-SERIIUS/2012. This
line represents the mobility if limited only by the corresponding work used the Extreme Science and Engineering Discovery
mechanism. Environment (XSEDE), which is supported by National

235123-10
AB INITIO ELECTRONIC TRANSPORT MODEL . . . PHYSICAL REVIEW B 91, 235123 (2015)

Science Foundation Grant No. OCI-1053575. We would like 2. Explicit solution of linear BTE
to thank Daniel Rode for helpful discussions during the To go beyond the relaxation time approximation, both
preparation of this manuscript. elastic scattering mechanisms, for which the kinetic energy
of the electrons remains constant, and inelastic scattering
APPENDIX A: BOLTZMANN TRANSPORT EQUATION mechanisms, for which there is a change in the electron
distribution, should be taken into account. If the system is
The Boltzmann transport equation (BTE) describes the governed only by elastic scattering mechanisms, the relaxation
nonequilibrium behavior of charge carriers (e.g., electrons time, , is equal to the inverse of the overall elastic scattering
or holes) by statistically averaging over all possible quantum rates, which is the sum of all individual rates. Evidently,
states. For the electron distribution, f , this is represented by is not constant but does depend on energy; however, it
the BTE: does not necessarily follow a power law dependence (e.g.,

 in InN [8]). However, inelastic scattering mechanisms also
df (k,T ,t) f (k,T ,t) dk
= k f (k,T ,t) limit the mobility, and, therefore, the conductivity, of the
dt t s dt
semiconductor; as an example, polar optical (PO) phonon
v (k) r f (k,T ,t) , (A1) scattering is the main electron-phonon interaction that limits
mobility at high temperatures in GaAs. Thus we need to first
where f is a function of state k, temperature T , and time t, calculate the perturbation, g, to the electron distribution due to
and v (k) are the electron group velocities. The three terms elastic and inelastic scattering mechanisms, and then integrate
on the right-hand side of Eq. (A1) refer, respectively, to the g over all states to obtain the mobility. Details on this approach
temporal rate of change of f due to all scattering processes, are given below.
rate of change of f due to external forces, and diffusion from The most relevant elastic scattering mechanism for com-
the carrier density gradient. pound semiconductors is expected to be ionized impurity
If the external forces consist only of a low electric field, E, scattering at low temperatures. Ionized impurity scattering
and no magnetic field, B, such that dk
dt
= eE
, then the low-field occurs when a charged center is introduced inside the bulk
BTE becomes material. As a result of Coulombic interactions between the
df (k,T ,t) eE electron and ion, electrons scatter to different states (i.e.,
+ v (k) r f (k,T ) + k f (k,T ) become distracted). The ionized impurity scattering rate, ii
dt 

 (i.e., a component of the overall ), may be expressed using
f (k,T ,t)
= . (A2) Brooks-Herring theory [25]:
t s
 
e4 N 4k 2
ii = D ln 1 + 2 B , (A5)
1. Constant relaxation time approximation 8 02 2 k 2 v
Furthermore, f can be described as a first-order (linear) where the charge screening potential, , is obtained by solving
perturbation, g (k), from the (equilibrium) Fermi-Dirac distri- Poissons equation:
bution, f0 , due to scattering:

 q
f (k,T ,t) f (k) f0 (k) g (k) = exp (r) (A6)
= = , (A3) 4 0 r
t s
and inverse screening length, , is given by
1
f0 [ (k)] = , (A4) e2
e[(k)F ]/kB T + 1 2 = DS () f (1 f ) d, (A7)
0 k B T
where the dependence of on k is given by the electronic band
structure, and the various scattering terms and time dependence where f is the electron distribution and 0 is the low-frequency
are lumped into the electronic relaxation time, . dielectric constant. Details on the , D, and B parameters are
If is a constant, then this major simplification results given in the literature [3].
in the BTE-cRTA. This assumption simplifies the theory to At high temperatures, after an inelastic (e.g., electron
an extent that closed form expressions for conductivity and phonon) scattering event, where the electron scatters from
Seebeck coefficient can be obtained [10]. In this approach, momentum state k to k  , the energy of an electron changes,
the details of all elastic and inelastic scattering mechanisms and, hence, the electron distribution also changes. (Note that
are lumped into the relaxation time constant, . While popular, the distribution may also be perturbed by external forces, such
this approach suffers from the following disadvantages: (1) is as an electric field or temperature gradient.) Thus f becomes
obtained by fitting to the experimental data for the conductivity a function of k, so it must be mapped via the electronic
of the material, which limits the predictability of the model, band structure, (k). This effect can be shown as a deviation
and (2) due to oversimplification of the transport mechanism, from Fermi-Dirac behavior [Eq. (1)]. After some mathematical
it may result in incorrect values and even incorrect trends manipulation, for which details can be found in the literature
with temperature or carrier concentration, as illustrated in Fig. [3], the BTE can be reformulated as
4. Therefore, by explicitly including all possible electronic  
scattering mechanisms, one can determine which mechanisms Si (g  ) f
z
eF f
 k
g= , (A8)
are physically relevant for a given semiconductor. So + el

235123-11
FAGHANINIA, AGER III, AND LO PHYSICAL REVIEW B 91, 235123 (2015)

FIG. 8. Phonon band structures of InN and GaAs calculated by Phonopy [31].

APPENDIX B: SENSITIVITY ANALYSIS

   
Si (g ) = dk Xg(k )[sinel (k ,k)[1 f (k)] 1. Sensitivity to the calculated dielectric constants
We have done sensitivity analysis for the calculated mobil-
+ sinel (k,k  )f (k)], (A9) ity of GaAs pure sample at different dielectric constants. As
shown in Fig. 9, the result is sensitive to dielectric constants
at low and high temperatures but much less sensitive at
So = dk  [sinel (k,k  )[1 f (k  )] + sinel (k  ,k)f (k  )].
temperatures in 100200 K range. Inaccurate calculation of
(A10) dielectric constant in this case can result in up to 41% (at
40 K) calculated value of mobility compared with measured
values in case of 20% from the base value for dielectric
Detailed integrated expressions for the scattering in, Si , and constant and it can go up to +43% (at 5 K) for +20% from
scattering out, So , terms are available in the literature [3]. the base value. The base values are the ones reported in Table
The reformulated BTE can then be solved   iteratively, using III calculated from ab intitio assuming the relaxed structure.
Rodes method [3,8,1114], since Si g  and f themselves This shows the importance of accurate calculation of these
are functions of g. First, the Fermi-Dirac distribution can constants at least with 5%10% accuracy.
be plugged into the right-hand side of Eq. (A8) to obtain
the first guess, g1 , which in turn is used to obtain a new
electron distribution to solve for the next guess, g2 ; this process
continues until g converges to a unique value. Typically, five -20% dielectric
2
iterations are required for the perturbation to converge for polar -10% dielectric
base case
optical phonon scattering in GaAs and InN. More details on 10
5
+10% dielectric
8 +20% dielectric
Eqs. (A8)(A10) are available in the literature [3].
Mobility (cm / Vs)

6 exp
4
2

2
3. Phonon dispersion
4
Polar optical phonon scattering originates from interactions 10 8
between electrons and high-frequency optical phonons. They 6

provide the dominant inelastic electron scattering mechanism 4

near (and above) room temperature in compound semiconduc- 2

tors. This is attributed to the high energies of optical phonons
being comparable to kB T at high temperatures. The scattering 100 200 300 400 500 600 700 800 900 1000
rates themselves are strongly dependent on the polar optical Temperature (K)
phonon frequencies. po . These frequencies can be calculated
using the Phonopy code [31] which solves for dynamical FIG. 9. (Color online) Sensitivity analysis of the mobility of
matrix from the force constants calculated using density GaAs pure sample (see Table IV). We changed here only the static,
functional perturbation theory (DFPT), as implemented in s , and high frequency, , dielectric constant from 20% to +20%
VASP. The phonon band structures for GaAs and InN are shown of the base case values reported in Table III. The results are sensitive
in Fig. 8. to dielectric constant at low and high temperatures.

235123-12
AB INITIO ELECTRONIC TRANSPORT MODEL . . . PHYSICAL REVIEW B 91, 235123 (2015)

2. Sensitivity to the lattice constants

6
We also applied 3% strain to the lattice constant of 10 -3% strain
base case
the relaxed GaAs and recalculated the band structure, DOS, +3% strain
exp
and optical phonon frequencies to see how sensitive is the

Mobility (cm / Vs)

calculated mobility with respect to the crystal structure. We as- 5
10
sume that everything else is kept constant according to the base

2
case (see Table III). According to Fig. 10, the calculated mobil-
ity is extremely sensitive to the crystal structure. This is mainly
due to the impact that the structure has on the band shape 10
4

(i.e., group velocity of the electrons) since the mobility at any

temperature is affected. For example, the GW band structure of
3% strained GaAs gives the effective mass of 0.026 while that
3
of +3% strained GaAs gives the effect of 0.10. Both of these 10
values are well outside of the range of the reported experiem-
100 200 300 400 500 600 700 800 900 1000
ntal values 0.0640.082 (see Table III). Also, these strained
structures are extremely unlikely to be relaxed with any func- Temperature (K)
tional since their built-in pressure with GGA-PBE functionals FIG. 10. (Color online) Sensitivity analysis of the mobility of
are already 10.66 kbar and 74.77 kbar, while the relaxed GaAs pure sample (see Table IV) calculated by aMoBT. We changed
structure that we have calculated and reported in Table II has here the crystal structure and subsequently the newly calculated
a built-in pressure of only 0.3 kbar. Nevertheless, Fig. 10 optical phonon frequencies. The calculated mobility is sensitive to
shows the importance of accurate calculation of the crystal the strain at all temperatures.
structure and subsequently the band structure (i.e., group
velocities).

[1] G. Hautier, A. Miglio, D. Waroquiers, G.-M. Rignanese, and X. [18] J. D. Wiley, Semiconductors and Semimetals (Ref. [3]),
Gonze, Chem. Mater. 26, 5447 (2014). Chap. 2.
[2] G. Hautier, A. Miglio, G. Ceder, G.-M. Rignanese, and X. [19] T. Dannecker, Y. Jin, H. Cheng, C. F. Gorman, J. Buckeridge,
Gonze, Nat. Commun. 4, 3292 (2013). C. Uher, S. Fahy, C. Kurdak, and R. S. Goldman, Phys. Rev. B
[3] D. L. Rode, Semiconductors and Semimetals (Academic Press, 82, 125203 (2010).
New York, 1975), Chap. 1. [20] C. Bera, M. Soulier, C. Navone, G. Roux, J. Simon, S. Volz, and
[4] D. L. Rode and S. Knight, Phys. Rev. B 3, 2534 (1971). N. Mingo, J. Appl. Phys. 108, 124306 (2010).
[5] O. F. Sankey, J. D. Dow, and K. Hess, Appl. Phys. Lett. 41, 664 [21] G. Kresse and J. Hafner, Phys. Rev. B 47, 558 (1993).
(1982). [22] G. Kresse and J. Hafner, Phys. Rev. B 49, 14251 (1994).
[6] H. Ehrenreich, Phys. Rev. 120, 1951 (1960). [23] G. Kresse and J. Furthmuller, Comput. Mater. Sci. 6, 15 (1996).
[7] J.-H. Bahk, Z. Bian, and A. Shakouri, Phys. Rev. B 89, 075204 [24] G. Kresse and J. Furthmuller, Phys. Rev. B 54, 11169 (1996).
(2014). [25] H. Brooks, Adv. Electron. Electron Phys. 7, 85 (1955).
[8] N. Miller, E. E. Haller, G. Koblmuller, C. Gallinat, J. S. Speck, [26] G. Stillman, C. Wolfe, and J. Dimmock, J. Phys. Chem. Solids
W. J. Schaff, M. E. Hawkridge, K. M. Yu, and J. W. Ager, Phys. 31, 1199 (1970).
Rev. B 84, 075315 (2011). [27] T. Mattila and R. M. Nieminen, Phys. Rev. B 54, 16676 (1996).
[9] O. D. Restrepo, K. Varga, and S. T. Pantelides, Appl. Phys. Lett. [28] A. Janotti, J. L. Lyons, and C. G. Van de Walle, Phys. Status
94, 212103 (2009). Solidi A 209, 65 (2012).
[10] G. K. Madsen and D. J. Singh, Comput. Phys. Commun. 175, [29] B. Liu, M. Gu, Z. Qi, X. Liu, S. Huang, and C. Ni, Phys. Rev. B
67 (2006). 76, 064307 (2007).
[11] D. L. Rode, Phys. Rev. B 2, 1012 (1970). [30] X. Gonze and C. Lee, Phys. Rev. B 55, 10355 (1997).
[12] G. Ng, D. Vasileska, and D. K. Schroder, J. Appl. Phys. 106, [31] A. Togo, F. Oba, and I. Tanaka, Phys. Rev. B 78, 134106
053719 (2009). (2008).
[13] M. Lundstrom, Fundamentals of Carrier Transport, 2nd ed. [32] Y. Takeda and A. Sasaki, Solid-State Electron. 27, 1127 (1984).
(Cambridge University Press, Cambridge, UK, 2009). [33] M. Cardona and N. E. Christensen, Phys. Rev. B 35, 6182 (1987).
[14] D. K. Ferry, Semiconductor Transport (Taylor & Francis, [34] A. Beya-Wakata, P.-Y. Prodhomme, and G. Bester, Phys. Rev.
London, 2000). B 84, 195207 (2011).
[15] A. T. Ramu, L. E. Cassels, N. H. Hackman, H. Lu, J. M. O. Zide, [35] K. Sarasamak, S. Limpijumnong, and W. R. L. Lambrecht, Phys.
and J. E. Bowers, J. Appl. Phys. 109, 033704 (2011). Rev. B 82, 035201 (2010).
[16] A. T. Ramu, L. E. Cassels, N. H. Hackman, H. Lu, J. M. O. Zide, [36] P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964).
and J. E. Bowers, J. Appl. Phys. 107, 083707 (2010). [37] W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
[17] A. T. Ramu and J. E. Bowers, Appl. Phys. Lett. 101, 173905 [38] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77,
(2012). 3865 (1996).

235123-13
FAGHANINIA, AGER III, AND LO PHYSICAL REVIEW B 91, 235123 (2015)

[39] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 78, [55] L. Hsu, R. E. Jones, S. X. Li, K. M. Yu, and W. Walukiewicz, J.
1396 (1997). Appl. Phys. 102, 073705 (2007).
[40] P. E. Blochl, Phys. Rev. B 50, 17953 (1994). [56] J. Wu, W. Walukiewicz, W. Shan, K. M. Yu, J. W. Ager, E. E.
[41] G. Kresse and D. Joubert, Phys. Rev. B 59, 1758 (1999). Haller, H. Lu, and W. J. Schaff, Phys. Rev. B 66, 201403 (2002).
[42] R. Juza and H. Hahn, Z. Anorg. Allg. Chem. 239, 282 (1938). [57] P. Schley, R. Goldhahn, G. Gobsch, M. Feneberg, K. Thonke,
[43] R. W. G. Wyckoff, Crystal Structures, 2nd ed. (Interscience X. Wang, and A. Yoshikawa, Phys. Status Solidi B 246, 1177
Publishers, New York, 1963), Vol. 1. (2009).
[44] A. A. Mostofi, J. R. Yates, Y.-S. Lee, I. Souza, D. Vanderbilt, [58] M. Goiran, M. Millot, J.-M. Poumirol, I. Gherasoiu, W.
and N. Marzari, Comput. Phys. Commun. 178, 685 (2008). Walukiewicz, and J. Leotin, Appl. Phys. Lett. 96, 052117 (2010).
[45] J. Paier, M. Marsman, K. Hummer, G. Kresse, I. Gerber, and J. [59] S. P. Fu and Y. F. Chen, Appl. Phys. Lett. 85, 1523 (2004).
Angyan, J. Chem. Phys. 124, 154709 (2006). [60] T. Inushima, K. Fukui, H. Lu, and W. J. Schaff, Appl. Phys. Lett.
[46] J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 92, 171905 (2008).
8207 (2003). [61] W. Walukiewicz, S. X. Li, J. Wu, K. M. Yu, J. W. Ager III,
[47] A. V. Krukau, O. A. Vydrov, A. F. Izmaylov, and G. E. Scuseria, E. E. Haller, H. Lu, and W. J. Schaff, J. Cryst. Growth 269, 119
J. Chem. Phys. 125, 224106 (2006). (2004).
[48] V. I. Anisimov, J. Zaanen, and O. K. Andersen, Phys. Rev. B 44, [62] D. Fritsch, H. Schmidt, and M. Grundmann, Phys. Rev. B 69,
943 (1991). 165204 (2004).
[49] A. Molina-Sanchez, A. Garca-Cristobal, A. Cantarero, A. [63] C. Bungaro, K. Rapcewicz, and J. Bernholc, Phys. Rev. B 61,
Terentjevs, and G. Cicero, Phys. Rev. B 82, 165324 (2010). 6720 (2000).
[50] W. Wei-Hua, Z. Guo-Zhong, and L. Xi-Xia, Chin. Phys. B 22, [64] S. E. Laux, M. V. Fischetti, and D. J. Frank, IBM J. Res. Dev.
120205 (2013). 34, 466 (1990).
[51] J. Furthmuller, P. H. Hahn, F. Fuchs, and F. Bechstedt, Phys. [65] Y. Takei and T. Nakayama, J. Cryst. Growth 311, 2767
Rev. B 72, 205106 (2005). (2009).
[52] C. Persson and S. Mirbt, Braz. J. Phys. 36, 286 (2006). [66] L. F. J. Piper, T. D. Veal, C. F. McConville, H. Lu, and W. J.
[53] I. Vurgaftman, J. R. Meyer, and L. R. Ram-Mohan, J. Appl. Schaff, Appl. Phys. Lett. 88, 252109 (2006).
Phys. 89, 5815 (2001). [67] V. Lebedev, V. Cimalla, T. Baumann, O. Ambacher, F. M.
[54] O. V. Emelyanenko, D. N. Nasledov, V. G. Sidorov, V. A. Morales, J. G. Lozano, and D. Gonzalez, J. Appl. Phys. 100,
Skripkin, and G. N. Talalakin, Phys. Status Solidi B 12, K93 94903 (2006).
(1965). [68] W. Kohn and J. M. Luttinger, Phys. Rev. 108, 590 (1957).

235123-14