Semiconductor Science and

Technology

You may also like

- Effects of a semiconductor matrix on the
Band anticrossing in highly mismatched III–V band anticrossing in dilute group II-VI
oxides
semiconductor alloys M Wena, R Kudrawiec, Y Nabetani et al.

- Tunable strong magnon-magnon coupling

in two-dimensional array of diamond
To cite this article: J Wu et al 2002 Semicond. Sci. Technol. 17 860 shaped ferromagnetic nanodots
Sudip Majumder, Samiran Choudhury,
Saswati Barman et al.

- Role of electronic excitations in magneto-

Raman spectra of graphene
View the article online for updates and enhancements. Oleksiy Kashuba and Vladimir I Fal'ko

This content was downloaded from IP address 193.255.3.69 on 26/09/2024 at 15:34

INSTITUTE OF PHYSICS PUBLISHING SEMICONDUCTOR SCIENCE AND TECHNOLOGY
Semicond. Sci. Technol. 17 (2002) 860–869 PII: S0268-1242(02)37916-1

Band anticrossing in highly mismatched

III–V semiconductor alloys
J Wu1,2, W Shan3 and W Walukiewicz2
1
Applied Science and Technology Graduate Group, University of California, Berkeley,
CA 94720, USA
2
Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720,
USA
3
OptiWork, Inc., Fremont, CA 94538, USA

Received 29 March 2002

Published 12 July 2002
Online at stacks.iop.org/SST/17/860
Abstract
In this paper we review the basic theoretical aspects as well as some
important experimental results of the band anticrossing effects in highly
electronegativity-mismatched semiconductor alloys, such as GaAs1−xNx and
InyGa1−yAs1−xNx. The many-impurity Anderson model treated in the
coherent potential approximation is applied to these semiconductor alloys,
in which metallic anion atoms are partially substituted by a highly
electronegative element at low concentrations. Analytical solutions of the
Green’s function provide dispersion relations and state broadenings for the
restructured conduction bands. The solutions also lead to the physically
intuitive and widely used two-level band anticrossing model. Significant
experimental observations, including large bandgap reduction, great electron
effective mass enhancement and unusual pressure behaviour of the bandgap,
are compared with the predictions of the band anticrossing model. The band
anticrossing model is extended over the entire Brillouin zone to explain the
pressure behaviour of the lowest conduction band minimum in GaP1−xNx.
Finally, we show that the band anticrossing can also account for the large
bandgap bowing parameters observed in GaAsxSb1−x, InAsySb1−y and
GaPxSb1−x alloys.

1. Introduction composed of materials with distinctly different properties.

The properties of these ‘highly mismatched’ alloys (HMAs)
The understanding of the electronic structure of random deviate drastically from the predictions of the VCA. The most
semiconductor alloys has been one of the interesting and prominent class of HMAs comprises the III–V1−x–Nx alloys,
important areas of semiconductor research for a long time. in which highly electronegative nitrogen substitutes group V
The simplest treatment of such alloys is the virtual crystal anions in group III–V compounds. Nitrogen incorporation
approximation (VCA) that is applicable to perfectly random into III–V semiconductors results in numerous effects, such
alloys [1, 2]. In this approximation, the electronic properties as a dramatic reduction in the fundamental bandgap [3], an
of the alloys are given by the linear interpolation between
increase in the electron effective mass [4] and a significant
the properties of the end-point materials. The alloy disorder
decrease in the electron mobility [5].
effects are typically included through a bowing parameter that
describes the deviations from the VCA. Most of the studies of The most striking and practically important feature of the
semiconductor alloy systems are restricted to the cases where HMAs was a large bandgap reduction. For example, it has
there are only small differences between properties of the end- been found that incorporation of only 1% of N into GaAs
point semiconductors. Such ‘well-matched’ alloys can be reduces the energy gap by about 180 meV [3]. Similarly,
easily synthesized and their properties are close to the VCA large bandgap reductions were also observed in other group
predictions. III–V1−x–Nx alloys [6–9]. The first attempts to explain
Recent progress in the epitaxial growth techniques has this unusual behaviour were based on a dielectric model
led to the successful synthesis of semiconductor alloys that predicted highly nonlinear composition dependencies

0268-1242/02/080860+10$30.00 © 2002 IOP Publishing Ltd Printed in the UK 860

 Band anticrossing in highly mismatched III–V semiconductor alloys

of the bandgap for the alloys of semiconductor compounds eigen-energy of the virtual state defines the width of the density
with very different properties [10]. The model predicted redistribution of the d-state, and determines the lifetime of the
a semiconductor to semi-metal transition in some of the state before the d-electrons are delocalized into the band states
alloys [11]. The large bandgap reduction in GaNxAs1−x through the exchange interaction.
alloys has also been explained by Wei and Zunger [12] in A many-impurity Anderson model has been proposed
terms of a large composition dependent bowing parameter to describe the electronic properties of semiconductor
that could be decomposed into three different contributions: crystals with low concentrations of deep-level transition-
a volume deformation, a charge exchange and a structural metal impurities [23, 24]. Very recently, the many-
relaxation. This approach was abandoned later and several impurity Anderson model has also been used to evaluate the
other theoretical explanations of the large bandgap reduction interaction between the randomly distributed localized states
in III–V–N alloys have also been proposed [13–17]. and the extended states in HMAs [25]. The calculations
Alternatively, the band structures of HMAs have been have reproduced the two-level BAC model results for the
explained in terms of the two-level band anticrossing (BAC) restructuring of the conduction band. The imaginary part of the
model. The BAC model has been developed to explain the Green’s function also yields new information on the electronic
pressure and composition dependencies of the bandgap of level broadening that is used to determine the broadening of the
InyGa1−yAs1−xNx alloys [18]. Later it was successfully applied optical transitions and to calculate the free electron mobility.
to other HMAs [19, 20]. Furthermore, it has predicted several In this paper, the theoretical aspects of the BAC model are
new effects, such as a N-induced enhancement of the electron reviewed on the basis of the many-impurity Anderson model.
effective mass [4], an improvement in the donor activation Comparisons are made between important theoretical results
efficiency [21] in InyGa1−yAs1−xNx alloys and the change in and basic experimental observations for III–V1−x–Nx alloys.
the nature of the fundamental bandgap from indirect to direct The basic BAC model is also extended to the entire Brillouin
in GaP1−xNx [20]. In the meantime, all these predictions have zone to explain new experimental findings.
been experimentally confirmed.
In the BAC model, the restructuring of the conduction
band is a result of an anticrossing interaction between highly 2. Theory
localized A1 states of the substitutional N atoms and the
extended states of the host semiconductor matrix. The newly 2.1. Band restructuring: many-impurity Anderson model in
formed sub-bands, named E+ and E−, have dispersion relations the coherent potential approximation
given by [18]

  Hjalmarson et al were the first to predict that the incorporation
E± (k) = 12 [E C (k) + E L ] ± [E C (k) − E L ]2 + 4V 2 · x , of isoelectronic impurities into semiconductors gives rise
(1) to highly localized impurity states [26]. The energy
of these states largely depends on the electronegativity
C
where E (k) is the energy dispersion of the lowest conduction of the substitutional impurity. In the case of highly
band of the host, and EL is the energy of the localized electronegative impurities substituting metallic anions in
states derived from the substitutional N atoms. The coupling compound semiconductors, the energy levels are located close
between the localized states and the band states of the host to the conduction band edge [26]. For example, substitutional
is described by the adjustable parameter V . The BAC nitrogen, an isoelectronic impurity in GaAs, generates an A1-
model provides a simple, analytical expression to calculate symmetry level located at ∼0.23 eV above the conduction
the electronic and optical properties of III–V1−x–Nx alloys. band edge of GaAs resonant with the conduction band. This
The two-level BAC model is a natural result of degenerate resonant level has been observed in optical experiments when
perturbation theory applied to a system of localized states it is moved into the GaAs bandgap by applying hydrostatic
and extended states. The interaction between these two pressure [27] or alloying with GaP [28]. With increasing N
types of states has been treated in the simplest possible impurity concentration, the interaction between the localized
manner that does not account for expected severe level N levels and the GaAs band states changes the electronic
broadening effects. It is enlightening to notice the similarities structure of the resulting GaAs1−xNx system.
between this localized extended hybridization problem and The electronic structure of HMAs (e.g., GaAs1−xNx)
the s–d exchange interaction for a single-impurity atom of can be described by considering an interaction between the
a transition metal or a rare-earth element in a non-magnetic localized states (|L) and extended states (|k) within the
metal. The single-site Anderson model has been developed to many-impurity Anderson model. The total Hamiltonian of
describe the latter. In Anderson’s s–d exchange model [22], the system is the sum of three terms [23, 24],
the electron system is separated into a delocalized part of the  
matrix metal that is described in terms of band theory, and a H = EkC ck+ ck + EjL dj+ dj
localized level of the d-shell electrons of the impurity atom. k j
A dynamical mixing term is introduced into the Hamiltonian 1  ik·j
of the system to describe the hybridization between the band +√ (e Vkj ck+ dj + h.c.), (2)
N j,k
states and the localized impurity states. While solving the
Anderson Hamiltonian, it has been found that, as a result where the first term is the Hamiltonian of the electrons in
of the hybridization, the impurity d-state becomes a virtual the band states with energy dispersion EkC , and the second
energy state with an imaginary energy part proportional to the term corresponds to the total energy of electrons localized on
strength of the s–d hybridization. The imaginary part of the impurity sites. The third term describes the change in the

861
J Wu et al

single electron energy due to the dynamical mixing between 2.2

the band states and the localized states. It is assumed that only GaAs
0.995
N
0.005
one band and one impurity level are involved in the process. 300K
Following Anderson’s scheme, the hybridization strength is 2.0
characterized by the parameter Vkj defined by the following
E (k)
equation [22], +
 1.8 C
1  ik·l E

E (eV)
k
k|HHF |L = √ e a ∗ (r − l)HHF (r)ϕL (r − j) dr
N l,j
L
1  ik·j  ik(l−j)
1.6 E
=√ e · e
N j l
 1.4
× a ∗ (r − l)HHF (r)ϕL (r − j) dr E (k)
-

1  ik·j 1.2
≡√ e Vkj , (3) -15 -10 -5 0 5 10 15
N j =1 -6 -1
k (10 cm )
where a(r − l) and ϕ L(r − j) are the Wannier function
belonging to the band and the localized wavefunction of the Figure 1. Conduction band restructuring according to equation (1)
impurity on the jth site, respectively. HHF (r) is the single for GaAs0.995N0.005. The broadening of the dispersion curves of the
newly formed sub-bands illustrates the energy uncertainties defined
electron energy described in the Hartree–Fock approximation. in equation (7). All the energies are referenced to the top of the
When there is only one impurity atom present, the model valence band of GaAs.
reduces to the original Anderson model. In general, one could
consider finite but dilute concentrations of impurities, 0 < the upper and lower conduction sub-bands. The flattened and
x  1. Assuming that the impurities are randomly distributed downshifted lower sub-band is responsible for most of the
in space, one can carry out a configurational averaging, unusual effects, such as the drastic bandgap reduction and the
neglecting correlations between positions of the impurities. electron effective mass enhancement. These effects will be
The coherent potential approximation (CPA) is usually used to addressed in section 3 of the paper.
treat such systems [29, 30]. In the CPA, consecutive multiple
√When the broadening L is nonzero  but small, so that
scatterings from each single-impurity atom are fully taken into 2V x  πβV 2 ρ0 (E L ) and EkC −E L   πβV 2 ρ0 (E L ), one
account, but correlations between scatterings from different obtains an approximate analytical solution for equation (5),
impurity atoms are neglected due to the lack of coherence
between the randomly distributed impurity sites. After the E± (k) − EkC
configurational averaging in the frame of CPA, the average Ẽ ± (k) ≈ E± (k) + iL
E± (k) − EkC + [E± (k) − E L ]
Green’s function restores the space translational invariance,
≡ E± (k) + i± (k), (6)
and k resumes its well-defined properties as a good quantum
number. In momentum space, the diagonal Green’s function where E± (k) is the real part of Ẽ ± (k) and is defined in
can be written as [23, 29, 30] equation (1). The imaginary part of the dispersion relations
 −1 defines the hybridization-induced energy uncertainties. It
V 2x
Gkk (E) = E − Ek − C
, (4) is worth noticing that the imaginary part in equation (6)
E − E L − iπβV 2 ρ0 (E L ) is proportional to the admixture of the localized states to
where V is the average value of Vkj , assuming weak the restructured wavefunctions in the two-level-perturbation
dependencies on k and j, and ρ0 is the unperturbed density picture described by equation (5),
of states of EkC per unit cell. Since ρ0 only weakly depends on
± (k) = |L|E± (k)|2 · L . (7)
energy, it can be assumed to be a constant in the lowest order
approximation, with an effective value equal to the unperturbed As an example, figure 1 shows the schematic dispersion
density of states evaluated at EL and multiplied by a prefactor relations for the sub-bands given by equation (1) for
β. The new dispersion relations are determined by the poles GaAs0.995N0.005 near the Brillouin zone centre. The
of Gkk (E), and the solutions are given by an equivalent two- broadenings of the dispersion relations are given by the
level-like eigenvalue problem, imaginary part of equation (6).
 
E C − E(k) V √x  We note that the original Anderson Hamiltonian
 k 
 √  = 0, (5) (equation (2)) and the coherent potential approximation ignore
V x E + iL − E(k)
L
interactions between neighbouring impurity states. The spatial
where L = πβV 2 ρ0 (E L ) is the broadening of EL in the distribution of the impurity atoms is assumed random and
original single-impurity Anderson model [22]. homogeneous. The application of this model is in principle
limited to dilute impurity concentrations for which the distance
between neighbouring impurity atoms is much larger than
2.2. Simplified representation: two-level band anticrossing
the spatial extension of the impurity wavefunction (e.g., the
model
wavefunction of the N resonant states). The nitrogen cluster
If  L = 0, equation (5) is reduced to the two-level BAC luminescence has been experimentally observed in GaAs1−xNx
model of equation (1) with two restructured dispersions for at dilute N concentrations [31]. It should be emphasized

862
 Band anticrossing in highly mismatched III–V semiconductor alloys

1.5

E1
1.4 GaAs N @ 295K E0 E0+∆ 0
1-x x

x=0
Energy gap (eV)

1.3

1.2 E_+∆0
E_
E+ E1

PR Signal (arb.units)
1.1 Uesugi, et. al. x=0.008
Keyes, et. al.
1 Malikova, et. al.
Bhat, et. al. E_+∆0
E+ E1
BAC theory E_
0.9
0 0.01 0.02 0.03 0.04 0.05
x=0.015
Nitrogen fraction, x
E1
Figure 2. Energy gap of GaAs1−xNx as a function of N E_ E_+∆ 0 E+
concentration from various reports [3, 31–33]. The solid curve is the
fitting of the BAC model.

x=0.02
however that the observation of photoluminescence lines
attributed to specific N clusters does not imply that the 295 K
concentration of such clusters is large enough to significantly GaAs1-xNx
change the concentration of the isolated N atoms and affect
0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4
the conclusions based on the BAC model. It has been shown
previously that small inter-impurity interaction effects may Photon Energy (eV)
be accounted for within the BAC model by assuming that (a)
EkC , EL and the coupling parameter depend on the N
3.0
concentration [32].

3. Comparison with experiment E

Transition Energy (eV)

-
2.5
E +∆
- 0
3.1. Fundamental energy gap
E
+
The bandgap reduction in GaAs1−xNx alloys is accounted for E
1
by the downward shift of the conduction band edge caused by 2.0
the anticrossing interaction. Figure 2 shows the fundamental
bandgap as a function of N concentration from various reports
[3, 33–35] together with the dependence calculated from 1.5
equation (1). The best fit is obtained with an interaction
constant of V = 2.7 eV.
The additional E+ transition at energies blue-shifted from
the previously observed N resonant level [18] suggests a 1.0
0 0.005 0.01 0.015 0.02
strong interaction between the N states and the host GaAs.
Figure 3(a) shows the photomodulated reflectance (PR) spectra x
(b)
for GaAs1−xNx for a range of N concentrations [36]. The
critical energies are plotted as a function of x in figure 3(b). Figure 3. (a) Photomodulated reflectance spectra showing the
In figure 3(b) it can be seen that the spin–orbit transition critical transitions. (b) The critical energies obtained from (a)
energy E− + 0 follows E− , indicating that the top of the plotted as a function of x.
valence band of GaAs is not affected by the N incorporation.
The symmetric shift of E+ and E− is a typical behaviour
for interacting two-level systems. On the other hand, the E1 in their pressure dependencies. Figure 4 shows the energy
energy that corresponds to the inter-band transition along the  position of the E+ and E− transitions for GaAs0.985N0.015
line in the Brillouin zone, exhibits a much weaker composition as a function of hydrostatic pressure [36]. The solid
dependence. This implies that away from the Brillouin zone curves represent the pressure dependencies calculated from
centre, the effect of N incorporation on the band structure of equation (1), using previously known pressure coefficients
GaAs is much weaker. dE C /dP = 10.8 meV kbar−1 for the GaAs conduction band
The anticrossing character of E+ and E− is not only edge, and dE L /dP = 1.5 meV kbar−1 for the N-localized
reflected in the composition dependence, but also manifested states.

863
J Wu et al

2.6 1.40
E
2.4 +
1.35
C
E
2.2 1.30 E
Energy (eV)

E (eV)
2 1.25
E L
X E
1.8 1.20 E
1
Ebulk
1.6 1.15
E GaAs0.985 N0.015
1.4 -
1.10
295K 0.01 0.015 0.02 0.025 0.03
1.2 x
0 20 40 60 80 100 120 (a)
Pressure (kbar)
1.40
Figure 4. Energies of the E+ and E− transitions obtained from PR
experiments as a function of hydrostatic pressure for GaAs0.985N0.015.
1.35
E1
1.40 1.30
E (eV)
InP1-xNx, 295K E
1.35 2
1.25
Band gap (eV)

1.30
1.20 E
1.25 bulk

1.20 1.15
BAC fitting 2 3 4 5 6 7 8 9
absorption, Bi et. al. well width (nm)
1.15
PR (b)

1.10 Figure 6. (a) First and second transition energies as a function of N

0.000 0.002 0.004 0.006 0.008 0.010 concentration for GaAs1−xNx/GaAs QWs with 7 nm well width.
x Solid curves: calculated values using the BAC model with the
GaAs1−xNx electron effective mass given by equation (8); Short
Figure 5. Energy gap of InP1−xNx as a function of x. The solid line dashed curves: calculated values assuming the GaAs1−xNx electron
is a fit based on the BAC model. effective mass is equal to m∗GaAs. Long dashed curve: bandgap of
bulk GaAs1−xNx given by the BAC model. (b) The transition
energies as a function of well width for x = 0.016. The different
Similar band anticrossing effects have also been observed types of curves represent the same calculations respectively as in (a).
in several other N-containing III–V alloys, such as GaP1−xNx
[6], InP1−xNx [7], GaSbyAs1−x−yNx [8] and InSb1−xNx [9].
band lineup with a very small negative valence-band offset of
Shown in figure 5 is the bandgap bowing of InP1−xNx
grown by gas-source molecular beam epitaxy with small N |EV | < 20 meV/%N.
concentrations [37]. The energy of the localized N-level EL = The inter-band optical transitions in GaAs1−xNx/GaAs
2.0 eV above the valence band edge of InP was estimated from multiple QWs have been measured by PR experiments [40].
the valence band offset of 0.35 eV between InP and GaAs. Figure 6(a) shows the transition energies as a function of the
A BAC fitting leads to a coupling constant of V = 3.5 eV in N concentration for a 7 nm multiple QW with 20 nm GaAs
InP1−xNx. barriers. These energies, denoted as E1 and E2, are attributed
to the critical transitions from the valence band edge to the
3.2. BAC effects in GaAs1−xNx/GaAs quantum wells ground state and first excited state, respectively, in the confined
E− sub-band. To understand the composition dependence
The band anticrossing interaction also has significant effects
quantitatively, we have applied a finite-depth single-square-
on the optical transitions in GaAs1−xNx/GaAs quantum wells
well model to calculate the confinement energies. The well
(QW). Since the perturbation of N on the band structure of
GaAs is mainly in the lowest conduction band, it can be depth for the confined electrons in the GaAs1−x Nx layer is given
expected that the band offset between GaAs1−xNx and GaAs is by the downward shift of the conduction band determined
mostly in the conduction band. Indeed, photoluminescence by equation (1). The effective mass inside the well is
[38] and x-ray photoelectron spectroscopy [39] studies of approximated by the density-of-state electron effective mass
GaAs1−xNx/GaAs heterostructures indicate a slightly type-II evaluated at the Brillouin zone centre,

864
 Band anticrossing in highly mismatched III–V semiconductor alloys

3.2
1.6 GaAs / GaAs N / GaAs
0.984 0.016
well width 7nm 3 E
+

1.5 2.8

Energy (eV)
E (eV)

2.6
1.4 GaP N @ 295K
1-x x
Ebulk 2.4

1.3 2.2

2 E
1.2 -
0 10 20 30 40 50 60 70
1.8
P (kbar)
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Figure 7. The first transition energy, E1, as a function of hydrostatic
pressure for x = 0.016 and well width = 7 nm. The different types N composition, x
of curves have the same meanings as in figure 6.
Figure 8. E+ and E− measured by PR as a function of x for
  GaP1−xNx.

2 k 
∗
m ≈ h̄  
dE− (k)/dk k=0 3.6
3.4
E C (0) − E L 3.2 EΓ
= 2m∗GaAs 1−  . (8) 3.0
(E C (0) − E L )2 + 4V 2 x
2.4
The effective mass calculated in equation (8) increases rapidly
with x from m∗GaAs = 0.067m0 for pure GaAs to about 0.1m0 GaP N , 30K
1-x x
E
at x = 0.01. The solid curves in figure 6(a) are the calculated 2.3 X
E (eV)

transition energies assuming a composition dependence of the

x=0.007
effective mass given by equation (8). To illustrate the effect of x=0.017
2.2 E
L
the heavier electron mass in the QW, we have also calculated x=0.023
the optical transition energies assuming that the effective mass
of GaAs1−xNx is unchanged from that of GaAs. The results 2.1
are shown as short-dashed curves in figure 6(a). Evidently,
much better agreement with the experiment is obtained when
2.0 L
the effect of mass enhancement is taken into account. It is L E
E X-
Γ-
also necessary to include the heavier mass effect to explain the
well width dependence of the transition energies. Figure 6(b) 1.9
0 20 40 60 80 100 120 140
shows E1 and E2 as a function of the well width for a constant
P (kbar)
N concentration of x = 0.016.
The pressure dependence of the transition energy further Figure 9. Photoluminescence peak energy at 30 K as a function of
supports the conclusion of heavier electron mass in the well pressure for three samples with different N concentrations. The
layer. In figure 7 E1 is shown as a function of hydrostatic dashed lines indicate the pressure dependencies of the N-localized
pressure for a QW. At high pressures, the experimental state, the  band and the X band minima of GaP. The solid curves
are the calculated lowest conduction band energies for the three N
data-points deviate from the predicted values based on the concentrations based on the BAC model.
assumption of constant effective mass, and can only be
accommodated by the heavier mass calculations. It should
of the conduction band and the N level, which is located at
be noted that with the increasing pressure, the electron
2.15 eV above the top of the valence band. The interaction
effective mass is further increased as a result of the stronger
strength has been determined to be V = 3.05 eV. Figure 8
flattening of E−(k). For example, for x = 0.016, the mass
shows the energies of E− and E+ observed using PR
increases from 0.11m0 at ambient pressure to 0.28m0 at 70 kbar,
spectroscopy as a function of N concentration, in comparison
two fold enhanced from the electron effective mass of the GaAs
with the BAC calculations.
host. Similar increases in the effective mass of GaAs1−xNx
Since the wavefunction of N states is highly localized in
have been reported using various experimental techniques
real space, its Fourier transform has significant contributions
[41, 42].
by off-zone-centre components in the Brillouin zone. It is,
therefore, expected to couple not only to the  conduction
3.3. Band anticrossing in the entire Brillouin zone
state, but also to other conduction states such as the L and X
It has been demonstrated that the incorporation of N into GaP band minima. We have investigated the hydrostatic pressure
changes the nature of the fundamental bandgap from indirect dependence of the fundamental bandgap of GaP1−xNx alloys
to direct [20]. A new conduction band minimum is formed at using photoluminescence (PL) technique at low temperature
the  point as a result of the anticrossing between the  edge (30 K) [43]. Our results are summarized in figure 9. It can be

865
J Wu et al

0.03
4.0 E
+ x=0
x=0.001 GaAs N
3.6 0.025 1-x x

ρ( E) (1/eV unit-cell)
x=0.005
x=0.01
3.2 C
E 0.02 x=0.02
2.8

2.4 0.015
E (eV)

0.01
2.3

0.005
L
2.2 E
0
1 1.2 1.4 1.6 1.8 2 2.2
2.1
E E (eV)
-

Figure 11. Density of states of GaAs1−xNx alloys for a range of

2.0
values of x as compared with the unperturbed density of states. The
two black dots on each curve indicate the energy positions of the E−
and E+ sub-band edges.
1.9
L Γ X
exponentially decaying function, specifically [25],
Figure 10. BAC-induced restructuring of the conduction band of
GaP1−xNx. The curves show the lower E−(k) and the upper E+(k) V
V (k) = . (9)
bands at ambient pressure. The symbols show the energy minima at [1 + (ak)2 ]2
100 kbar: open triangles, N-localized level EL; open circles, EC
(including E and EX); solid triangles, E− (including EL− and
In equation (9), a is a parameter describing the spatial extent
ELX−); solid circles, E+ (including EL+ and ELX+). Note the change of the N localized wavefunction, and has been estimated to
in the nature of the lowest conduction band edge, from direct at be of the order of the lattice constant of the host, so that VX
ambient pressure to indirect, mostly X-like at 100 kbar. agrees with the value determined from the X–N anticrossing
fitting. The energy minima restructured from the –N and
seen that at high pressures, the fundamental bandgap shifts to X–N anticrossing interactions at 100 kbar are also shown as
lower energy, approaching the negative pressure dependency symbols in figure 10. The fundamental bandgap of the material
of the X band minimum of GaP. The non-monotonic pressure switches from direct at ambient pressure to indirect at high
dependence of the fundamental bandgap is explained by pressures.
the interactions of the N-localized states with the  band
edge of GaP at low pressures and with the X band edge at 3.4. State broadening and related effects
high pressures. The interaction constant is determined to It is interesting to see the effects of band anticrossing on the
be V = 0.90 eV at the X point, indicating a much weaker density of states of the conduction band. The original density
coupling than at the  point. of states per unit cell for the GaAs conduction band edge is
Evidence of similar off-zone-centre coupling in assumed to have a parabolic form following the effective-mass
GaAs1−xNx alloy has been observed by Seong et al using theory,
resonant Raman scattering [44] and by Perkins et al using   3/2
electroreflectance technique [45]. In these ambient-pressure ρ0 (ε) = 4π ε − E0C εB , (10)
experiments, a coupling between N states and the conduction
where εB = h̄2 (2π/b)2 /(2m∗GaAs ) is of the order of the
band edge at the L point of the Brillouin zone was observed.
conduction band width, b = 5.65 Å is the lattice constant
This is because in GaAs, the N level is closer to the L
of the unit cell and m∗GaAs is the electron effective mass of
band minima (separation = 0.12 eV) than to the X minima
GaAs.
(separation = 0.27 eV), whereas in GaP it is located much
The perturbed density of states can be calculated from
closer to the X band minima than to the L minima, with
the imaginary part of the Green’s function and is given by the
separations equal to 0.17 eV and 0.46 eV, respectively. In expression
GaAs, the negative pressure derivative resulting from the 
1  1  
coupling between EL-localized levels and EX minima can be ρ(E) = Im Gkk (E) = ρ0 EkC Im[Gkk (E)] dEkC .
observed only at very large pressures when the X band minima π k
π
are shifted down to the band edge. In fact, the onset of this (11)
effect has shown up in GaAs1−xNx at a pressure of about
100 kbar, as can be seen in figure 4. Shown in figure 10 is a The integration converges rapidly with EkC in a small range
schematic diagram of the conduction bands of GaP0.977N0.023 that is proportional to x. The calculated perturbed density of
at ambient pressure. In the calculation, V (k) is assumed states for GaAs1−xNx with several small values of x is shown
to take the functional form of the Fourier transform of an in figure 11. Note that the anticrossing interaction leads to

866
 Band anticrossing in highly mismatched III–V semiconductor alloys

4.0 1.0
In Ga
0.04 0.96
As N
0.99 0.01
, 295K E =1.6 eV
3.5 As

Conduction band edge (eV)

C =0.95 eV E (x)
total AsM M
3.0 0.8
α (E ) (104 cm-1)

heavy hole
light hole
2.5 split off
experimental
2.0 0.6 GaAs Sb
x 1-x

1.5

1.0 0.4

0.50
E (y)
M
0.2
0.0
1 1.2 1.4 1.6 1.8 2 2.2 InAs Sb
y 1-y
E (eV)
0.0
Figure 12. Calculated optical absorption coefficient in comparison 0 0.05 0.1 0.15 0.2 0.25 0.3
with room-temperature experimental data [44] for free-standing
In0.04Ga0.96As0.99N0.01. The oscillations below the absorption edge
Composition, x or y
are due to Fabry–Perot interference.
Figure 13. Energy of the conduction band edge in GaAsxSb1−x [46]
and InAsySb1−y [47]. The energies are relative to the valence band
a considerable redistribution of the electronic states in the maximum of GaSb. The solid lines are calculated using the BAC
conduction band. The most striking feature of the density of model. EM(x) and EM(y) represent the energies of the conduction
band minimum obtained from the virtual crystal approximation.
states curves is the clearly seen gap between E+ and E− that
evolves with increasing N content.
In order to illustrate the effect of the state broadening on of materials. It has been shown recently that it includes
the optical properties, we consider the spectral dependence of group II–VI–O alloys in which highly electronegative O
the interband absorption in InyGa1−yAs1−xNx alloys. The partially replaces more metallic group VI elements. Large
band anticrossing effects were observed in Cd1−yMnyOxTe1−x
optical absorption coefficient due to the transitions from the
with less than one per cent of oxygen [47]. Because of the
valence bands to the restructured conduction bands can be
high electronegativity of N and O atoms, both group III–V–N
written in the form of the joint density of states as
 and group II–VI–O alloys represent extreme cases of HMAs
1      in which the band anticrossing interaction leads to dramatic
α(E) ∝ ρ0 EkC Im Gkk E + Ekv dEkC . (12)
E v changes of the conduction band structure. There are, however,
many of the III–V or II–VI alloys formed of anions with a
In this expression, the sum over v represents the sum of
smaller but still significant electronegativity difference. For
the contributions from the heavy-hole, the light-hole and the
example, it has been shown that the composition and pressure
spin–orbit split-off valence bands. Assuming parabolic forms
dependencies of the energy gap of ZnTe1−xSx and ZnTe1−ySey
for the dispersions of the valence bands near the Brillouin
can be described well by an anticrossing interaction of the
zone centre, we have calculated the optical absorption for
localized levels of S and Se, respectively, with the extended
In0.04Ga0.96As0.99N0.01 for which experimental results are
states of the ZnTe matrix [19]. The effects of the anticrossing
available [46]. The comparison between the calculation and
are clearly observed despite much smaller electronegativity
the experimental data is shown in figure 12. In the calculation,
difference of 0.4 for ZnTe1−xSx and 0.3 for ZnTe1−ySey alloys.
the only parameter that has been adjusted is the prefactor β It is well known that large bandgap bowings are observed
used to scale the energy broadening. The best fitting with the in all group III–Sb alloys in which metallic Sb is replaced
experimental data is obtained with β = 0.22. The calculation by either P or As. An interesting question arises: could the
clearly reproduces the two bumps on the absorption curve, strong bowings also be explained by the band anticrossing
i.e., one starting at ∼1.8 eV due to the onset of the transitions interaction? Figure 13 shows the composition dependence of
from the heavy-hole and light-hole valence bands to E+, and the energy of the conduction band edge in GaAsx Sb1−x [48] and
one starting at ∼1.5 eV due to the onset of the transition from InAsySb1−y [49] along with the results of calculations based
the split-off valence band to E−. The more rapid rise of the on the BAC model. The calculations account well for the
experimental data at the absorption edge near 1.2 eV is most conduction band shifts assuming the energy of the localized
likely due to the continuum exciton absorption effect, which As level lies at 1.6 eV above the valence band of GaSb and the
is not considered in the calculation. coupling constant is 0.95 eV. The band anticrossing effects
in these materials are significantly weaker compared with
4. Band anticrossing in other group III–V alloys III–N–V alloys because of smaller electrongativity difference
of 0.2 between Sb and As.
The band anticrossing model has explained successfully the Somewhat larger anticrossing effects are expected for
strong bandgap bowing of all III–V–N HMAs. It should be III–V alloys in which Sb with electronegativity of 1.8 is
emphasized that these alloys belong to a much broader class partially replaced by P with electronegativity of 2.1. Such

867
J Wu et al

1.2 Acknowledgments
GaP Sb
1.1 x 1-x J Wu acknowledges the Berkeley Fellowship from the
Conduction band edge (eV)

C =1.25 eV E University of California, Berkeley. We wish to gratefully

PM M
1.0 E =1.5 eV acknowledge collaborations with Dr K M Yu, Dr J W Ager,
P
Professor E E Haller, Dr J F Geisz, Dr Skierbiszewski and
0.9 Professor C W Tu. The work at Lawrence Berkeley National
Laboratory is part of the project on the ‘Photovoltaic Materials
0.8 Focus Area’ in the DOE Centre of Excellence for the Synthesis
and Processing of Advanced Materials, and was supported
0.7 by the Director, Office of Science, Office of Basic Energy
Sciences, Division of Materials Sciences of the US Department
0.6 of Energy under Contract No DE-AC03–76SF00098.

0.5 References
0.4 [1] Van Vechten J A and Bergstresser T K 1970 Phys. Rev. B 1
0 0.05 0.1 0.15 0.2 0.25 0.3 3351
[2] Hill R and Richardson D 1971 J. Phys. C 4 L289
Composition, x [3] Weyers M, Sato M and Ando H 1992 Japan. J. Appl. Phys. 31
L853
Figure 14. Conduction band minimum as function of composition Uesugi K, Marooka N and Suemune I 1999 Appl. Phys. Lett.
in GaPxSb1−x. The points represent experimental data of [48]. The 74 1254
solid line shows the results of calculations with EM representing the [4] Skierbiszewski C et al 2000 Appl. Phys. Lett. 76 2409
conduction band edge obtained within the virtual crystal [5] Kurtz S R, Allerman A A, Seager C H, Sieg R M and
approximation. Jones E D 2000 Appl. Phys. Lett. 77 400
Geisz J F, Friedman D J, Olson J M, Kurtz S R and Keyes B M
1998 J. Cryst. Growth 195 401
alloys containing P and Sb are difficult to synthesize because [6] Baillargeon N, Cheng K Y, Hofler G F, Pearah P J and
of a large immiscibility. A successful synthesis of GaPxSb1−x Hsieh K C 1992 Appl. Phys. Lett. 60 2540
has shown that this alloy system exhibits a large bandgap [7] Bi W G and Tu C W 1996 J. Appl. Phys. 80 1934
bowing [50]. The energy of the conduction band edge as [8] Harmand J C, Ungaro G, Ramos J, Rao E V K,
Saint-Girons G, Teissier R, Le Roux G, Largeau L and
function of the P content is shown in figure 14. As expected, Patriarche G 2001 J. Cryst. Growth 227–228 553
the anticrossing effects are more pronounced in this case than [9] Murdin B N et al 2001 Appl. Phys. Lett. 78 1558
in the case of GaAsxSb1−x or InAsySb1−y alloys. Note that [10] Van Vechten J A and Bergstresser T K 1970 Phys. Rev. B 8
3351
there is quite a large deviation of the conduction band energy
[11] Sakai S, Ueta Y and Terauchi Y 1993 Japan. J. Appl. Phys. 32
from the prediction of the VCA represented by EM. Again, 4413
calculations based on the BAC model account very well for [12] Wei Su-Huai and Zunger A 1996 Phys. Rev. Lett. 76 664
the downward shift of the conduction band edge. [13] Jones E D, Modine N A, Allerman A A, Kurtz S R,
Wright A F, Tozer S T and Wei X 1999 Phys. Rev. B 60
4430
[14] Mattila T, Wei Su-Huai and Zunger A 1999 Phys. Rev. B 60
5. Summary R11 245
[15] Wang Lin-Wang 2001 Appl. Phys. Lett. 78 1565
Szwacki N G and Boguslawski P 2001 Phys. Rev. B 64 161201
In this paper we have reviewed the theoretical origin and [16] Al-Yacoub A and Bellaiche L 2000 Phys. Rev. B 62 10 847
resultant effects of band anticrossing between localized [17] Zhang Y, Mascarenhas A, Xin H P and Tu C W 2001 Phys.
impurity states and the extended band states of the host Rev. B 63 161303
in highly electronegativity-mismatched semiconductor alloys. [18] Shan W, Walukiewicz W, Ager J W III, Haller E E, Geisz J F,
Friedman D J, Olson J M and Kurtz S R 1999 Phys. Rev.
Some important experimental results of GaAs1−xNx, InP1−xNx, Lett. 82 1221
and GaP1−xNx have been discussed in the context of the [19] Walukiewicz W, Shan W, Yu K M, Ager J W III, Haller E E,
band anticrossing model. It is demonstrated that the band Miotkowski I, Seong M J, Alawadhi H and Ramdas A K
anticrossing model can account successfully for the unusual 2000 Phys. Rev. Lett. 85 1552
[20] Shan W, Walukiewicz W, Yu K M, Wu J, Ager J W III,
behaviour of the highly mismatched alloys. The band Haller E E, Xin H P and Tu C W 2000 Appl. Phys. Lett. 76
anticrossing interaction does not only exist between the 3251
localized states and the conduction band near the Brillouin [21] Yu K M, Walukiewicz W, Shan W, Ager J W III, Wu J,
Haller E E, Geisz J F, Friedman D J and Olson J M 2000
zone centre, but it also extends to the zone edge. As examples
Phys. Rev. B 61 R13337
for the generality of the application of the band anticrossing [22] Anderson P W 1961 Phys. Rev. 124 41
model, we have shown that the large bowing parameters [23] Kocharyan A N 1986 Soc. Phys. Solid State 28 6
observed in group III–V alloys with As or P substituting Sb [24] Ivanov M A and Yu G Pogorelov 1979 Sov. Phys.–JETP 49 510
Ivanov M A and Yu G Pogorelov 1985 Sov. Phys.–JETP 61
can also be explained by the interaction between localized As
1033
or P levels and the extended conduction band states of the [25] Wu J, Walukiewicz W and Haller E E 2002 Phys. Rev. B 65
semiconductor matrix. 233210

868
 Band anticrossing in highly mismatched III–V semiconductor alloys

[26] Harold P Hjalmarson, Vogl P, Wolford D J and John D Dow Tanaka S, Takahashi M, Moto A, Tanabe T, Takagishi S,
1980 Phys. Rev. Lett. 44 810 Karatani K, Nakanishi T and Nakayama M 1998 25th
[27] Wolford D J, Bradley J A, Fry K and Thompson J 1984 International Symposium on Compound Semiconductors
Physics of Semiconductors ed J D Chadi and W A Harrison (Nara, Japan)
(New York: Springer) [40] Wu J, Shan W, Walukiewicz W, Yu K M, Ager J W III,
[28] Hsu W Y, Dow J D, Wolford D J and Streetman B G 1977 Haller E E, Xin H P and Tu C W 2001 Phys. Rev. B 64
Phys. Rev. B 16 1597 85320
[29] Yonezawa Fumiko and Morigaki Kazuo 1973 Prog. Theor. [41] Jones E D, Modine N A, Allerman A A, Fritz I J, Kurtz S R,
Phys. 53 Supplement Wright A F, Tozer S T and Wei Xing 1999 Proc. 195th
[30] Elliott R J, Krumhansl J A and Leath P L 1974 Rev. Mod. Electrochem. Soc. Meeting (Seattle, WA 99–11) 170
Phys. 46 465 [42] Hetterich M, Dawson M D, Egorov A Yu, Bernklau D and
[31] Liu Xiao, Pistol M-E, Samuelson L, Schwetlick S and Riechert H 2000 Appl. Phys. Lett. 76 1030
Seifert W 1990 Appl. Phys. Lett. 56 1451 [43] Wu J, Walukiewicz W, Yu K M, Ager J W III, Haller E E,
[32] Lindsay A and O’Reilly E P 1999 Solid State Commun. 112 Hong Y G, Xin H P and Tu C W 2002 Phys. Rev. B 65
443 241303(R)
[33] Keyes B M, Geisz J F, Dippo P C, Reedy R, Kramer C, [44] Seong M J, Mascarenhas A and Geisz J F 2001 Appl. Phys.
Friedman D J, Kurtz Sarah R and Olson J M 1999 NCPV Lett. 79 1297
photovoltaics program review AIP Conf. Proc. 462 511 [45] Perkins J D, Mascarenhas A, Geisz J F and Friedman D J 2001
[34] Malikova L, Pollak F H and Bhat R 1998 J. Electron. Mater. Phys. Rev. B 64 121301
27 484 [46] Perlin P, Wisniewski P, Skierbiszewski C, Suski T,
[35] Bhat R, Caneau C, Salamanca-Riba L, Bi W and Tu C 1998 Kaminska E, Subramanya S G, Weber E R, Mars D E and
J. Cryst. Growth 195 427 Walukiewicz W 2000 Appl. Phys. Lett. 76 1279
[36] Shan W, Walukiewicz W, Ager J W III, Haller E E, Geisz J F, [47] Yu K M, Walukiewicz W, Wu J, Beeman J W, Ager J W III,
Friedman D J, Olson J M and Kurtz Sarah R 1999 J. Appl. Haller E E, Miotkowski I, Ramdas A K and Becla P 2002
Phys. 86 2349 Appl. Phys. Lett. 80 1571
[37] Yu K M, Walukiewicz W, Wu J, Beeman J W, Ager J W III, [48] Thomas M B, Coderre W M and Wooley J 1970 Phys. Status
Haller E E, Shan W, Xin H P, Tu C W and Ridgway M C Solidi A 2 K141
2001 J. Appl. Phys. 90 2227 [49] Coderre W M and Wooley J C 1968 Can. J. Phys. 46 1207
[38] Sun B Q, Jiang D S, Luo X D, Xu Z Y, Pan Z, Li L H and [50] Loualiche S, Le Corre A, Salaun S, Caulet J, Lambert B,
Wu R H 2000 Appl. Phys. Lett. 76 2862 Gauneau M, Lecrosnier D and Deveaud B 1991 Appl. Phys.
[39] Kitatani Takeshi, Kondow Masahiko, Kikawa Takeshi, Lett. 59 423
Yazawa Yoshiaki, Okai Makoto and Uomi Kazuhisa 1999 Jou M J, Cherng Y T, Jen H R and Stringfellow G B 1987
Japan. J. Appl. Phys. 38 5003 Appl. Phys. Lett. 52 549

869