Tunnelling and Relaxation in Semiconductor Double Quantum Wells

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 44

Home Search Collections Journals About Contact us My IOPscience

Tunnelling and relaxation in semiconductor double quantum wells

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

1997 Rep. Prog. Phys. 60 345

(http://iopscience.iop.org/0034-4885/60/3/002)

View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 134.68.190.47
The article was downloaded on 11/06/2013 at 18:10

Please note that terms and conditions apply.


Rep. Prog. Phys. 60 (1997) 345–387. Printed in the UK PII: S0034-4885(97)64909-X

Tunnelling and relaxation in semiconductor double


quantum wells

R Ferreira and G Bastard


Laboratoire de Physique de la Matière Condensée-ENS, 24 rue Lhomond, F-75005 Paris, France

Received 31 July 1996, in final form 28 October 1996

Abstract

Double quantum wells are among the simplest semiconductor heterostructures exhibiting
tunnel coupling. The existence of a quantum confinement effect for the energy levels
of a narrow single quantum well has been largely studied. In double quantum wells, in
addition to these confinement effects which characterize the levels of the isolated wells,
one faces the problem of describing the eigenstates of systems interacting weakly through
a potential barrier. In addition, the actual structures differ from the ideal systems studied
in the quantum mechanics textbooks in many aspects. The presence of defects leads, for
instance, to an irreversible time evolution for a population of photocreated carriers. This
irreversible transfer is now clearly established experimentally. The resonant behaviour of
the transfer has also been evidenced, from the study of biased structures. If the existence
of an interwell transfer is now clearly established from the experimental point of view, its
theoretical description, however, is not fully satisfactory.
This review focuses on the theoretical description of the energy levels and of the
interwell assisted transfer in double quantum wells. We shall firstly outline the problem of
tunnel coupling in semiconductor heterostructures and then discuss the single particle and
exciton eigenstates in double quantum wells. In the remaining part of the review we shall
present and critically review a few theoretical models used to describe the assisted interwell
transfer in these structures.

0034-4885/97/030345+43$59.50
c 1997 IOP Publishing Ltd 345
346 R Ferreira and G Bastard

Contents

Page
1. Introduction 347
2. Electronic states in double quantum wells 352
2.1. Flat-band case 353
2.2. Biased double quantum wells 355
2.3. Valence states 359
2.4. Electron and hole states in biased ADQW: the tight-binding approach 362
3. Exciton states in double quantum wells 366
3.1. Introduction 366
3.2. Direct and crossed excitons in double quantum wells 368
4. Assisted tunnelling 370
5. Vertical transport in double quantum wells 376
5.1. Time evolution in perfect structures 377
5.2. Summary of experimental results 378
5.3. Time evolution in real structures: simple theoretical models 379
5.3.1. Damping of the probability amplitudes. 380
5.3.2. Density matrix formulation. 381
5.3.3. Beyond the previous models. 382
5.4. Assisted scatterings and interwell transport 383
5.4.1. Rate equations for the populations. 383
5.4.2. Role of the coherences. 384
6. Conclusion 385
Acknowledgments 386
References 386
Tunnelling and relaxation in semiconductor double quantum wells 347

1. Introduction

The wave-like nature of the electron implies the tunnel effect, i.e. that the electron has a non-
zero probability to be found in regions of space where the potential energy exceeds the total
energy and thus are classically forbidden. Consider, for instance, a one-dimensional square
potential barrier (height, V0 ; thickness, b) along the z-axis. The transmission coefficient
T (εz ) of an electron impinging on this barrier with an energy εz < V0 is zero for a classical
particle while one finds readily
T (εz ) = 1/[[1 + 1/(4x(1 − x))] sinh2 {b[(2mV0 /h̄2 )(1 − x)]1/2 }] (1.1)
with x = εz /V0 , in a quantum mechanical description (Schiff 1955). In this ideal situation,
the tunnelling current per electron is simply T (εz )vz , where vz is the electron velocity. The
electrons which have passed across the barrier have done so coherently. Their eigenstates are
exactly known and are linear superposition of plane waves (either propagating or evanescent)
everywhere in the structure. If we allow for the transverse degrees of freedom of an actual
electron, i.e. one which moves in a three-dimensional space, we find that the z and transverse
motions decouple and that the transverse quantum numbers are conserved.
There are other manifestations of tunnelling which do not involve current flow. For
instance, in the ammonia tetrahedral molecule NH3 the hydrogens form an equilateral
triangle. The N atom has two equilibrium positions symmetrically placed on each side
of the triangle. If classical mechanics were always applicable there would exist no possible
hopping for the N atom between the two equilibrium positions. Quantum mechanics allows
for such a hopping due to the tunnelling across the potential barrier which separates the two
minima. Compared to the degenerate unperturbed eigenstates with wavefunctions localized
around the two minima, the tunnel coupling mixes the unperturbed eigenstates and lifts the
degeneracy. The NH3 molecule provides a simple example of what will be later called
resonant tunnelling between discrete levels.
The tunnel effect in solids is often associated with carrier motion, and one loosely
speaks of carrier hopping between wells due to tunnelling, tunnel current in resonant
tunnelling diodes, etc. Although tunnelling is by no means restricted to semiconductors it
plays a decisive role in quantum semiconductor heterostructures and is best illustrated there
because of the sample quality (interface sharpness, control of doping, dislocations etc). By
quantum semiconductor heterostructures we mean a stack of several semiconductor layers
with thickness which are comparable to the effective de Broglie wavelength of the carrier
(with an effective mass m∗ ) and are such that the carrier motion needs to be described by
quantum mechanics and not by classical or semiclassical approaches. There are many ways
to design such quantum heterostructures and, as foreseen (and demonstrated) by Capasso
et al (1986), this ‘bandgap engineering’ is very effective in creating heterostructures which
fulfil a given device purpose.
In quantum heterostructures one very often uses (and we shall do so in the following) the
notion of a conduction (valence) band edge profile which is the position dependence of one
of the bulk conduction (valence) band edges along the growth axis of the heterostructure.
Take the example of a single barrier structure. It is composed of a stack of three layers, two
of which are semi-infinite (in practice very large compared to the de Broglie wavelength)
and identical, and separated by a thin slab (a few nanometres) of another material whose
conduction band edge is at higher energy than the two others. In reality, these materials are
solids and the potential energy is by no means piecewise constant. If, however, one neglects
the details which take place at the scale of the hosts’ unit cell (typically 0.5 nm), it is well
known that for a single layer of macroscopic size (a bulk material) the electron motion is
348 R Ferreira and G Bastard

Figure 1. Conduction band edge profiles of several heterostructures exhibiting pronounced


tunnelling effects: unbiased (a) and biased (b) superlattices; unbiased (c) and biased (d) resonant
tunnelling diodes; unbiased (e) and biased (f) double quantum wells.

very well approximated by that of a free particle with an effective mass m∗ corresponding
to the conduction band of interest. In heterostructures, a similar coarse graining exists. If
one is only interested in the electron dynamics on the scale of the slab thickness, typically
5–10 nm, and not on what happens at the nanometre size, the quantum motion of the particle
is in many instances satisfactorily described by an effective Hamiltonian where the kinetic
energy involves the carrier effective mass (a scalar for electrons, a 4 × 4 matrix for the
holes) and where the potential energy is the position-dependent conduction band edge for
the electron (the valence band edge for the hole).
The position-dependent conduction band edge of the quantum heterostructures discussed
in this review are shown in figure 1. They are the superlattices (unbiased or biased), the
resonant tunnel diode and the double quantum well (unbiased or biased).
The field of quantum semiconductor heterostructures started in the early 1970s after
Esaki and Tsu’s famous proposal (Esaki and Tsu 1970) that the stacking of a large number
of identical layers of two different semiconductors, a superlattice, should give rise to novel
electronics devices with a negative differential resistance (NDR). The physical origin of
this NDR is the resonant tunnelling between wells. In superlattices, the discrete levels
of the isolated wells hybridize together and give rise to quasicontinuums, the minibands,
whose dispersions εn (q), where q is the carrier wavevector along the growth axis, when
translated in terms of an effective mass, display a change of sign of the effective mass.
Hence the NDR follows because in the presence of an electric field q varies linearly with
time (q(t) = −eF t/h̄) and describes the miniband dispersion which is periodic upon q.
The carrier starting, say, with a positive effective mass ends up with a negative effective
mass. The material quality during this early period was not high enough to allow a full
experimental proof of this kind of resonant tunnelling. Of paramount importance for Esaki
Tunnelling and relaxation in semiconductor double quantum wells 349

and Tsu’s superlattice performance is the existence of a miniband spectrum. Equivalently,


if one analyses an Esaki and Tsu’s superlattice (which is necessarily of finite thickness)
in terms of carrier transmission across a finite stack of N wells and N + 1 barriers, it is
necessary that all the wells (barriers) are as identical and perfect as possible so as to allow
the carrier, once injected on one side of the stack, to be successfully transmitted on the
other side and that the carrier experiences as few collisions as possible.
It appears that we shall have to make a distinction between the tunnel coupling which
occurs between energy-aligned or energy-misaligned levels. They will be referred to as
resonant and non-resonant tunnelling, respectively. Although in both cases the tunnel
coupling induces a delocalization of the carrier wavefunction, the former is complete while
the latter is only partial. In addition, the tunnel coupling can never be treated perturbatively
in the resonant case while the non-resonant tunnelling, if the tunnel matrix element is small
enough, can be satisfactorily described by perturbative approaches.
The tunnel coupling between two (or M) wells leads to level anti-crossings. In the
NH3 molecule discussed earlier, the two degenerate levels in the absence of tunnelling
had wavefunctions which were localized on each side of the hydrogen triangle. Once
tunnelling is taken into account, the degeneracy is lifted and the corresponding eigenstates
are delocalized over the two minima (actually, in the simplest approach, the new eigenstates
are the normalized sum or difference of the two unperturbed states). Likewise, in tunnel
coupled double or multiple quantum wells, the carrier delocalization over the structure
results from the anti-crossings of two (or M) eigenstates of an unperturbed Hamiltonian
which are localized in different regions of the heterostructures (the bound states of each well
when considered as isolated in the double well example) and which, because of a suitable
coupling term, interact. This interaction mixes the unperturbed levels and the resulting
wavefunctions extend more than they did initially. If the interacting levels are close enough
in energy compared to the effect of the perturbation, the delocalization is almost complete,
while in the limit of large energy detuning between the uncoupled levels it can be very small.
The anti-crossing can be monitored by an externally controlled parameter, for example an
electric field, and, thus, one is able to control the amount of delocalization of the resulting
eigenstates.
A number of theoretical studies were undertaken in the early 1970s concerning the non-
resonant tunnelling in an attempt to account for the tunnelling current. Consider again Esaki
and Tsu’s superlattice structure and let us assume it has been biased (along the growth axis,
hereafter referred to as the z-axis). If one assumes for simplicity that the electric field is
constant inside the structure, the energy levels of consecutive wells are misaligned by eF d,
where d is the superlattice period. The corresponding wavefunctions localize inside the
superlattice and, in principle, there is no possible vertical transport (i.e. along the growth
axis) through a given miniband of a perfect heterostructure since localized states carry
no current and since the initial in-plane wavevector is conserved due to the translational
invariance in the layer plane. Translated in terms of carrier transmission across N wells and
barriers if N eF d exceeds the superlattice bandwidth the transmission becomes vanishingly
small.
Yet current flows because semiconductor heterostructures are never perfect and never
reducible to one-dimensional structures. There always exist a finite amount of residual
doping, interface roughness and/or lattice vibrations (phonons) to which the electrons are
coupled. If weak enough the imperfections can be treated perturbatively. Each of the
discrete energy levels for the z-motion carries a two-dimensional subband which arises from
the continuum of eigenvalues of the carrier in-plane motion. If the heterostructure is perfect,
these continuums have a constant density of states due to the two-dimensional free motion.
350 R Ferreira and G Bastard

Translational invariance in the layer plane implies that any electron transition conserves
the in-plane wavevector and it is therefore permissible to neglect the in-plane degrees of
freedom since their associated quantum numbers are conserved. An actual, i.e. disordered,
structure has lost this translational invariance and electron transitions can take place with
a change of the in-plane wavevector. Thus, such a heterostructure will, in addition to its
energy level scheme for the z-motion, be characterized by transition rates between the in-
plane continuums attached to each level. If disorder-induced transitions take place between
states of different in-plane wavevectors, a variation of the in-plane kinetic energy occurs.
Even if the scatterers are static the global energy conservation for an elastic scattering does
not imply a separate conservation of the in-plane and longitudinal kinetic energies. Thus,
by conversion of in-plane kinetic energy into kinetic and potential energies along the z-axis,
the carrier actually moves along the growth axis by hopping between wells. Since there
is a bias the transition probabilities to move along or against the field are different, which
leads to a net current. This hopping motion can also be termed assisted tunnelling and was
investigated by a number of authors (see, e.g., Kazarinov and Suris 1972, Tsu and Döhler
1975). It was shown, in the early eighties, to account for the measured electron and hole
mobilities in GaAs–Ga(Al)As superlattices (Calecki et al 1984).
In a way the superlattices are too complicated if one has the intention of investigating
the nature of the tunnelling (resonant versus non-resonant, assisted versus coherent) in
heterostructures. Two heterostructures fulfil the same purposes which are much simpler and
therefore much more reliably grown. These are the double barrier diodes and the double
quantum well structures.
A double barrier diode consists of a single quantum well with left and right barriers of
finite thickness (see figure 1(c)). The bound state of the well is actually a virtual bound state
which can decay in the continuum of the two semi-infinite wells on each side of the structure.
Conversely, if there is a way to align the populated levels on one side of the well with the
quasibound state, one expects a very large increase of the electron transmission across the
double barriers. The two semi-infinite contacts are heavily doped, say n type. By biasing
the diode (figure 1(d)) the Fermi level of one part (the emitter) is brought into coincidence
with the virtual bound level, which ensures a good transmission to the other side of the
structure (the collector). By further biasing the diode the bottom of the conduction band of
the emitter moves above the virtual level. This leads to a low transmission to the collector
and the current decreases. A double barrier diode is, therefore, a negative differential
resistance device. As in the previous discussion of superlattices, defects and phonons blur
or even change significantly the picture of the coherent tunnelling that has been adopted
for the diode. In particular, when one does not expect any current (in the ‘valley’ of the
diode current–voltage characteristics) the actual devices do show a non-negligible current.
The latter is associated with tunnelling events which do not conserve the electron in-plane
wavevector, i.e. one again deals with assisted tunnelling. The first experimental evidence
of negative differential resistance in a double barrier diode was obtained by Chang et al
(1974). For recent reviews on resonant tunnelling in semiconductors see Chang et al (1990).
A double quantum well is the shortest possible superlattice since it comprises only
two quantum wells separated by a thin barrier, the two outer cladding barriers being, in
principle, infinite (figure 1(e)). As in a double barrier diode, it is possible to study in a
double quantum well the resonant and non-resonant tunnelling between levels by an external
bias. The two structures differ, however, in a number of points. The double barrier diode
is a unipolar device where one injects from the outside (the emitter) electrons (or holes)
and collects outside the transmitted particles (open structure). Double quantum wells are
generally undoped and at equilibrium have no free carriers. They are closed structures
Tunnelling and relaxation in semiconductor double quantum wells 351

where the carriers (electrons and holes) may be photogenerated and their fate tracked inside
the structure by studying, for example, photoluminescence or absorption. The carriers may
have moved, however, inside the double well because the average carrier location may be
different in the initial state and in the final state.
Photogeneration implies that electron–hole pairs are created. The motions of the
electrons and holes are correlated by Coulombic interaction at their creation. Soon after,
this correlation is lost during the relaxation towards the lowest-lying excited state, the ones
from which light emission takes place, because of the numerous phase breaking events such
as phonon emission or absorption. When an electron and a hole, decorrelated by scattering,
reach their respective ground states they form again a correlated entity, an exciton, before
radiating. In quantum structures, however, under suitable excitation, i.e. not too far from
the light emitting levels and at low temperature, the phase coherence may be sufficiently
maintained between the creation and annihilation of the excitons. In particular, in double
quantum wells it is not at all clear that the persistence of a correlated motion for the
electron and the hole may not invalidate the description of tunnelling from one well to
the other in terms of single particle events. In a biased asymmetrical double well, the
two lowest electronic levels anti-cross at a given bias. One then expects that there should
exist experimental methods, such as recording the photoluminescence peaks position versus
bias, to measure this anti-crossing (since usually the hole anti-crossings are not found at the
same bias as the electron ones). These attempts are based on the assumption of uncorrelated
motions of electrons and holes, but we shall see later on that they are deceptive in that one
does not observe the free electron anti-crossing but that of the electron when it is bound
to the hole by the Coulombic field in the form of an exciton. The bias where the exciton
anti-crossing takes place is, therefore, not the one where the bare electron anti-crossing
would have been found.
Another important feature of the tunnel effect in double quantum wells is brought
about by population effects. We have so far reasoned in terms of a single electron, or
hole or exciton. In reality, even at very low optical injection, many carrier effects can
show up. Usually, the many-body effects in quantum heterostructures are strong photo-
injection features. They are due to the Pauli exclusion principle or the screening effect
and typically take place when the mean distance between excitons becomes comparable to
the exciton effective Bohr radius. This occurs at an exciton areal concentration in excess
of say 1011 cm−2 . Under such conditions, the description of the optical response of the
medium in terms of single excitons loses sense and, in fact, one observes experimentally
significant alterations of the optical spectra; for instance, the excitonic peaks become very
broad and the absorption edge is red shifted (bandgap renormalization). For a review on
strong photo-injection features in semiconductors see Haug and Koch (1990).
The effects which are peculiar to double wells are associated with the different main
localization of the electron and the hole in the initial (or final) state of an exciton anti-
crossing, that is to say are due to photo-created dipoles. The more polarized the excitons
along the z-axis, the larger their lifetime against radiative recombination. If the non-radiative
channels are not too effective, the steady-state regime corresponds to a population of spatially
separated electrons and holes small enough to still consider the excitons as independent. In
turn these dipoles give rise to an electrostatic potential which distorts the band edges, thus
perturbs the level scheme and may bring the system out of (or in) resonance (Sauer et al
1988). These induced dipole effects have also been observed in time-resolved experiments,
see, for example, Ferreira et al (1992).
In this review we shall concentrate on the behaviour of electrons, holes and excitons
along the growth axis. This does not imply that the carrier motion in the layer plane is
352 R Ferreira and G Bastard

not affected by the coupling of the eigenstates along the growth axis. In particular, it
has been shown in the last few years that new fractional Quantum Hall states occur in
double-layer two-dimensional electron systems (Eisenstein et al 1992a) and that the strong
Coulomb correlations between electrons located in parallel two-dimensional electron gases
in double quantum wells profoundly affect the current voltage characteristics for the current
flow along the growth axis (Eisenstein et al 1992b). Both experimental findings rely on
the application of a very strong magnetic field parallel to the growth axis. Such a magnetic
field discretizes the in-plane motion and gives rise to one-electron densities of states which
consist of narrow Landau subbands. As often with narrow bands, the electron gas becomes
very sensitive to Coulomb correlations. Due to a lack of space and to the fact that these
novel physical effects are only marginally related to the central theme of this review, the
tunnel effect along the growth axis, we shall not discuss them any further (for a recent
review see, e.g., Manoharan et al (1996)).
As this introduction aimed at showing, the conceptually simple tunnel effect in a double
quantum well is in practice not as simple as one might have anticipated. The remainder
of this review is devoted to a more detailed presentation of the various effects which have
been alluded to in this introduction and which are directly related to the electron, hole and
exciton motion along the growth axis. Since this topic has reached gigantic proportions in
the literature we have had to focus tightly on the central theme to keep this review article
within reasonable limits. This implies that a number of subjects will not be discussed
in the present review. Broadly speaking, most of the phenomena which are related to
many-particle effects will not be covered. This includes double quantum wells which have
been purposedly doped and, therefore, display band bending, Pauli exclusion effects and
bandgap renormalization. This also includes highly-excited double quantum wells which
contain carriers and excitons in such a number that it is illusory to attempt to describe their
motions without an allowance of again the Pauli exclusion effects and possible band bending
if the electrons and holes are spatially separated. In short, our objective is to describe the
motion of a single electron or hole or exciton in a double quantum well allowing for the
imperfections which are necessarily present. In section 2 we shall discuss the electron and
hole energy levels in an ideal double well. In section 3 we shall deal with exciton effects,
while section 4 will detail the various mechanisms which gives rise to the assisted tunnelling
for electrons, holes and excitons in a simple model where the scattering agents are taken
into account in the lowest order of perturbation theory over the eigenstates of the perfect
double well. Section 5 will be devoted to a discussion of various time-dependent aspects
of the tunnel effect in double quantum wells and we shall see that the simple model of
section 4 needs to be corrected, in particular, when the broadening becomes comparable
to the tunnel coupling. Finally, we shall summarize the (provisional) conclusions about
tunnelling that can be drawn from the studies of double quantum wells.

2. Electronic states in double quantum wells

The electronic levels in heterostructures have been the subject of numerous theoretical
investigations ranging from first principles calculations to effective mass approaches (for
a general review see, for instance, Smith and Mailhiot (1990)). On the one hand, one
searches in fully microscopic calculations for the complete carrier wavefunctions which
are thus known on every length scale, the hosts’ unit cell (0.5 nm) and the slab thickness
(a few nanometres). These methods include all electrons calculations, pseudo-potentials,
tight-binding and, at least in principle, do not require the knowledge of the band offsets.
The latter are by-products of these elaborate computational schemes. On the other hand, the
Tunnelling and relaxation in semiconductor double quantum wells 353

envelope function method focuses its attention on the modulation of the carrier wavefunction
which is due to the heterostructure itself, i.e. the slab thickness (Bastard 1981a, b, 1988,
White and Sham 1981, Altarelli 1983, see also Ando et al 1982). The rapid variations due
to the natural periodicity of the hosts are ignored, or rather are taken implicitly into account
by using effective masses, or more generally approximate hosts’ dispersion relations. The
band offsets should be inputs to envelope function calculations. Clearly, the envelope
function approaches provide a less detailed knowledge than the first principle ones. They
are, however, much more manageable, in particular they handle naturally the effect of an
external potential, and much easier to implement on the numerical side.
The hosts’ dispersions should be known a priori. In the simpler approach that we
shall follow in this review, one assumes a parabolic description of the hosts’ bands. For a
twofold (or spin) degenerate band edge such as the 06 conduction band edge of III–V or
II–VI semiconductors, the effective Hamiltonian is the scalar p2 /2m∗ . The valence band
of the cubic semiconductors is invariably fourfold degenerate at the centre of the Brillouin
zone (08 point). Long ago, the effective Hamiltonian of this degenerate band edge was
derived by Luttinger and Kohn (1955). Due to the fourfold degeneracy, this Hamiltonian
can be expressed in terms of k and of a 32 spin J (J = 3/2):
H{08 (k)} = αk 2 1 + β(kx2 Jx2 + CP) + γ [kx ky {Jx , Jy } + CP] (2.1)
where CP denotes cyclic permutations, { } is the anticommutator, α, β and γ are three
effective mass constants to be determined by experiment (such as cyclotron resonance) and
1 is the 4 × 4 unity matrix. When k k J , the eigenstates are also eigenfunctions of Jz with
eigenvalue mJ = ± 32 for the heavy-hole branch and mJ = ± 12 for the light-hole branch. In
heterostructures, we shall take z parallel to the growth direction. For a general k direction
the eigenfunctions of the valence Hamiltonian are 4 × 1 spinors and do not exhibit a definite
± 32 or ± 12 character. This mJ mixing effect at non-vanishing k is a general feature of the
valence band states. As we shall see, it plays a significant role in heterostructures.
Specific to the heterostructures is the role played by the evanescent (non-propagating)
states, forbidden in infinite bulk crystals on account of their exponential variation along one
or several directions. To build up the heterostructure states we should, strictly speaking,
consider all the possible (propagative and evanescent) solutions within each layer and
match them properly at the interfaces. This procedure requires one to include a number of
evanescent states which are more localized near the interfaces when derived from the more
remote bands. These are the remote bands which ensure the total wavefunction and current
conservation at the boundaries between two layers. Being interested in spatial modulations
which are slow, it is justified to discard these rapidly varying evanescent states and retain
only those which display smaller spatial variations. In our parabolic treatment of the hosts’
dispersions, we shall in fact only retain the evanescent states in the barrier layer which
belong to the same bulk extremum as that in the well layer. This simplification is well
justified in a large number of practical cases. It will fail whenever two different extrema
in the hosts’ Brillouin zone have to be simultaneously considered in the building of the
heterostructure states. A notorious example is provided by the GaAs/AlAs pair which in
the limit of narrow (2–3 nm) GaAs layers involve a mixing of 06 -related and Xz -related
states. In the following such complicated situations will not be discussed.

2.1. Flat-band case


In the absence of band bending (and for the kx = ky = 0 valence states) the problem is
purely one-dimensional and scalar, and involves a piecewise constant potential energy. The
354 R Ferreira and G Bastard

latter can be set equal to zero in the well material and to Ve (Vh ), the conduction (valence)
band offset in the barrier material. For any kind of particle (electron, heavy or light hole)
and for a given energy the bulk solutions inside any layer are given by linear combinations
of either harmonic (sin and cos) or exponential (sinh and cosh) functions. Inside each layer
the envelope solution χ (z) is built up as a linear combination of these bulk solutions. The
eigenvalues are obtained by matching both the envelope function χ(z) and current density
[1/m∗ (z)] dχ (z)/dz at the interfaces and by imposing appropriate asymptotic behaviour
at infinity (for instance, the envelope of a bound state should decrease exponentially as
|z| → ∞). In a word, for a fixed energy the heterostructure state is formed by the bulk
states inside the different layers and at this same energy. Correspondingly, the mass entering
in the effective mass equation is, for any layer, the bulk mass at the same energy.
The double quantum well system is one of the simplest systems exhibiting tunnel
coupling (which in the superlattices leads to the formation of minibands). These systems
are composed of two ‘well’ layers separated by an intermediate ‘barrier’ layer and clad by
semi-infinite ‘barrier’ layers (see figure 1(e)). In the following we consider two different
methods to evaluate their electronic states. The first can be applied for any set of layer
thicknesses. The second is a tight-binding approximation limited to sufficiently thick
intermediate barriers. Numerical calculations are required in the first approach, whereas the
second leads to simple analytical results in terms of a small number of relevant parameters
(see section 2.4). Finally, except where explicitly mentioned, the results presented were
obtained for systems based on GaAs (well)–Ga0.7 Al0.3 As (barrier) materials. Also, in the
following we denote as L1 –B–L2 the thicknesses of the three intermediate layers.
Let us start with the symmetric double quantum well (SDQW) case. We show in figure 2
the variations of the energies En = En (B) of the bound conduction levels as a function
of the intermediate barrier thickness B for L1 = L2 = 40 Å (2(a)) and 80 Å (2(b)) wells.
We note that: (i) at B = 0 the structure reduces to a single quantum well of thickness
2L1 ; (ii) with increasing B the pairs of levels E1 and E2 and E3 and E4 are less and less
separated, for B > 50 Å the splitting between the two lowest-lying levels E2 (B)–E1 (B) is
very small (6 1 meV); (iii) when B → ∞ the tunnel coupling through the barrier decreases
to zero and the double quantum well system becomes a system with two isolated identical
single quantum wells. We show in figure 3 the envelope functions of the first two levels
in figure 2(a): B = 0 (3(a)), 20 Å (3(b)), 60 Å (3(c)) and for a 40 Å wide single well
(3(d)). All the functions are found to be delocalized over the whole structure and to have
a well defined parity: odd (even) envelope functions vanish (are non-zero) at the centre of
the structure. This delocalization persists for an arbitrarily thick intermediate barrier and
suggests that even the smallest tunnel coupling is not amenable to perturbative approaches
in the symmetrical situation. As will be shown later, this somewhat paradoxical situation
is linked to the assumed exact identity of the two wells.
In contrast to symmetric structures, the asymmetric double quantum wells (ADQW) have
no inversion plane. Then the corresponding envelopes are generally unequally distributed
between the two wells and a preferential localization appears. We show in figure 4(a) the
variations of the energies En = En (B) of the bound conduction levels as a function of B
for L1 = 100 Å and L2 = 50 Å. We note that the energies En (B) are almost B-independent
for B > 30 Å. The tight-binding method is reliable in this regime of weak dependence
of the eigenvalues with the thickness of the intermediate barrier. We show in figure 4(b)
the envelopes for the first three levels of the 150 Å single well (B = 0). Increasing the
barrier width (10 Å in figure 4(c) and 50 Å in figure 4(d)) leads to an increasing spatial
localization of the ADQW eigenstates inside one of the two wells. This spatial localization
with increasing B is faster the larger the energy difference between the levels of the two
Tunnelling and relaxation in semiconductor double quantum wells 355

Figure 2. B dependence of the conduction eigenstates of a 40 Å–B–40 Å (a) and of a 80 Å–


B–80 Å (b) double quantum well. B is the intermediate barrier thickness.

isolated wells forming the ADQW. In an asymmetrical double well the localization of the
lowest-lying state always takes place in the thick barrier limit which contrasts with the
symmetrical situation. If one excited level of, say, the left well is in resonance with a level
of the right well, the eigenfunctions for these two levels no longer localize spatially when
B increases but remain delocalized over the two wells, as in the SDQW case.

2.2. Biased double quantum wells

As shown in the previous section for the asymmetric DQW case, the tunnel coupling is
strongly dependent on the relative energy positions of the interacting levels. The tunnel
coupling can thus be externally monitored by the application of an external perturbation.
The application of an electric field along the growth axis, as discussed in the following,
is the most widely used method to probe the tunnel coupling effects. The application of
a strain (see, for instance, Lefebvre et al (1991)) and of a magnetic field (in II–VI based
structures; see, for instance, Lawrence et al (1993)) have also been considered.
We shall mostly discuss the tunnel effect in nominally undoped structures where the
band bending arises from charges which are located outside the active part (the double well).
In this way, the band bending reduces to that due to a nearly constant electric field F . Other
situations, not discussed in the present review, involve double quantum wells which have
been purposely doped. By selectively doping the barriers adjacent to the double quantum
356 R Ferreira and G Bastard

Figure 3. Envelope probability density (χ 2 (z)) against position of the two lowest-lying
conduction states of a 40 Å–B–40 Å double quantum well. B = 0 (a), 20 Å (b), 60 Å (c)
and ∞ (a single 40 Å thick quantum well) (d).

well, carriers are transferred from the dopants to the double well structure, thus creating a
two-dimensional electron gas which, depending upon the doping and possible external bias
through a back gate, can be mostly localized in one well or delocalized over the double well
(see, e.g., Ensslin et al (1989) for a study of these biased modulation doped double quantum
wells and the review by Manoharan et al (1996) for their use in the study of correlation
effects in bi-layers).
For a nominally undoped double quantum well we take the origin of the electrostatic
potential at the centre of the intermediate barrier. The electrostatic potential is eF z and,
therefore, lifts up the right-hand side well with respect to the left-hand side well if F > 0
(see figure 1(f)). We discuss in the following the effect of the external field on the bound
levels of the unbiased structure. First, these levels become unstable in the presence of
the field and give rise to a series of resonances since the electrostatic potential becomes
arbitrarily large and negative when z → −∞. However, it turns out that the field-induced
broadenings of the first few electronic states for most of the actual structures are very small
for field intensities currently reached experimentally (6 100 kV cm−1 ), a situation very
much like the one met in biased single wells (see, for instance, Bastard et al 1991). A first
approximation of these resonances is thus obtained by neglecting their finite energy width
and by considering them as true bound states. Hence, to calculate their energy positions,
we can diagonalize exactly the electrostatic potential inside the subspace spanned by the
bound levels at zero field and obtain useful information regarding the field dependence of
the spatial localization of the eigenstates in the biased double well.
We present in figure 5(a) the field variation of the energies of a biased SDQW
(L1 = B = L2 = 40 Å). Let 1E = E2 (0) − E1 (0) be the zero field splitting due to
the tunnel coupling, d = L1 + B the interwell separation and E0 ≈ (E2 (0) + E1 (0))/2
the ground energy of a 40 Å thick isolated unbiased well. For small fields, such that
eF d < 1E, the energies vary quadratically with the field, a feature reminiscent of the
so-called quantum confined Stark effect (mostly investigated in biased single quantum wells
Tunnelling and relaxation in semiconductor double quantum wells 357

Figure 4. Conduction eigenstates and envelope functions χ (z) (in arbitrary units) of a 100 Å–
B–50 Å asymmetrical double quantum well. (a) energy levels against B. Wavefunctions χ1 (z)
(full curve), χ2 (z) (dashed curve) and χ3 (z) (dashed dotted curve) for B = 0 (b), 10 Å (c) and
50 Å (d).

(Miller et al 1985, Schmitt–Rink et al 1989)). At large fields (eF d  1E) the energies
converge asymptotically to the linear relations E± = E0 ± eF d, which is independent of the
interwell tunnel coupling. The critical field Fc = 1E/ed gives the transition region between
the quadratic and linear regimes (Fc ≈ 15 kV cm−1 in figure 5). We also show in figure 5
the corresponding envelope functions at various fields. The eigenfunctions are delocalized
over the two wells at zero bias, start localizing around a given well with increasing field
and finally strongly localize around this well at higher fields. This preferential localization
has the same physical origin as the one observed in asymmetrical structures: the external
field breaks the initial symmetry by energetically favouring one of the two wells. If the
mismatch due to this asymmetry becomes stronger than the tendency to delocalization,
which is provided by the tunnel coupling between wells, the eigenstates become more
and more polarized around one of the wells. If in the former example one keeps the
electric field constant but varies the intermediate barrier thickness, a similar tendency to
localization results: the tunnel coupling weakens and ultimately becomes a perturbation
358 R Ferreira and G Bastard

Figure 5. The electric field effect on a symmetrical double quantum well 40 Å–40 Å–40 Å.
Conduction states. (a) Energy levels against field. Envelope probability densities against position
for F (kV cm−1 ) = 0 (b), 15 (c) and 60 (d).

compared to the isolated well situation. It becomes possible to express the modification of
the eigenenergies by non-degenerate perturbation calculations, an impossible procedure in
the resonant tunnelling situation of the unbiased SDQW where the two interacting levels
are energy degenerate. We show in the figure 6 the F -dependence of the energies of an
L1 = 100 Å, B = 20 Å SDQW. In this case each isolated well binds two levels and the
SDQW spectrum presents field-induced anti-crossings when the second level of one well
becomes lined up with the ground level of the other by the field. The envelopes for the
second and third levels are delocalized over the two wells at zero bias, start localizing
around one well with increasing field, become delocalized again at the energy anti-crossing
region and finally localize in the other well for larger fields.
We show in figure 7 the field evolution of the energy levels of an L1 = 100 Å, B = 50 Å,
L2 = 47 Å ADQW. In this asymmetrical structure we have En (F ) 6= En (−F ) and the
energy anti-crossings are no longer symmetrically placed around F = 0. In addition, the
field-induced anti-crossings for the conduction and valence levels generally do not appear
Tunnelling and relaxation in semiconductor double quantum wells 359

Figure 6. The electric field dependence of the conduction states in a 100 Å–20 Å–100 Å double
quantum well.

Figure 7. The electric field dependence of the conduction states in a 100 Å–50 Å–47 Å
asymmetrical double quantum well.

at the same field, which helps us in the following to distinguish between the effects related
to the electrons and to the holes when considering the interwell transfer of optical excitation
observed experimentally in such systems.

2.3. Valence states

The valence levels and in-plane dispersions of single and multiple quantum wells have been
extensively discussed in the literature (see, for instance, Altarelli 1983). In the |J = 32 , Jz i
360 R Ferreira and G Bastard

Figure 8. Valence band in-plane dispersion relations of a single 100 Å thick GaAs–Ga0.7 Al0.3 As
quantum well. Dashed curves, diagonal approximation; full curves, full Luttinger matrix.

basis the kinetic energy term reads


 
Hhh c b 0

 c Hlh 0 −b 
H{08 (k)} =  ∗  (2.2a)
b 0 Hlh c
∗ ∗
0 −b c Hhh
Hhh = −(γ1 + γ2 )h̄ k⊥ /2m0 − (γ1 − 2γ2 )pz2 /2m0
2 2

Hlh = −(γ1 − γ2 )h̄2 k⊥


2
/2m0 − (γ1 + 2γ2 )pz2 /2m0

b = [h̄γ3 3/m0 ](kx − iky )kz

c = [h̄2 3/2m0 ](γ2 (kx2 − ky2 ) − 2iγ3 kx ky ) (2.2b)
where k⊥ = (kx , ky ) and γ1 , γ2 , γ3 are the so-called Luttinger parameters of the valence
band (Luttinger 1956). They are usually different in different materials. To account for their
possible position dependence any combination involving these terms should be properly
anti-symmetrized. In addition to the kinetic energy, one should include a term which is
diagonal in the mJ index and represents the effect of the valence band offset (a piecewise
constant term Vhs (z)) and of any external potential Vext (z) (e.g. eF z for a linearly biased
structure).
In order to discuss more easily the valence states of single and double quantum wells it
is better to consider first the results obtained within a simple diagonal approximation where
the non-diagonal terms b and c of the Luttinger matrix equation (2.2) are neglected and
subsequently consider the role played by these non-diagonal terms. In the following we
consider separately (i) the effect of mass reversal, (ii) the coupling between the confined
motion along the growth axis and the free in-plane motion for carriers in the heterostructures
and (iii) the role of the spin–orbit terms on the valence levels.
(i) If b and c are zero, the heterostructure eigenstates are also eigenstates of Jz at
any k⊥ . The heavy-hole states (mJ = ± 32 ) have indeed a heavy mass along the z-axis
Tunnelling and relaxation in semiconductor double quantum wells 361

(= m0 /(γ1 − 2γ2 )) but their mass for the in-plane motion is light (= m0 /(γ1 + γ2 )). The
converse is true for the light-hole levels: their mass along z is light (= m0 /(γ1 + 2γ2 )) and
their in-plane mass is heavy (= m0 /(γ1 − γ2 )). The diagonal approximation to the valence
Hamiltonian leads to crossings between the heavy-hole-related and light-hole-related levels
at finite in-plane wavevectors. In reality, these crossings are replaced by anti-crossings once
the off-diagonal terms of the Luttinger matrix are taken into account. The result is generally
a complicated series of levels at a given k⊥ = (kx , ky ) or a series of strongly non-parabolic
dispersions for the in-plane motion (see, e.g., Altarelli 1983). The ‘camel-back’ observed
for certain branches illustrates these considerations (see below).
(ii) The non-separability of the valence Hamiltonian into two independent parts
corresponding to the motions parallel and perpendicular to the growth axis indicate a strong
correlation between the confined and in-plane motions. This point is important in double
and multiple quantum wells since holes propagating freely in the layer plane are unavoidably
scattered in the presence of defects, which changes their in-plane motion. Since in-plane and
confined motions are coupled, these scatterings can influence the transfer from a given well
to a neighbouring one (assisted ‘vertical’ transfer). This kind of mechanism is discussed in
section 4.
(iii) The non-diagonal terms of the Luttinger matrix indicate a strong coupling between
the mJ character and orbital motions for the holes in these quasi-two-dimensional systems.
In the parabolic approximation only a single component of the 4 × 1 spinor is non-zero.
The non-diagonal terms of the Luttinger matrix mix parabolic states with different orbital
motions and different mJ values. This kind of effect is important when considering the
‘spin’ relaxation for holes photocreated in real structures since the hole scattering by defects
changes their in-plane motion which, therefore, leads to spin relaxation (Twardowski and
Hermann 1987, Ferreira and Bastard 1989b).
These three effects are illustrated in figure 8 where we show the in-plane dispersion
for an unbiased single quantum well of L = 100 Å. The dashed curves give the parabolic
levels. The mass reversal effect leads to a series of crossings between the heavy and
light dispersions. The full curves give the actual dispersions, after diagonalizing the
Luttinger matrix. We see that the previous crossings are replaced by anti-crossings and
that the dispersions are far from following a parabolic law. In particular, the first light-hole
dispersion (or equivalently the in-plane dispersion associated to the first k⊥ = 0 light state)
presents a ‘camel-back’ shape: it has an electron-like character for small wavevectors. We
also obtain that each energy level is twofold degenerate at a given k⊥ . In fact we can show
on very general grounds that any even potential Vtot (z) = Vhs (z) + Vext (z) = Vtot (−z) in the
valence Hamiltonian leads to a twofold degenerate heterostructure spectrum. This ‘parity’
degeneracy at a given k⊥ is lifted in asymmetrical structures (like ADQWs and unbiased
single quantum wells with different left and right barriers) and in biased structures.
Figure 9(a) shows the in-plane dispersions for two independent quantum wells of
L = 100 Å (full curves) and L = 60 Å (dashed curves). Figure 9(b) shows the in-plane
dispersion for a double quantum well 100 Å–40 Å–60 Å. We see that the interwell crossings
are replaced by energy anti-crossings and that the twofold degeneracy is lifted. Let us
consider the ninth dispersion in figure 9(b), i.e. the ninth eigenvalue of the hole Hamiltonian
when k⊥ varies. Figure 9(c) presents the squared amplitude of the total envelope (which is
equal to the sum of the squared amplitudes of the four spinor components) for three different
values of the in-plane wavevector. At k⊥ = 0 (full curves), the ninth state corresponds to
a light state strongly confined in the thin well. At k⊥ = 0.016 Å−1 (dashed curve) the
envelope is delocalized over the two wells, as due to the tunnel coupling with two heavy
dispersions mainly related to the wide well. Finally at k⊥ = 0.025 Å−1 (dotted curve) the
362 R Ferreira and G Bastard

Figure 9. (a) Valence band in-plane dispersion relations of a GaAs–Ga0.7 Al0.3 As single quantum
well with L = 100 Å (full curves) and L = 60 Å (dashed curves). (b) Same as before but for a
100 Å–40 Å–60 Å double quantum well. (c) Envelope probability densities against position for
three different in-plane wavevectors of the ninth state of the double well, k⊥ = 0, 0.016 Å−1
and 0.025 Å−1 .

envelope is localized mainly in the wide well and has become predominantly heavy-like
(see Ferreira 1992, Goldoni and Fasolino 1992).
We show in figure 10 the field variations of the energies for a 50 Å–40 Å–50 Å
SDQW at two different wavevectors: k⊥ = 0 (a) and 0.025 Å−1 (b). We note that the
twofold degeneracy is lifted when F and k⊥ 6= 0. Note also that for a vanishing in-plane
wavevector the electric field induces crossings between the heavy and light levels. This is
due to the fact that the non-diagonal terms of the Luttinger matrix vanish at k⊥ = 0. For
finite k⊥ values, on the other hand, these field-induced crossings are replaced by energy
anti-crossings.

2.4. Electron and hole states in biased ADQW: the tight-binding approach

In this subsection we consider the effects of the tunnel coupling on the first few levels of a
biased ADQW in the framework of a tight-binding model. This approach is not as accurate
as the numerical methods discussed above, but is physically more transparent. In the general
case, the tunnel coupling depends on the in-plane motion. However, in aiming to discuss
here the main features of the tunnel coupling, we shall consider in the following only the
k⊥ = 0 states (or parabolic in-plane dispersions). The double well is initially considered
as two non-interacting wells. We label by eN(p) the energy of the N th level of the isolated
quantum well ‘p’ (p = 1 (width L1 ) or 2 (width L2 )), see figure 11(a)).
Tunnelling and relaxation in semiconductor double quantum wells 363

Figure 10. The electric field dependence of the valence eigenstates in a 50 Å–40 Å–50 Å
GaAs–Ga0.7 Al0.3 As double quantum well. (a) k⊥ = 0; (b) k⊥ = 2.5 × 106 cm−1 .

We are mainly interested in the case where level ‘n’ in L1 (energy en(1) ) interacts mainly
with level ‘m’ in L2 (energy (em(2) )), all the other levels being far apart (|en(1) − em(2) | 
|en(1) − en0 (1) | and |em(2) − em0 (2) | with n0 6= n and m0 6= m). In order to focus on the effect
of the interwell coupling, we retain in the following only the levels en(1) and em(2) . (The
effect of distant states can be accounted for perturbatively, which permits to recover, for
instance, the intrawell quantum confined Stark effect.) Thus, the electric field only shifts
rigidly the isolated well energies en(1) and em(2) :
eN (p) (F ) = eN (p) (F = 0) + eF zp (2.3)
where z1 = −(L1 + B)/2 and z2 = (L2 + B)/2. Then, for positive F , en(1) (F )(em(2) (F ))
decreases (increases) with increasing electric field (note that F < 0 in figure 11). The
corresponding envelopes are given by
ϕN (p) (z) = ϕloc(N(p)) (z − zp ) (2.4)
where ϕloc(N (p)) (z), a solution of the unbiased isolated quantum well problem, is strongly
localized in the pth quantum well region. Eigenstates of the whole structure result from
the tunnel coupling of the√two localized (in different wells) states. The strength of this
coupling is given by λc = |λ1 λ2 |, where λ1 and λ2 are the transfer integrals:
λp = −Vc hϕloc(n(1)) (z − z1 )|ϕloc(m(2)) (z − z2 )ipth well . (2.5)
The integration is to be performed over the pth well region and Vc is the conduction band
offset. |λ1 | and |λ2 | vary exponentially with the barrier width and λc → 0 when B → ∞.
The tight-binding energies (E1 = E− and E2 = E+ in figure 11) are
E± = e0 ± [(1e/2)2 + λ2c ]1/2 (2.6)
364 R Ferreira and G Bastard

Figure 11. Schematics of the biased double quantum well. (a) Conduction band edge profile
and energy levels. (b) Field dependence of the two tunnel interacting states. Full lines, without
tunnel effect; dashed curves, with tunnel effect.

where 1e = |en(1) (F ) − em(2) (F )| and e0 = [en(1) (F ) + em(2) (F )]/2. The envelopes have
the form

ϕ±,n,m (z) = A±,n,m ϕn(1) (z) + B±,n,m ϕm(2) (z) (2.7)

where the coefficients A and B account for the interwell mixing of the two decoupled en(1)
and em(2) states.
For a given structure the field-dependent parameter is the energy ‘detuning’ 1e =
1e(F ). Whenever 1e(F )  2λc , the ADQW energies and wavefunctions are given
essentially by the two decoupled solutions (|A|, |B| ≈ 0 or 1). On the other hand, if
1e(F ) 6 2λc the tunnel interaction effectively mixes the two localized states and each√of
the two ADQW eigenstates spreads over the whole ADQW structure (|A| ≈ |B| ≈ 1/ 2
if 1e(F ) ≈ 0). The inequality 1e(F ) 6 2λc can be rewritten as |1F | = |F − F0 | 6
2λc /ed ≡ |1Fc |, where F0 is the resonance field defined by en(1) (F0 ) = em(2) (F0 ). For field
values in this interval the ADQW energies E± deviate from the isolated well values en(1) and
em(2) and, in particular, the crossing en(1) = em(2) at F = F0 is replaced by an anti-crossing
and an energy splitting 1E(F0 ) = E+ (F0 ) − E− (F0 ) = 2λc (figure 11(b)). A variation
of the external electric field by |1Fc | around the resonant value F0 enables a change from
a situation of two quasi-decoupled wells to a situation of two strongly interacting wells.
Tunnelling and relaxation in semiconductor double quantum wells 365

Note that for thick barriers (B > 100 Å) or very thin barriers (B 6 20 Å), |1Fc | is either
extremely small or very large and no experimental resonant features are expected in any
physical property depending on the spatial extention of the ADQW states when varying the
external field. In the thin barrier case the resonances would be very smooth; in the wide
barrier case the broadening of the lines due to the inhomogeneity of the structure washes
out the resonance. To be specific, let us take the 78 Å–55 Å–35 Å structure and consider
the two ground levels (n = m = 1). We find 1E(F0 ) ≈ 3 meV, 2|1Fc | ≈ 6 kV cm−1 and
F0 ≈ −54 kV cm−1 .
For the valence states, similarly, we have

hN (p) (F ) = −|hN (p) (F = 0)| + eF zp (2.8)

and for the same 78 Å–55 Å–35 Å structure the energy crossing h1(1) = h1(2) is located in
the positive field region. The decoupled hole envelopes χloc(N (p)) (z) are strongly localized
in one of the wells. For the ground hole states λh ≈ 0, and a good approximation is to
consider the decoupled envelopes as solutions of the ADQW problem.
Optical studies of double quantum wells are sensitive to the variations with the external
voltage of both the relative position of the various energy levels and the spatial localization
of the corresponding envelopes (Chen et al 1987, Islam et al 1987, Andrews et al 1988,
Tada et al 1988, Ferreira et al 1990, Golub et al 1990, Fox et al 1991). In the following
we consider briefly the interband optical transitions (absorption and luminescence) in these
ADQW structures. In the simple tight-binding approach four band-to-band transitions are
obtained by combining the two hole levels and the two electron levels.
Far from the resonance region (|F − F0 |  |1Fc |) one has both direct and crossed
edges. Direct (crossed) transitions occur between a hole and an electron located in the same
well (different wells), with energies roughly given by ≈ en(1) − hm(1) and ≈ en0 (2) − hm0 (2)
(respectively, by ≈ en(1) − hm0 (2) and ≈ en0 (2) − hm(1) ). We would like to stress here two
features: (i) the fundamental transition associated with each hole level changes from a
spatially direct one to a spatially indirect one (or vice versa) as we go from one side of
F0 (the resonance field for e1(1) and e1(2) ) to the other. In particular, when F0 < F < 0
the first energy edge E1 − H1 corresponds to the intrawide well transition e1(1) − h1(1) ,
whereas for F < F0 it corresponds to the interwell e1(2) − h1(1) transition. Then, for large
negative electric fields the ground electronic transition corresponds to a crossed one in
real space; (ii) the second remark concerns the interband optical absorption in an ADQW
under off-resonance condition. First, it is worth remarking that the absorption is essentially
an intrawell process (vertical transition in real space) since the absorption probability is
proportional to the squared overlap of the electron and hole envelopes. The latter is very
small for carriers in different wells (interwell overlaps are roughly smaller by a factor
λc /Vc  1 with respect to the intrawell ones). Since, in addition, the intrawell edges are
well energy separated, the spatial origin of the optical processes can be tracked back far
from resonances, which is very convenient from a practical point of view.
On the other hand, in the anti-crossing region ‘inter’-well and ‘intra’-well contributions
to the optical properties compete. In fact, due to the increasing spread of the eigenstates over
the whole structure, the two terms ‘inter’ and ‘intra’ become meaningless as F approaches
F0 . In addition, when the energy mismatch between the two a ADQW levels is of the order
of a few millielectronvolts, the experimental optical resolution of the two contributions
becomes difficult because of the broadening of the lines (see, e.g., Ferreira et al 1992).
366 R Ferreira and G Bastard

3. Exciton states in double quantum wells

3.1. Introduction
The study of the optical response of semiconductors (bulk or heterostructures) implies that
one deals with the excited states of the crystal. If the electrons were not interacting these
excited states would be electron–hole pairs characterized by well defined wavevectors for
their centre-of-mass and relative motion, on account of the translation invariance of the
ideal crystal. The electrons do interact, however, and the lowest-lying excited states of
the crystal, the excitons, are no longer characterized by free relative motion. Instead, this
relative motion displays both a discrete and a continuous spectra. They roughly correspond
to the bound and dissociated states of the hydrogen atom. The advances of laser fabrication
enables one to study the dynamics of an excited semiconductor on a subpicosecond time
scale. In particular, the return of the excited crystal to equilibrium by emission of photons is
of considerable current interest (for a recent review on hot carrier effects in semiconductors
see Shah (1992)). One may analyse these time-resolved optical experiments by means of rate
equations for the populations of electrons, holes and excitons. This approach will be used in
this review. We consider it to be satisfactory when the semiconductor is weakly excited, i.e.
when there are few excitons. High excitation effects, in particular lasing action, are better
handled by a more microscopic approach, using the semiconductor Bloch equations (for a
recent review see Haug and Koch (1990)). Under such a scheme, the equations of motion for
the electron and hole populations and the interband polarization are solved simultaneously.
This allows one to include many-body effects such as phase space filling, screening, etc,
which necessarily show up when the material is strongly excited. The semiconductor Bloch
equations approach has been used by Binder et al (1994) to analyse the coherent response of
perfect double quantum wells. One difficulty with this approach is that it does not lend itself
easily to the inclusion of scattering of carriers by defects and phonons. Since we believe
that these scattering-induced tunnelling events in double quantum wells are of paramount
importance and since this review does not cover highly-excited double quantum wells, we
shall not discuss the Bloch equation approach any further.
Coulombic states in bulk semiconductors have been extensively discussed in the past
and used as keys to the understanding of the band structures as well as a test for theoretical
models (Elliott 1963, Knox 1963). Excitons are not easy to analyse in detail in zinc-blende-
like semiconductors due to the complex valence band of these compounds (Baldereschi
and Lipari 1971). Apart from their energy levels, one is often interested in knowing the
exciton dynamics, for example the interactions with phonons and impurities. A convenient
approximation is often performed which consists of treating separately the heavy and light
states within the parabolic effective mass approximation. The exciton problem then reduces
to that of an electron–hole pair bound by the three-dimensional screened Coulombic potential
VCoul (re −rh ). This leads to hydrogen-like states for the reduced motion and to a free motion
for the centre of mass. We recall briefly that:
(i) an infinity of ‘bound’ levels populate the energy gap around the ground interband
transition
ε(υ, K) = −Ry ∗ /n2υ + h̄2 K 2 /2M + εg (3.1)
where υ labels the different symmetries (1S, 2S, 2P, . . . ; nυ = integer), Ry ∗ is the bulk
(three-dimensional) effective Rydberg and the kinetic energy is due to the free centre-of-
mass propagation;
(ii) these levels are not actually stationary states but correspond to resonances, the
electron–hole pairs can, for instance, de-excite by emission of a photon or become excited
Tunnelling and relaxation in semiconductor double quantum wells 367

(de-excited) by absorption (emission) of phonons to higher (lower) energy levels;


(iii) bulk excitons are very unstable in the presence of an electric field, some few
kilovolts per centimetre dissociate the bulk pairs (eF a ∗ ≈ Ry ∗ , where a ∗ is the effective
Bohr radius).
Excitons play a crucial role in low-dimensional structures. Excitonic features dominate
the optical response of such systems up to very high temperatures, due to their enhanced
binding with respect to the bulk three-dimensional case. Also, and in contrast to the bulk
case, a series of absorption peaks is currently observed when exciting far above the ground
exciton transition (see, e.g., Dingle et al 1975).
The exciton states in quantum wells have been thoroughly investigated theoretically
(see, e.g., Andreani (1994) and references cited therein). In the diagonal (or parabolic)
approximation one has two independent series of levels: heavy excitons and light excitons.
Generally speaking, we can envision the excitonic states in multiple quantum wells as
follows. In the absence of a Coulombic interaction the system presents a series of interband
edges: 1n,m = En − Hm (we restrict ourselves to the interband transitions between hole and
electron states bound for the z-motion). As far as the in-plane relative motion is concerned,
in the presence of a Coulombic interaction there is: (i) a continuum of levels corresponding
to dissociated (but correlated) pairs; (ii) a series of correlated states characterized by bound
relative motions for the pair, which are associated to a given (n, m) edge and with energies
εn,m (υ, K = 0) < 1n,m . In addition, bound states with different (n, m, υ) are coupled
by the Coulombic potential. Their interactions lead to the final bound excitonic states of
the multiple quantum wells. Finally, the Coulombic interaction also possibly couples the
excited bound states (for instance, the ones linked to the 11,3 = E1 − H3 edge) with the
dissociated states from low-energy edges (the former states become resonances; see, for
instance, Brum and Oberli (1993) and references therein). Altogether, these effects explain
the series of features currently observed in the optical absorption in quasi-two-dimensional
systems (see, e.g., Andreani (1994) and references therein).
Exciton states become resonances in the presence of a longitudinal electric field
(F kz). However, and in contrast to the bulk case, the field-induced broadenings of the
heterostructure levels are very often much smaller than the energy shifts (like the one-
particle states) and can be neglected in a first evaluation. This statement is corroborated
by the marked features observed in the optical response of highly biased structures
(F ≈ 100 kV cm−1 ).
Actual samples present both intrinsic (phonons) and extrinsic (residual doping) defects.
These scatter the carriers and broaden the excitonic levels. They are, in particular,
responsible for the energy relaxation of excited excitons and assist the interwell transfer
in coupled quantum wells. These scattering processes are considered in the next section.
A number of methods have been proposed to evaluate the exciton states. We classify
them into two broad groups. In the first group, one works with real space wavefunctions.
The peculiarities of the structure are accounted for in the establishment of the model, and
no general approach has so far been presented. On the one hand, one diagonalizes the
Coulombic interaction in a chosen basis for the envelopes (e.g. Gaussians). Variational
calculations in single and double quantum wells, on the other hand, take advantage of
physical trends which are genuine to the heterostructure: (i) for thin isolated wells the
electron and hole motions along the growth axis are forced by the confining walls whereas
the in-plane motion is dominated by the Coulombic interaction (see, for instance, Bastard
1988); (ii) in double quantum wells the Coulombic interaction simultaneously mixes the
various uncorrelated motions along the growth axis and governs the in-plane motion of
the pair (Galbraith and Duggan 1989, Fox et al 1991), which is equivalent to stating
368 R Ferreira and G Bastard

that the Coulombic interaction mixes the set of independent exciton states, each of them
being linked to a different couple of electron–hole states (Kamizato and Matsuura 1989,
Ferreira et al 1990; see also section 3.2 below). Leavitt and Little (1990) and Dignam
and Sipe (1991) present more elaborate variational procedures to describe the exciton states
in DQWs. In the second group, the calculations are performed in the momentum space.
Efficient algorithms exist to solve the set of coupled equations for the exciton envelopes
and an accurate description is obtained for both the bound and continuum states. These
approaches are particularly suited to handle the correlated continua and thus to evaluate
the absorption spectrum of quantum wells (see, for instance, Winkler (1995) and references
therein).
In the following we use a description based on real space envelopes to describe the
tunnelling effects on the exciton states in double quantum wells.

3.2. Direct and crossed excitons in double quantum wells


In double quantum wells the combined action of the narrow intermediate barrier and of the
long-range Coulombic attraction gives rise to two types of excitons:
(i) the direct excitons, formed between electron and hole states essentially localized in
the same well;
(ii) the crossed excitons, formed by pairing an electron and a hole mainly localized in
different wells.
In addition, important admixture effects are expected to occur each time two interband
transitions involving a common conduction or valence state anti-cross.
In the parabolic effective mass approximation the exciton Hamiltonian reads
Hexc = HCM + Hrel + He + Hh − (e2 /κ)/r (3.2)
where the four first terms account for the in-plane centre-of-mass, in-plane relative electron–
hole motion, independent electron and independent hole z-motions, respectively. The last
term is the Coulombic interaction with r denoting the electron–hole distance and κ the
relative dielectric constant of the heterostructure. The Hh Hamiltonian is taken as the heavy
or light diagonal terms of the Luttinger matrix. In this diagonal approximation the centre-
of-mass motion is completely decoupled from the relative motion and thus contributes to
the exciton energy only by an additive term (h̄2 K 2 /2M, where M is the sum of the electron
and hole in-plane masses and the wavevector K is related to the free in-plane propagation
of the centre of mass). He and Hh Hamiltonians are diagonalized in the standard way, and
the eigenvalues and envelopes are denoted by En , ϕn (ze ) for He , and Hm , χm (zh ) for Hh .
Let 9 be a linear combination of all the 1S-like exciton states that can be generated
in the decoupling approximation. That is, to each pair of states (ϕn (ze ), χm (zh )) we can
associate a 1S-like exciton state, with envelope given by
9n,m (ze , zh , ρ) = (2/πλ2n,m )1/2 ϕn (ze )χm (zh ) exp{−ρ/λn,m } (3.3)
where λn,m are variational parameters which are determined by minimizing H for the pair
(n, m) of electron and hole states. For this pair the exciton eigenenergy is labelled εn,m . In
the second step 9 is written as
X
9= αn,m 9n,m . (3.4)
n,m

The coefficients αn,m and the exciton energies ε are then determined after solving the
(Nc Nv )2 eigenvalue matrix det{h9|Hexc − ε1|9i} = 0, where Nc (Nv ) is the number of
electron (hole) states that generate the basis 9n,m . The only non-diagonal terms are due to
Tunnelling and relaxation in semiconductor double quantum wells 369

the coupling of different exciton states induced by the Coulomb potential (Kamizato and
Matsuura 1989, Ferreira et al 1990).
Another (simpler) variational envelope
X 

9(r) = (2/πλ2 ) exp[−ρ/λ] · ci,j ϕi (ze )χj (zh )
i,j
X
|ci,j | = 1
2
(3.5)
i,j

has also been used to deal with the ground exciton level (Galbraith and Duggan 1989) and
the direct and crossed exciton states (Fox et al 1991) in DQW heterostructures.
Optical studies (Ferreira et al 1990, Fox et al 1991) have evidenced the coupling
between direct and crossed exciton states in biased DQW structures, as discussed in the
rest of this section. Direct and crossed excitons have also been shown to play an important
role in the low-temperature vertical transport of excitation in such systems (see section 4).
In figure 12 we show the electric field dependence of the band-to-band (dashed curves)
and excitonic transitions for a 78 Å–55 Å–35 Å ADQW. We focus on an electric field
region around F0 ≈ −54 kV cm−1 , where the E2 and E1 electron levels anti-cross. Two
independent exciton states are obtained in the simplest approximation given in equation (3.3).
They correspond either to coupling ϕ1 (ze ) or ϕ2 (ze ) to the hole state χ1 (zh ) (the ground
hole level in this field region). The εn,m (n = 1, 2; m = 1) exciton energies at this
approximation are shown as full curves in figure 12. The presence of a direct (X in
figure 12) and a crossed (X00 in figure 12) exciton with very different binding energies
(≈ 8.1 and 3.5 meV, respectively) accounts for the shift in the electric field position of
the energy anti-crossing of the two excitonic levels (at F1 ≈ −57 kV cm−1 < F0 ). In
this decoupled approximation the excitonic minigap (≈ 1.6 meV at F1 ) is smaller than the
band-to-band gap (≈ 3.5 meV at F0 ). However, strong coupling effects between 92,1 and
91,1 are expected if |ε2,1 − ε1,1 | 6 Ry ∗ ≈ 5.5 meV. The results of the calculations for
admixed excitons are plotted as dotted curves in figure 12. The interaction between the
two 9n,m states (non-diagonal terms in h9|Hexc − ε|9i displaces the field where the anti-
crossing takes place to F2 ≈ −58 kV cm−1 < F1 < F0 and widens the excitonic minigap
(≈ 3.5 meV at F2 ). Note finally that 92,1 and 91,1 represent the direct and the crossed
states, respectively, if F  F0 (the crossed and the direct states if F0  F < 0).
In the dipole approximation only K ≈ 0 exciton states couple to the light on account of
the translational invariance in the layer plane. The strength of an optical transition is given
by its oscillator strength (OS), which for the states in equation (3.3) is proportional to
X 2

OS = (2/π) αn,m (a0∗ /λn,m )hϕn (z)|χm (z)i (3.6)
n,m

where a0∗ is the bulk effective Bohr radius. OS depends on the spatial overlap of the electron–
hole envelopes. Thus, direct excitons interact much more strongly with the light than
crossed ones. This feature is consistent with optical studies performed in biased structures
(see, e.g., Chen et al (1987) for photoluminescence excitation spectroscopy experiments).
In photoluminescence experiments (see, e.g., Liu et al 1989) a line, attributed to a crossed
transition, appears below a given threshold reverse voltage (−3.5 V for the 78 Å–55 Å–
35 Å ADQW; F ≈ F1 ). Actually, a crossed photoluminescence signal seems to appear
only when it corresponds to the ground interband transition.
370 R Ferreira and G Bastard

Figure 12. Calculated field dependences of several transitions energies in a 78 Å–55 Å–35 Å
GaAs–G0.7 Al0.3 As double quantum well. Dashed curves, band-to-band energies; full curves,
decoupled excitons; dotted curves, coupled excitons.

4. Assisted tunnelling

We have so far assumed the heterostructures to be perfect. In particular, the translational


invariance in the layer plane was assumed, in order to be able to define k (in this section
the in-plane wavevector k⊥ is denoted by k). Under such conditions, there is no possible
single-particle tunnelling, if tunnelling means an irreversible transfer from one eigenstate
to a group of others with different spatial localization. This is because the eigenstates we
have discussed are stationary solutions of the heterostructure Hamiltonian. Even if one
is able to prepare an initial state which is not an eigenstate, there will only be reversible
oscillations of the carrier mean position along the z-axis. Excitons, even without defects,
can undergo irreversible transfer through the Fano mechanism: excitons formed between
excited subbands can decay into the continuum of the lowest-lying electron–hole subbands
(see, e.g., Filoramo et al (1996) and references therein).
Static defects or coupling to the lattice vibrations are the necessary agents to induce
irreversibility in the single-particle case. Once this is acknowledged, it becomes a matter of
calculations to compute the transfer time (Sawaki et al 1988, Ferreira and Bastard 1989a, b,
Lary et al 1989). Such calculations may in fact be involved because the irreversibility is
brought about by a coupling Hamiltonian which is very difficult to handle beyond the lowest
order of perturbation theory.
The actual single-particle heterostructure Hamiltonian we have to deal with can be
formally written

H = H0 + W (4.1)
H0 |n, ki = En (k)|n, ki (4.2)
Tunnelling and relaxation in semiconductor double quantum wells 371

where W is the potential describing the interaction with static defects:


X
W = w(r − Ri ).
i
Ri denotes the random location of the scatterers. These scatterers can be ionized impurities
and interface defects, etc. The coupling with lattice vibrations is represented by the electron–
phonon interaction, either of Fröhlich type for the coupling with longitudinal (LO) phonons
or the deformation potential for the interaction with the longitudinal acoustical phonons. The
piezoelectric scattering with acoustical phonons usually produces smaller transition rates (for
a review on scattering in semiconductors see, e.g., Ridley (1988)). Similar expressions exist
for holes with the additional complications that the holes have 4 × 1 envelope functions and
that their coupling with the phonons can be more complicated to handle.
To lowest order in perturbation theory (Fermi golden rule) the effect of the interaction
between the carriers and defects, phonons, etc, amounts to inducing irreversible transitions
from one given initial state |n, ki to the continuum of states |n0 , k0 i with a transition rate
per unit time 1/τnk . For elastic scattering this rate is
X
h̄/(2τnk ) = π |hn0 k0 |W |nki|2 δ(Enk − En0 k0 ). (4.3)
k0
On account of the random positions of the impurities this transition rate is simply the number
of scatterers times the transition rate for one scatterer. Let (0, zi ) denote the position of
this scatterer (it is necessary to specify its position along the growth axis since there is no
translational invariance along the z-axis) and assume for the sake of definiteness that this is
a Coulombic centre (very often the heterostructure, in particular those which involve GaAs,
display a residual p-type doping). One finds readily that for electrons
Z Z
h̄/(2τnk ) = Nimp π(e2 /κ)2 dq/q dθ|hφn0 | exp(−q|z − zi |)|φn i|2 δ(1En0 ,n )
(4.4)
∗ ∗
1En0 ,n = En0 − En + h̄ q /2m + h̄ kq cos θ/m
2 2 2

where Nimp denotes the areal concentration of scatterers (Ferreira and Bastard 1989a). For
a zero initial wavevector this expression simplifies to
h̄/(2τnk=0 ) = π 2 Nimp (e2 /κ)2 (En − En0 )(−1) |hφn0 | exp(−q|z − zi |)|φn i|2
(4.5)
q 2 = (2m∗ /h̄2 )(En − En0 ).
This expression shows very clearly the resonances one should expect in the interwell transfer:
the transfer time shortens considerably when by some means (e.g. an electric field) the energy
difference (En − En0 ) is made to decrease. Thus, when |n0 i and |ni anti-cross and that one
follows the escape time from the upper branch |ni, one finds that this time decreases,
reaches a minimum and increases again. There are two factors which contribute to a more
efficient transfer rate. Not only does the energy difference (En − En0 ) decrease but also
the overlap between the envelopes φn0 and φn is bigger which increases the matrix element
appearing in equation (4.5). Note, however, that the transfer time never vanishes because
even if (En − En0 ) were to vanish (which never actually happens) the matrix element would
also vanish on account of the orthogonality of the initial and final states, which cancels the
apparent divergency of the denominator. In the limit of large detuning, one may use the fact
that exp(−q|z − zi |) ≈ (2/q)δ(z − zi ) to obtain the asymptotic dependence of the assisted
tunnelling:
h̄/(2τnk=0 ) = (π 2 /2m∗ )Nimp (2h̄e2 /κ)2 (En − En0 )(−2) |φn∗0 (zi )φn (zi )|2 . (4.6)
This expression clearly displays the exponential decrease of the assisted tunnelling upon
the intermediate barrier thickness since either φn or φn0 is evanescent at zi . It also predicts
372 R Ferreira and G Bastard

a power-law increase of τnk=0 upon the detuning. When the detuning becomes very large,
the assisted interwell tunnelling becomes very ineffective. It can, in practice, be overcome
by other mechanisms, also due to defects, phonons, etc, but of quite different origin since
these are the intrawell scattering events, neglected in the previous analysis, which play
a decisive part. Let us assume that the double well is far from resonance (λ  1e in
equation (2.5)) and that there exists scatterers, say static, which broaden the isolated well
eigenstates. Our aim is to evaluate the lifetime of the levels of one of the wells, well (1)
for definiteness, due to the combined action of the intrawell scattering and of the weak
tunnel coupling between the levels n and m localized, respectively, in the wells (1) and
(2). Following section 2.4, we label the eigenstates forming the nth dispersion of well (1),
when isolated, by |n(1), ki without scatterers and tunnel effect. Due to scattering, |n(1), ki
remain approximate eigenstates of the well (1) but the corresponding eigenenergies become
complex: en(1)k − iγn(1)k . A similar expression holds for the states in well (2). Since
tunnelling is a small effect it can be treated by second-order perturbation theory. As a
result, the imaginary part of the eigenenergies of well (1) become
γn(1)k + λ2 [γm(2)k − γn(1)k ]/[em(2)k − en(1)k ]2 . (4.7)
Thus, the intrawell scatterings inside one well contribute to the lifetime of the eigenstates
of the other well (see section 5).
The previous considerations hold for all the scattering mechanisms and for both the
electrons and holes. In the case of the emission of LO phonons the resonance in the
transfer rate involves the matching of the energy difference (En − En0 ) with the LO phonon
energy h̄ωLO .
We illustrate the resonances which occur in the rates of assisted transfer in biased double
quantum wells in figure 13 for electrons and in figure 14 for holes. In the latter case, there
is not only an enhancement of the assisted tunnelling when two heavy-hole branches anti-
cross, but also when a light-hole related branch anti-crosses a heavy-hole related one. This
is because the final states of the assisted transition have a non-zero in-plane wavevector and
therefore an mJ mixing, which allows a coupling with any other branch despite the fact that
the coupling Hamiltonian is diagonal in the mJ index (Ferreira and Bastard 1989b). On
comparing the shape of the τ (F ) curves for electrons and holes, one notices that the electron
resonances are broader and shallower than the hole resonances. This is a result of the smaller
magnitude of the hole anti-crossings. One needs to come very close to the anti-crossing field
to get an appreciable delocalization of the hole wavefunction and thus an enhancement of the
interwell assisted tunnelling of holes. In agreement with equation (4.5) the assisted transfer
times for holes are at resonance shorter than that of the electrons, since the anti-crossing gaps
are smaller. However, one should be very cautious in giving physical significance to such
very deep hole resonances. They are the result of a lowest-order perturbation calculation and
one should check a posteriori that the predicted results are compatible with the underlying
assumptions. In the specific case of the transfer rates a qualitative applicability criterion can
be set by requiring that the initial level broadening h̄/(2τnk ) associated with the transfer rates
remain smaller than the energy spacings between the unperturbed levels. The upper and
lower branches are at least separated by En (k = 0)−En0 (k0 = 0). To make sense the transfer
rate should be such that h̄/(2τnk=0 )  |En (k = 0) − En0 (k0 = 0)|. We show in figure 15
the field dependences of these two quantities involving the first two heavy-hole levels in
a 50 Å–40 Å–50 Å DQW. It is seen that the narrowness of the hole resonance implies a
failure of the simple Born type of calculations and call for more elaborate approaches (see
next section). For electrons instead, the larger anti-crossing gaps ensure that the criterion is
always fulfilled, at least approximately.
Tunnelling and relaxation in semiconductor double quantum wells 373

Figure 13. Calculated resonances in the assisted transfer of electrons in a 100 Å–50 Å–
47 Å GaAs–G0.7 Al0.3 As DQW. Transfer times (log scale) for E2 (τ2 = τ2→1 ) and E3 (1/τ3 =
1/τ3→2 + 1/τ3→1 ). The transfer is assisted by 1010 cm−2 ionized impurities assumed to be
located on the two ‘inverted’ interfaces separating GaAs and GaAs–G0.7 Al0.3 As.

The exciton assisted transfer can be treated along the same lines as used for single
particles. In the exciton case, one evaluates the transition rate for an exciton in the νth
branch (ν refers to the internal degrees of freedom of the exciton) and with an in-plane
wavevector K to make a transition to the µth branch with any in-plane wavevector K 0 .
Consider again the static defects case. The exciton is coupled to the defects through electron
coupling and hole coupling to the defects. It is convenient to express the in-plane position
vectors of the hole and the electron in terms of those of the centre of mass (R) and of the
relative position (ρ). The exciton defect Hamiltonian is thus
X
Hexc−def = Ve (ze − zi , R − ρi + mh /Mρ) + Vh (zh − zi , R − ρi − me /Mρ) (4.8)
i
where ρi is the in-plane (random) position of the scatterers, zi their location along the
growth axis and M is the total exciton mass in the layer plane. The assisted tunnelling rate
for the exciton will thus be equal to
X
h̄/(2τνK ) = π |hνK|Hexc−def |µK 0 i|2 δ(ενK − εµK 0 ). (4.9)
K0
In the simple model of the exciton, which assumes a factorization of the exciton envelope in
terms of the envelopes of the electron and hole z-motion (equation (3.3)) and due to the shape
of the exciton–defect interaction, namely the sum of the electron and hole contributions, it
374 R Ferreira and G Bastard

Figure 14. Calculated resonances in the assisted transfer of holes in a 50 Å–40 Å–50 Å GaAs–
G0.7 Al0.3 As double quantum well without and with allowance of the band mixing effects (dashed
and full curves, respectively).

can be very easily checked that the only possible exciton transitions are those which ensure
the conservation of either the electron quantum number for the z-motion or that of the hole
quantum number along the z-axis. For instance, in the specific case of Coulombic impurities
located at zi , with an areal concentration equal to Nimp , we find the assisted tunnelling time
τnm0 from the level |En , Hm , 1S, K = 0i to the branch |En0 , Hm0 , 1S, K 0 i to be equal to
h̄/(2τnm0 ) = 2π 2 MNimp (2πe2 |81 − 82 |/κh̄Q)2 (4.10)
0
81 = δm,m0 hn| exp{−Q|ze − zi |}|n ih1Snm |J0 (Qρmh /M)|1Sn0 m0 i
82 = δn,n0 hm| exp{−Q|zh − zi |}|m0 ih1Snm |J0 (Qρme /M)|1Sn0 m0 i (4.11)

Q = 2M(En + Hm −
2
Rnm −E −Hn0 m0 + Rn∗0 m0 )/h̄2 (4.12)

where Rnm is the exciton binding energy when the electron and the hole are respectively in
the En and Hm states for their z-motion and J0 is the Bessel function. Clearly the transition
rate vanishes when both n 6= n0 and m 6= m0 . In other words, only oblique exciton transfers
are allowed while the assisted tunnellings of the exciton as a whole entity are forbidden.
More elaborate exciton envelopes which account for the Coulomb-induced coupling between
the various exciton states (see section 3.1 and Michl et al (1995)) allow the latter kind of
transfers. Their rates are, however, found to be often smaller than those of the oblique
tunnelling.
It is also important to realize that the exciton transition rates are maximum at the
exciton anti-crossings and not at the single-particle anti-crossings, in spite of the fact that
in both cases this is the same particle which undergoes the assisted tunnelling but in one
case irrespective of the other particle while in the exciton transfer the two particles remain
strongly correlated (Ferreira et al 1990, Fox et al 1991). To give an example, consider
the case of a biased asymmetric double quantum well. One may line up the two ground
Tunnelling and relaxation in semiconductor double quantum wells 375

Figure 15. Failure of the Born approximation in the case of a hole resonance in a biased 50 Å–
40 Å–50 Å GaAs–G0.7 Al0.3 As double quantum well. Dashed curve, electric field dependence
of the energy difference between the two interacting hole states; full curve, field dependence of
the broadening h̄/2τ2→1 of the upper level. At resonance the broadening exceeds the energy
difference.

electron states which anti-cross. If the electron tunnels from the upper branch to the lower
one irrespective of the hole, the maximum transition rate will take place at a field F such
that

eF (B + (L1 + L2 )/2) = e1(1) − e1(2) (4.13)

while the maximum exciton transfer will happen when

eF (B + (L1 + L2 )/2) = e1(1) − Rd − e1(2) + Ri (4.14)

where e1(p) denote the ground electron energy of the pth isolated well and where Rd and Ri
denote the binding energies of the spatially direct e1(1) − h1(1) and of the spatially indirect
e1(2) − h1(1) excitons, respectively (see sections 2.4 and 3.1). The two exciton binding
energies Rd and Ri are different (≈ 10 meV against 3 meV) and thus the resonance in the
exciton transfer does not take place at the same field as the bare electron resonance. In
addition, the optically detected resonance field should not be the same if one looks at the
e1(1) − e1(2) resonance through another optical transition, say one which involves h1(2) or the
light-hole related ones. These considerations are illustrated in figure 16 where we compare
the transfer times for an electron alone (dotted curve) to those of excitons which involve
the anti-crossing electron levels in a 78 Å–55 Å–35 Å GaAs–Ga(Al)As double quantum
well. The transfer is assisted by Coulombic impurities (Nimp = 1010 cm−2 ) located on
the two ‘inverted’ GaAs–Ga(Al)As interfaces. The minimum in the three transfer times
are reached when, in each case, the interacting entities anti-cross. The three resonance
fields are therefore different and, in particular, one does not find the same resonance field
upon looking at two different optical transitions. This feature is the key signature of the
excitonic nature in transfer resonances. It should not take place if one of the particles were
transferring irrespective of the other.
376 R Ferreira and G Bastard

Figure 16. Electric field dependence of the impurity assisted electron (dotted curve) and exciton
(X → X00 and X0 → X000 ) transfer times. The shorthand notations X, X0 , X00 and X000 replace the
1S-like states X1S (e1(1) − h1(1) ), X1S (e1(2) − h1(2) ), X1S (e1(2) − h1(1) ) and X1S (e1(1) − h1(2) ),
respectively. Full (dashed-dotted) curves denote coupled (uncoupled) exciton states.

5. Vertical transport in double quantum wells

We have discussed in sections 2 and 3 the stationary eigenstates of a DQW structure. In


a tight-binding picture the coherent tunnel interaction couples states localized in different
wells but with the same in-plane wavevector. The energy levels of the isolated wells anti-
cross and the DQW envelopes delocalize over the whole structure. Also, an electric field
applied along the growth axis weakens the resonant coupling and the envelopes localize
inside the two wells. Optical experiments are able to probe these effects. In particular,
the energy splitting due to the interwell coupling in unbiased SDQWs and the localization
effects in asymmetric and/or biased structures have been observed optically (Dingle et al
1975, Delalande et al 1984, Nelson et al 1986, Chen et al 1987, Andrews et al 1988,
Charbonneau et al 1988, Sauer et al 1988, Liu et al 1989, Golub et al 1990, Alexandrou
et al 1990).
In this section we are mainly interested in the time evolution of a photoexcited
population, as obtained for instance in time-resolved experiments, where carriers are
photocreated by a fast laser pulse and the optical response of the structure is observed
as a function of time (see section 5.2). These experiments probe the vertical (or z-related)
transport of carriers as follows. After the resonant excitation of a transition linked to one of
the two wells, the population of correlated and/or uncorrelated carriers redistribute inside the
two wells. Scattering processes are able to assist this transfer. As a consequence, a new low-
energy PL line possibly appears, the intensity of which depends directly on the efficiency
of the tunnelling. Thus, finally, in this picture the interwell transport is proportional to
the interwell scattering rate. Intrawell processes are neglected in this picture, since they
do not directly affect the carriers’ localization along the growth axis. In this last section
we both apply this simple picture to some experimental situations and critically review its
applicability. In particular, we are interested in discussing the role played by the intrawell
scattering processes on the interwell vertical transport. That these intrawell processes should
play a role on the carriers’ tunnelling can be more easily seen if we consider the opposite
picture for the vertical transport, namely, the situation where the interwell scatterings can
Tunnelling and relaxation in semiconductor double quantum wells 377

be neglected and where only intrawell scatterings perturb the coherent interwell coupling.
As we shall see, in the strong scattering regime the tunnelling probability is inversely
proportional to the scattering rate.
We can state the same point in a different way. On very general grounds, the transport
in bulk materials is limited by the presence of defects and the net current is related to a
sequence of incoherent events (scatterings of the carriers). The carriers’ mean free path is
generally much smaller than the extension of the bulk unperturbed eigenstates, that is the
crystal dimension. In quasi-two-dimensional structures the electronic states are localized in
a narrow region, with respect to the crystal length along the growth axis. If, in addition,
this confinement length becomes smaller than the mean free path, as is often the case in
practice, coherent effects are expected for the carriers motions, among which we have the
energy splitting due to the tunnel coupling in double quantum wells and its counterpart in
the time domain, the ‘Rabi’-like oscillations. (Other examples, such as Bloch oscillations
in biased superlattices, are not discussed in this review; see, e.g., Waschke et al 1993.) The
same kind of reasoning does not apply to the in-plane motion, which is free and thus more
easily affected by the defects in actual structures. Such defects, as discussed in the previous
section, couple the in-plane and z-motions. The question to be solved is: To what extent do
such perturbations for the in-plane motion affect the coherent interwell coupling in actual
double quantum wells?
In this last section we deal with the main physical aspects and theoretical difficulties
underlying the description of the broadened levels of coupled quantum wells. We review
a few models which are used to describe the assisted interwell transfer and relaxation in
these systems. We start from a two-level description of a perfect structure (section 5.1).
Section 5.2 presents a brief review of some experimental results. In the remaining sections
(5.3 and 5.4) we consider a few broadening mechanisms and discuss how to account for
each of them in a simple way.

5.1. Time evolution in perfect structures


Note first that the tunnel coupling conserves the in-plane motion. Also, for a fixed in-plane
wavevector, the bound states of a perfect (defect free) structure form a finite set of discrete
levels. Thus, as far as only bound levels are concerned, the time evolution corresponding
to a given (arbitrary) initial condition is necessarily oscillatory. We recall here a simple
model to describe the interwell oscillations in perfect DQWs. Let us take only one bound
level per well. The time-dependent envelope function for the resulting two-level problem
reads

9(z, t) = a1 (t)φ1 (z) + a2 (t)φ2 (z) (5.1)

where φn (z) is the isolated well solution related to en and n = 1, 2 labels the two
wells (in this section (e1 , φ1 ) and (e2 , φ2 ) are an abbreviated notation for (en(1) , φn(1) ) and
(em(2) , φm(2) ), respectively, defined in section 2.4). In the tight-binding approximation we
have

ih̄ dan (t)/dt = en an (t) + λc am6=n (t) (5.2)

where λc is the transfer integral defined in section 2. Let E1 and E2 denote the stationary
eigenenergies of the DQW. The non-stationary solutions are

an (t) = αn,1 exp{−iE1 t/h̄} + αn,2 exp{−iE2 t/h̄} (5.3)


378 R Ferreira and G Bastard

where αn,m depend on the initial conditions. For instance, α1,1 = −α1,2 = −λc /(E2 − E1 )
and α2,1 = 1 − α2,2 = (e1 − E1 )/(E2 − E1 ) for a carrier localized in well ‘2’ at t = 0
(a1 (0) = 0 and a2 (0) = 1).
One easily shows that the amplitudes |a1 (t)|2 and |a2 (t)|2 are periodic in time with
periodicity T = 2πh̄/1E, where 1E = E2 − E1 = [(1e)2 + 4λ2c ]1/2 and 1e = e2 − e1 .
In the tight-binding approach each squared amplitude gives the probabilities of finding the
carrier in one of the two wells. Thus, the carrier strongly confined in one well at t = 0
starts oscillating between the two wells. The averages over a period of these probabilities
Z
1 t0 +T
In = |an (t)|2 dt (5.4)
T t0
are I1 = 1 − I2 = 2λ2c /(1E)2 . In are independent of the time origin t0 and give the
‘residence’ times in the two wells per oscillation period. In the weak coupling regime
|1e/λc |  1, I1 → 0, I2 → 1 and the probability of finding the carrier in well ‘1’ is
negligible (we recover the temporal counterpart of the spatial localization of the eigenstates
of a system when the detuning 1e is larger than the tunnel coupling λc ). When |1e/λc |  1
(the strong coupling regime) one finds I1 ≈ I2 ≈ 12 and the carrier ‘visits’ almost equally
the two wells.

5.2. Summary of experimental results


One easily finds that the amplitude of the oscillations is very sensitive to the initial
conditions. In particular, no oscillation should appear any time a1 (0) and a2 (0) are
such as to reproduce at t = 0 one of the two stationary tight-binding solutions. In
practice, the existence of oscillations is linked to the optical preparation of the system.
In a very simple picture, we have to consider three characteristic energy widths: 1E,
the separation between the two stationary levels; the broadenings Gi (due to radiative or
non-radiative processes) and the ‘intrinsic’ width h̄1ωL of the laser line. In practice the
inequalities Gi < 1E < h̄1ωi need to be fulfilled to ensure: (i) that the two levels are
optically resolved (Gi < 1E); (ii) that they are coherently excited (1E < h̄1ωL ); and
(iii) that the level lifetimes are much longer than the pulse duration (h̄/Gi > 1/1ωL ).
Under these conditions carriers are possibly photocreated in a non-stationary state of
the structure, which is a condition for the observation of such oscillations. In practical
units we have T (ps)1E(meV) ≈ 4.13. At resonance, T ≈ 4 ps for a tunnel splitting
2|λc | ≈ 1 meV.
‘Rabi’-like oscillations have been observed (Leo et al 1991) in ADQW biased in a
field region around one interwell energy resonance of two electron levels by using an
optical system operating in the subpicosecond regime (time resolution ≈ 100 fs). The
results show that the oscillations disappear as one goes far from the resonant bias. In
addition, the oscillations are considerably damped, and in practice only a few periods are
resolved. This latter result should, at least in part, be related to incoherent scatterings of
the carriers.
To our knowledge, oscillations related to an interwell hole resonance have not been
reported up to now. This can be qualitatively understood as follows. For holes the energy
splittings are generally much smaller than those for electrons. On the one hand, one obtains
that the electric field intervals for the strong coupling regime are much narrower and
the field positions for the interwell resonances are probably broadened by the interface
fluctuations. On the other hand, one has much longer periods for the oscillations (since
the anti-crossings are smaller) and thus various scattering events can possibly occur before
Tunnelling and relaxation in semiconductor double quantum wells 379

a complete oscillation, since the condition Gi < 1E often does not hold. Then, the
time evolution for holes in double quantum wells should be dominated by incoherent
processes.
By using excitation and detection techniques with time resolution τL > 1 ps, one is able
to selectively excite one of a series of well separated (1E > h̄/τL ) interband transitions.
Although no evolution is expected when an (0) reproduces one stationary solution, a time
dependence of the luminescence lines is nevertheless observed which suggests the existence
of an interwell transfer of carriers. As discussed in section 4, this is, at least partly, due
to the presence of defects and, in contrast with Rabi-like oscillations, is an irreversible
phenomenon. cw and time-resolving techniques are widely used to study the tunnelling
processes (Tada et al 1988, Liu et al 1989, Charbonneau et al 1988, Norris et al 1989,
Alexander et al 1989). The relative intensities of the cw luminescence lines and the observed
rise and decay times of the optical signals depend on the interwell transfer, on the existence
of non-radiative and radiative channels and on various intrawell relaxation processes. A
number of studies have been performed in double quantum wells and the transfer times
have been directly measured or indirectly evaluated. For instance, one is able to excite
quasiresonantly the ground transition of the narrow well and record the various emission
lines. The decay of the luminescence of the same narrow well should give roughly the
interwell transfer time (if it is much shorter than the recombination time). The resonant
nature of the interwell coupling has been evidenced in experiments where the relative
position of the DQW levels could be externally monitored (for instance, by applying an
electric field): the measured decays have been shown to rapidly vary with the energy
detuning around one interwell resonance. Electronic (Oberli et al 1989, Deveaud et al
1990, Livescu et al 1990) and optical phonon related (Oberli et al 1990) interwell resonances
have been considered. The transfer of holes (Leo et al 1990, Norris et al 1991, Nido et
al 1990, 1991, Roussignol et al 1991) and excitons (Ferreira et al 1990) have also been
experimentally investigated.

5.3. Time evolution in real structures: simple theoretical models


The presence of defects affects the time evolution of a photocreated population around an
interwell energy resonance as described in section 5.1. As we shall see later, a pertinent
parameter at resonance is the ratio η = Gsc /|λc |, where Gsc is a characteristic broadening
energy (to be considered below). For η < 1, one should observe damped interwell
oscillations, whereas a diffusive regime should appear for η > 1. To be more quantitative
we need to explicitly consider the scattering potentials. Here we describe simple theoretical
models. Only a few works treat the time evolution of photocreated carriers in imperfect
structures on a more rigorous basis (beyond the simple models to be described below; see,
for instance, Kazarinov and Suris (1972), Vas’ko and Raichev (1995), and references therein,
and Ferreira (1996)). The main reason is that any complete theory should incorporate the
rather different underlying physics of the weak and strong scattering regimes: in the first
case we have a collection of two-level systems coupled by a random distribution of three-
dimensional defect potentials, whereas in the strong scattering case the tunnel interaction
couples the in-plane scattering states of two independent wells. Perturbative methods,
although unable to treat the whole problem, can nevertheless be used in some limiting cases
(see section 5.4.1).
For simplicity, and as in sections 2.4 and 5.1, we assume that two subband edges are
quasiresonant and well separated in energy from the others. We shall make correspondingly
a distinction between three different broadening mechanisms (see figure 17): Pext , processes
380 R Ferreira and G Bastard

Figure 17. Sketch of the various paths a carrier may follow in a double quantum well with two
tunnel coupled subbands and an ‘external’ subband where the carrier may escape.

which empty the populations of the two ‘main’ subbands (such as couplings with ‘external’
subbands, radiative recombination and escape out of the DQW region); Pinter , interwell
scatterings between states of the two ‘main’ subbands; Pintra , intrawell scatterings within
each one of the two ‘main’ subbands. Thus, due to the continua of in-plane states, the two-
level model of section 5.1 should be replaced by a ‘two-subband’ model for the description
of the time evolution in actual (not perfect) double quantum wells. Note that not all the
processes Pext are necessarily linked to the presence of defects. Note also that processes
Pinter and Pintra mix the states of the two main subbands, whereas processes Pext can be (to
lowest order) considered within a two-level model. We consider in sections 5.3.1 and 5.3.2
two different ways to deal with the broadening due to Pext , which includes the dominant
broadening mechanisms in a few practical cases. In section 5.3.3 we discuss the applicability
of the simple two-level models to processes Pinter and Pintra .

5.3.1. Damping of the probability amplitudes. The first model consists of replacing the
single well energies en by en − iγn /2, with γn = h̄0n > 0, in the tight-binding equations
for a1 (t) and a2 (t). γn are either phenomenological or calculated (for instance, by using
the Born approximation) broadenings. A dependence upon k⊥ of en and γn is implicitly
assumed. The complex DQW energies read

E± = (e2 + e1 )/2 − i(γ2 + γ1 )/4 ± {λ2c + [(1e/2) − i(γ2 − γ1 )/4]2 }1/2 . (5.5)

For clarity let us first consider the resonant case (1e = 0). The resulting time evolutions
depend strongly on the ratio |(γ2 − γ1 )/4λc |, which plays the role of the parameter η
defined above. Note that the interwell coupling is not destroyed when the broadenings
for the two isolated wells are equal, in the sense that although |an (t)|2 present envelopes
decaying exponentially with time constant 1/ 01 , they nevertheless oscillate in time with
the unperturbed period T = 2π/1E = π/|λc |. A rather different behaviour is observed
for unequal broadenings. As a consequence of the asymmetry introduced by the unequal
broadenings, both damped and diffusive (non-oscillatory or overdamped) regimes are
possible for the time evolutions of the amplitudes, as follows. In the weakly damped regime
the energy splitting 1E ≈ 2|λc |−(γ2 −γ1 )2 /16|λc | slightly diminishes, the oscillation period
T increases and the amplitudes decay with equal rate 00 = (02 + 01 )/2 in the presence of
scatterings. In the strongly diffusive regime, on the other hand, two decay times τA and τB
Tunnelling and relaxation in semiconductor double quantum wells 381

govern the time evolutions of a1 (t) and a2 (t):


2h̄/τA ≈ γmax − γ0 2h̄/τB ≈ γmin + γ0
γ0 = 4λ2c /1γ 1γ = |γ2 − γ1 |
γmax = max{γ2 , γ1 } γmin = min{γ2 , γ1 }. (5.6)
If, in addition, γmax  γmin ≈ 0, then we have
τB ≈ h̄γmax /(2λ2c )  τA ≈ 2h̄/γmax . (5.7)
For long times (t  τA ) the probabilities decay with the rate r0 where
r0 = 2/τB ≈ (1/2)τA 2 (5.8)
where h̄ = 2|λc | is the tunnel splitting for the unperturbed structure. It is worth noting
that the decay rate at long times decreases with increasing scattering rate, a ubiquitous
regime of two-level systems (motional narrowing in NMR, Dyakonov–Perel spin-relaxation
in semiconductors, etc). This model of two coupled oscillators, one of them highly damped,
has been used by Leo et al (1990) and Gurvitz et al (1991) to study the interwell transfer
in double quantum wells where one of the interacting states is strongly coupled to the
continuum of lower-lying states of the structure. The damping is then related to the
existence of fast escape processes for the carriers, which selectively empty one of the
coupled levels.
In the non-resonant case (1e = h̄ω0 6= 0) one obtains for the two times τA and τB in
equation (5.6) that γ0 = 1γ λ2c /[(ω0 )2 + (1γ /2)2 ] in the limit (ω0 )2 + (1γ /2)2  2 . If
γmax  γmin ≈ 0, then for long times (t  τA ≈ 2h̄/γmax ) the probabilities decay with the
rate
r = 2/τB ≈ (1/2)τA 2 /[1 + (ω0 τA )2 ]. (5.9)
Thus, the electric-field-induced energy detuning in biased DQWs then acts much like the
magnetic-field-induced suppression of the spin depolarization in an external longitudinal
field (see, for instance, Pikus and Titkov (1984)).

5.3.2. Density matrix formulation. In a second model one uses the density matrix formalism
to treat the interplay between the coherent tunnel coupling and the incoherent processes (for
reviews on the use of the density matrix approach to transport phenomena in solids see,
e.g., Jones and March (1973) and to the coupling between material systems and radiation
see, e.g., Cohen-Tannoudji et al (1988) and Haug and Koch (1990) and references cited
therein). This more general approach not only reproduces the previous results but enables
one to go beyond the first model, as discussed in section 5.3.3. Here we consider the time
evolution of the density matrix ρ for two levels coherently coupled by the tunnel interaction
and broadened by their interactions with a continuum. Let (ρ)m,n and
(H0 )m,n = (en − iγn /2)δm,n + (1 − δm,n )λc (5.10)
be the matrix elements for ρ and for the Hamiltonian in the basis of the isolated wells states
(m, n = 1, 2) and for a fixed k⊥ value (the dependence upon k⊥ is implicitly assumed).
Then one solves (with suitable initial conditions) the Liouville equation
ih̄∂ρ/∂t = (H0 )ρ − ρ(H0 )+ . (5.11)
One finds that the broadenings γn not only decrease the populations (ρ)n,n but also affect
the coherences (non-diagonal terms (ρ)1,2 and (ρ)2,1 ). One can also show that at resonance
the relevant parameter is η = |1γ /4λc |, as in the first model, and all the limiting situations
382 R Ferreira and G Bastard

discussed above (for η  1; η = 0 or η  1) are recovered in the density matrix formalism.


Actually, the time evolution of (ρ)n,m in equations (5.11) equals by definition the time
evolution of the product an (t)[am (t)]∗ , as one can easily check by using equations (5.2).

5.3.3. Beyond the previous models. The equivalence of the two models in sections 5.3.1
and 5.3.2 for two-level systems is stressed, for example, in Cohen-Tannoudji et al (1988),
where the applicability of such a kind of formalism is also critically reviewed. In
particular, one cannot find a two-level Hamiltonian (even generalized to complex energies)
to describe, in terms of a Schrödinger-like equation, several processes which conserve
the total population, such as (i) interwell assisted transfer from one well to its neighbour
and (ii) intrawell scatterings which affect the coherences but not the subband populations.
These two scattering processes can nevertheless be dealt with within the density matrix
formalism.
It is important to realize that such processes are unavoidably present in any actual
coupled semiconductor quantum wells. In ideal structures the in-plane motion is a good
quantum number and the DQW can be described by a collection of independent (for a fixed
k⊥ ) two-level systems. In real structures, on the other hand, k⊥ is no longer a constant of
motion. We separate three contributions for the total broadening due to the defects (which
respectively belong to the Pext , Pinter and Pintra processes) and consider separately their effects
on the time evolution of the density matrix. Intrawell intersubband processes are responsible
for the escape of carriers and then contribute to the γn broadenings in the previous sections.
Interwell assisted scatterings lead to an effective transfer. Finally, intrawell intrasubband
processes lead to an additional damping of the ‘dipole’ (ρ)2,1 −(ρ)1,2 (see below). Note that
higher-order interactions for both the intrawell intersubband and the interwell scatterings
couple states of a dispersion branch with different in-plane wavevectors, very much like
Pintra processes. We assume that these higher-order contributions can be either neglected
or incorporated in Pintra . This ensures that k⊥ remains a constant of motion if only Pext
scatterings are present, as implicitly assumed in sections 5.3.1 and 5.3.2. Note finally that
Pintra processes possibly dominate when the escape probability decreases (for instance, when
no optical phonon can be emitted or when the interacting levels are the ground levels of
each well). Thus, even if the non-radiative and radiative escape rates vanish (γn = 0) and
there is an absence of interwell scatterings, the interwell coherent evolution in time can
nevertheless be modified due to the existence of intrawell scatterings. This is not very
surprising if one remembers that the presence of defects strongly couples the in-plane and
z-motions in coupled quantum wells.
First, let us consider Pintra processes. The (dense) set of independent oscillators, each
related to a different k⊥ , becomes coupled by the intrawell scatterings. These do not change
the populations of the two wells and can roughly be accounted for by a phenomenological
damping of the coherences:
[∂(ρ)1,2 ]coll /∂t = −0intra (ρ)1,2 . (5.12)
A similar expression holds for (ρ)2,1 . In other words, we assume that the part played by
the non-diagonal (in k⊥ ) components of the scattering and density matrices on the time
evolution of the diagonal in k⊥ components of the density matrix can be roughly accounted
for by the diagonal in k⊥ broadening parameter 0intra (which is possibly k⊥ dependent).
Now let us discuss an important limiting case. Adding the Pintra and Pext processes, one
obtains from equations (5.11) and (5.12) that the coherences decay with a characteristic
time ≈ [0intra + 00 ]−1 . If the populations vary sufficiently slowly on this time scale (see
below), then the coherences can be assigned their time-independent value obtained from
Tunnelling and relaxation in semiconductor double quantum wells 383

[∂(ρ)1,2 ]total /∂t = 0 at each time. In this case, the population (ρ)n,n fulfils
∂(ρ)n,n /∂t ≈ −[0n + 0c ](ρ)n,n + 0c (ρ)m,m (5.13a)
0c ≈ (1/2)τT 2 /[1 + (ω0 τT )2 ] (5.13b)
where n, m = 1 or 2, n 6= m and τT = 1/(0intra + 00 ). Equations (5.13) are valid if the
condition 0n +0c  0intra +00 is fulfilled. When the Pintra processes dominate, τT ≈ 1/ 0intra
and equations (5.13) are valid if 2  (0intra )2 + (ω0 )2 . It is worth comparing the resulting
expression for 0c to the result obtained previously in the large detuning limit. The short time
τA in equation (5.9) is related to an escape process while the rate 0intra is related to intrawell
scatterings. In fact, for a given oscillator with fixed k⊥ , 0intra can also be considered as
an ‘escape’ probability. Note finally that equations (5.13) do not hold when 0intra = 0, no
matter how large is |1e| and for any 0n values.
Let us now consider the Pinter scatterings. Strictly speaking, elastic and inelastic
Pinter processes couple states with different in-plane wavevectors and affect not only the
populations but also the coherences (like Pext ). As a crude approximation for the large
detuning limit we have
∂(ρ)2 /∂t ≈ −[02 + 0c + 0inter ](ρ)2 + 0c (ρ)1 + G2 (t) (5.14a)
∂(ρ)1 /∂t ≈ −[01 + 0c ](ρ)1 + [0c + 0inter ](ρ)2 + G1 (t) (5.14b)
where 1e = e2 −e1 > 0, 0c is given by equation (5.13b) with τT = 1/(0intra +00 +0inter /2),
Gn account for the photocreation process, (ρ)n is the total population of subband ‘n’ and the
broadenings are average values over k⊥ for the two wells. The three processes Pext , Pinter
and Pintra are incorporated in equations (5.14). We have just added the three contributions
to the time evolutions and neglected any interference between these processes. Finally, we
have neglected the coupling of the various coherences, related to states with different |k⊥ |
which all oscillate at the same frequency ω0 . In the following we adopt this set of equations
to roughly estimate the transfer probability for a population photocreated in a biased DQW.

5.4. Assisted scatterings and interwell transport


We consider below the part played by the assisted scatterings on the interwell transport
in coupled quantum wells. We describe two approaches: in section 5.4.1 we deal with
simple rate equations for the populations alone (0c = 0 in equations (5.14)), whereas the
contribution of the coherence terms to the transfer is accounted for in section 5.4.2.

5.4.1. Rate equations for the populations. As shown in section 3, the calculated
broadenings are often much less than the energy separation of the unperturbed levels,
justifying a posteriori the perturbative treatment of the defects. Here we consider only
interwell scatterings (Pinter processes) and associate the resulting broadening to the inverse
of the interwell transfer time. We suppose an initial population of independent carriers
photocreated in one excited stationary state of the DQW which is mainly centred in one well.
Elastic and inelastic processes are possible, as follows. Phonon emissions, on the one hand,
lead to an irreversible transfer of the carriers, since at low temperature the reverse process of
absorption is much less probable. Elastic interwell scatterings, on the other hand, should be
supplemented by inelastic intrawell processes, otherwise reverse scatterings take place and
lead the carriers back to the original well (with the same rate, since τ2→1 = τ1→2 for elastic
scatterings between two-dimensional parabolic subbands). We can show, however, that the
decay time for the photocreated population is roughly the interwell elastic one whenever
τinter /(2τintra ) > 1, which in fact is a not very restrictive condition in real structures, even
384 R Ferreira and G Bastard

though the intrawell relaxation is due to the emission of acoustical phonons (if optical
phonons cannot be emitted, see Ferreira 1992). These considerations are illustrated in the
following by two practical examples.
Rate equations for the populations have been used by Liu et al (1989) to interpret the
evolution with the applied field of the intensity of the various luminescence lines observed
in a biased ADQW under cw excitation. The relative intensities are related not only to
the radiative recombination and non-radiative captures, but also to the interwell transfer.
Another ingredient for the balance equations, not included in equation (5.14), is the ‘reverse’,
thermally activated transfer. If 1E = E2 − E1 > 0 and 1/τ2→1 is the transfer rate from
subband ‘2’ to subband ‘1’, then at finite temperature for thermalized populations the reverse
rate is 1/τ1→2 = exp{−1E/KB T }/τ2→1 . The two main results of Liu et al are: (i) the
crucial role played by the excitonic interactions in the transfer of the photocreated pairs;
(ii) the fact that the experimental results could be relatively well described by simple rate
equations only if a fast hole transfer is possible.
Rate equations for the populations have also been used to interpret the observed time
evolution of the luminescence lines of a DQW (Ferreira et al 1992). The structure was
biased so that the ground interband transition was the crossed one. In this case, after
quasiresonant excitation of the wide well, part of the photocreated pairs are converted
into crossed ones, leading to a final luminescence strongly dependent upon the excitation
intensity. Three main factors govern the time evolution of the photocreated population: (i)
the low recombination rate for the crossed pairs, as compared to the rate for direct pairs; (ii)
the build-up of a transient internal field opposite to the external bias; and (iii) the existence
of a reverse transfer for the crossed pairs. The opposite field appears because only the
electrons effectively transfer. This field is due to the spatial separation of the electrons
and holes forming the crossed pairs and is roughly proportional to their areal density (the
appearance of such a space charge dipole was first discussed by Sauer et al (1988)). The
resulting rate equations for the populations is nonlinear in laser intensity, since the transfer
rates and the total field inside the DQW depend on each other.

5.4.2. Role of the coherences. We now discuss some results from equations (5.14). We
disregard effects leading to a modification of the transfer, such as band filling or the built-up
of an internal field: the rates 0 are taken to be independent of time and the populations.
The time evolution of (ρ)n is then easily obtained when the ‘pumping’ terms are either
time independent or represent a very short (delta-like) photocreation process. For instance,
when G1 = 0 one can easily show that the steady-state population of the higher level is
(ρ)2 = G2 / 02T where 02T = 02 + 01 (0c + 0inter )/(0c + 01 ). Note that 02T = 02 + 0inter if
0c = 0, as expected. In the same way, when G1 = 0, G2 (t) = δ(t) and 0n = 0 one easily
obtains
(ρ)1 = 1 − (ρ)2 = α{1 − exp[−(20c + 0inter )t]} (5.15a)
α = (0c + 0inter )/(20c + 0inter ). (5.15b)
Note that at large times we have (ρ)2 /(ρ)1 = 0c /(0c + 0inter ): the dephasing-related
broadening tends to equally populate the two wells whereas Pinter processes lead to an
effective transfer. Of course, both populations vanish at large times if 0n 6= 0.
Let us finally consider the electric field dependence of the population decay times in the
large detuning limit. First, it is important to realize that this limit is easily realized in biased
DQWs. The detuning is 1e(F ) ≈ e(Fr − F )d where d = B + (L1 + L2 )/2 and Fr is the
resonance field. In practical units this is 1e(meV) ≈ 1F (kV cm−1 )d(Å)/100. If we take
1/ 0intra = 1 ps and λc = 2 meV (a large interwell coupling: ||  0intra ), for d = 100 Å
Tunnelling and relaxation in semiconductor double quantum wells 385

DQW the large detuning condition is easily verified since |1F | should only be much larger
than 4 kV cm−1 . For holes this condition is even more easily fulfilled. Second, we note
that 0intra is roughly field independent when the wells are not too wide. If, in addition, we
assume that τT ≈ 1/ 0intra in equation (5.13b), then one has 0c ∝ 1F −2 for large detunings.
This is similar to the dependence on 1F obtained in section 4 for the interwell scattering
rate 1/τinter (see equation (4.6) with 1E ≈ 1F ).
A more careful analysis of the two mechanisms for the case where the scatterings are
due to the ionized impurities shows that the one related to the intrawell processes dominates.
Also, a more complete work is needed to estimate its role for excitons and the importance
of other levels present in the structure, which perturb the two interacting subbands of the
ADQW far from their resonance.
In conclusion for section 5, we recall that a rather different description of the DQW
eigenstates is needed whenever the total broadening h̄/2τsc becomes larger than the energy
correction δE induced by the interwell tunnel coupling (see figure 11). In this case, one is
unable to distinguish between the levels of the whole ADQW and those of the decoupled
wells, a genuine feature of a motional-narrowing-like regime. In the two-level case, one has
δE = [(1e/2)2 + λ2c ]1/2 − 1e/2. Then, a motional-narrowing-like regime should appear at
resonance if the condition τsc  1 is fulfilled and far from resonance if |ω0 |  (1/2)τsc 2 .
Note that it may turn out that only the second inequality is fulfilled. This is due to the fact
that the external bias weakens the tunnel coupling but essentially does not influence the
intrawell scatterings. Finally, note that for actual biased DQWs the last inequality is in the
general case easily reachable at large detunings.

6. Conclusion

In this review we have considered the role of assisted scatterings on the interwell transfer
in double quantum wells. As pointed out in the introduction, these structures correspond to
one of the simplest systems exhibiting a tunnel coupling. In section 2 we have studied
the conduction and valence stationary eigenstates of defect free structures. As shown
in this section, the interwell coupling is more effective at resonance, and energetically
distant levels of the two wells are only weakly coupled by the tunnel interaction. We have
also discussed in detail the effects of an external bias applied along the growth axis on
the energy levels and stressed the possibility to experimentally enter or exit the regime
of resonant coupling. Finally, the excitonic states in double quantum wells were briefly
reviewed in section 3. The role of defects on the electronic states of double quantum wells
was discussed in sections 4 and 5. As shown in these sections, the presence of random
scatterings broadens the DQW levels, perturbs the coherent tunnel coupling and leads to
an irreversible transfer between the two wells. In section 4 we have considered (at the
Born approximation) the assisted scatterings by some mechanisms of interest and focused
on the case where the mean carrier’s position along the growth axis changes from one well
to its neighbour after scattering, leading thus to an effective interwell transfer. We have
shown, in particular, the existence of a large increase for the evaluated transfer rates as a
function of the applied field when the field approaches one interwell (elastic or inelastic)
resonance. Finally, we have stressed in section 4 the importance of a realistic description
of the valence and exciton eigenstates in the evaluation of the transfer rates. We have
considered in section 5 the evolution in time of a population photocreated in the double
quantum well. We have reviewed some theoretical models which account for the levels’
broadening and have, in particular, stressed the role of intrawell scatterings on the interwell
transfer.
386 R Ferreira and G Bastard

Acknowledgments

It is a pleasure to thank Drs C Delalande, A Filoramo and Ph Roussignol for their active
participation in the work described in this report. The Laboratoire de Physique de la Matière
Condensée de l’Ecole Normale Supérieure is ‘Unité Associée au CNRS (URA1437) et aux
Universités Paris 6 et Paris 7’.

References

Alexander M G, Rühle W W, Sauer R and Tsang W T 1989 Appl. Phys. Lett. 55 885
Alexandrou A, Kash J E, Mendez E E, Hong J M, Fukuzawa T and Hase Y 1990 Phys. Rev. B 42 9225
Altarelli M 1983 Phys. Rev. B 28 842
Ando T, Fowler A and Stern F 1982 Rev. Mod. Phys. 54 437
Andreani L C 1994 Confined Electrons and Photons: New Physics and Applications (Lecture Notes for the Int.
School of Material Science and Technology) ed E Burstein and C Weisbuch (New York: Plenum)
Andrews S R, Murray C M, Davies R A and Kerr T M 1988 Phys. Rev. B 37 8198
Baldereschi A and Lipari N O 1971 Phys. Rev. B 3 439
Bastard G 1981a Phys. Rev. B 24 5693
——1981b Phys. Rev. B 25 7584
——1988 Wave Mechanics Applied to Semiconductor Heterostructures (Les Ulis, France: Les Editions de Physique)
Bastard G, Brum J A and Ferreira R 1991 Solid State Phys. 44 229
Binder E, Kuhn T and Mahler G 1994 Phys. Rev. B 50 18 319
Brum J A and Oberli D 1993 J. Physique IV C5 191
Calecki D, Palmier J-F and Chomette A 1984 J. Phys. C: Solid State Phys. 17 5017
Capasso F, Mohammed K and Cho A Y 1986 IEEE Quantum Electron. QE22 1853
Chang L L, Esaki L and Tsu R 1974 Appl. Phys. Lett. 24 593
Chang L L, Mendez E E and Tejedor C (eds) 1990 Resonant Tunnelling in Semiconductors (NATO ASI Series B:
Physics, vol 277) (New York: Plenum)
Charbonneau S, Thewalt M L W, Koteles E S and Elman B 1988 Phys. Rev. B 38 6287
Chen Y J, Koteles E S, Elman B S and Armiento C A 1987 Phys. Rev. B 36 4562
Cohen-Tannoudji C, Dupont-Roc J and Grynberg G 1988 Processus d’interaction entre photons et atomes (Paris:
InterEditioins and Editions du CNRS)
Delalande C, Ziemelis U O, Bastard G, Voos M, Gossard A C and Wiegmann W 1984 Surf. Sci. 142 498
Deveaud B, Clerot F, Chomette A, Regreny A, Ferreira R, Bastard G and Sermage B 1990 Europhys. Lett. 11 367
Dignam M M and Sipe J E 1991 Phys. Rev. B 43 4085
Dingle R, Gossard A C and Wiegmann W 1975 Phys. Rev. Lett. 34 1327
Elliott R J 1963 Theory of Excitons I (Polarons and Excitons) ed C G Kuper and G D Whitfield (New York:
Plenum)
Eisenstein J P, Boebinger G S, Pfeiffer L N, West K W and Song He 1992a Phys. Rev. Lett. 68 1383
Eisenstein J P, Pfeiffer L N and West K W 1992b Phys. Rev. Lett. 69 3804
Ensslin K, Heitmann D, Dobers M, von Klitzing K and Ploog K 1989 Phys. Rev. B 39 11 179
Esaki L and Tsu R IBM 1970 J. Res. Dev. 14 61
Ferreira R 1992 PhD Thesis Université Pierre et Marie Curie, Paris, France
——1996 Solid-State Electron. 40 425
Ferreira R, Delalande C, Liu H W, Bastard G, Etienne B and Palmier J-F 1990 Phys. Rev. B 42 9170
Ferreira R, Rolland P, Roussignol Ph, Delalande C, Vinattieri A, Carraresi L, Colloci M, Roy N, Sermage B and
Palmier J-F 1992 Phys. Rev. B 45 11 782
Ferreira R and Bastard G 1989a Phys. Rev. B 40 1074
——1989b Europhys. Lett. 10 279
Filoramo A, Ferreira R and Roussignol Ph 1996 to be published
Fox A M, Miller D A B, Livescu G, Cunningham J E and Jan W Y 1991 Phys. Rev. B 44 6231
Galbraith I and Duggan G 1989 Phys. Rev. B 40 5515
Goldoni G and Fasolino A 1992 Phys. Rev. Lett. 69 2567
Golub J E, Kash K, Harbison J P and Florez L T 1990 Phys. Rev. B 41 8564
Gurvitz S A, Bar-Joseph I and Deveaud B 1991 Phys. Rev. B 43 14 703
Haug H and Koch S W 1990 Quantum Theory of the Optical and Electronic Properties of Semiconductors
(Singapore: World Scientific)
Tunnelling and relaxation in semiconductor double quantum wells 387

Islam M N, Hillman R L, Miller D A B, Chemla D S, Gossard A C and English J H 1987 Appl. Phys. Lett. 50
1098
Jones W and March N M 1973 Theoretical Solid State Physics Volume 2: Non equilibrium and Disorder (London:
Wiley)
Kamizato T and Matsuura M 1989 Phys. Rev. B 40 8378
Kasarinov R F and Suris R A 1972 Sov. Phys.–Semicond. 6 120
Knox R S 1963 Solid State Physics suppl. 5 Theory of Excitons ed F Seitz and D Turnbull (New York: Academic)
Lary J, Goodnick S M, Lugli P, Oberli D and Shah J 1989 Solid-State Electron. 32 1283
Lawrence I, Rühle W W, Feuillet G, Tuffigo H, Mariette H, Bodin C and Cibert J 1993 J. Physique IV C5 405
Leavitt R P and Little J W 1990 Phys. Rev. B 42 11 774
Lefebvre P, Bonnel P, Gil B and Mathieu H 1991 Phys. Rev. B 44 5635
Leo K, Shah J, Gordon J P, Damen T C, Miller D A B, Tu C W and Cunningham J E 1990 Phys. Rev. B 42 7065
Leo K, Shah J, Göbel E O, Damen T C, Schmitt-Rink S, Schäfer W and Köhler K 1991 Phys. Rev. Lett. 66 201
Liu H W, Ferreira R, Bastard G, Delalande C, Palmier J-F and Etienne B 1989 Appl. Phys. Lett. 54 2082
Livescu G, Fox A M, Miller D A B, Sizer T, Knox W H, Cunningham J E, Gossard A C and English J H 1990
Semicond. Sci. Technol. 5 549
Luttinger J M 1956 Phys. Rev. 102 1030
Luttinger J M and Kohn W 1955 Phys. Rev. 97 869
Manoharan H C, Suen Y W, Santos M B and Shayegan M 1996 Surf. Sci. 361 106
Michl F C, Winkler R and Rössler U 1995 Nuovo Cimento D 17 1613
Miller D A B, Chemla D S, Damen T C, Gossard A C, Wiegmann W, Hood T C and Burrus A C 1985 Phys. Rev.
B 32 1043
Nelson D F, Miller R C, Kleinman D A and Gossard A C 1986 Phys. Rev. B 34 8671
Nido M, Alexander M G W, Rühle W W and Köhler K 1991 Phys. Rev. B 43 1839
Nido M, Alexander M G W, Rühle W W, Schweizer T and Köhler K 1990 Appl. Phys. Lett. 56 355
Norris T B, Vodjdani N, Vinter B, Costard E and Böckenhoff E 1991 Phys. Rev. B 43 1867
Norris T B, Vodjdani N, Vinter B, Weisbuch C and Mourou G A 1989 Phys. Rev. B 40 1392
Oberli D Y, Shah J, Damen T C, Tu C W, Chang T Y, Miller D A B, Henry J E, Kopf R F, Sauer N and
DiGiovanni A E 1989 Phys. Rev. B 40 3028
Oberli D Y, Shah J, Damen T C, Kuo J M, Henry J E, Lary J and Goodnick S M 1990 Appl. Phys. Lett. 56 1239
Pikus G E and Titkov A N 1984 Modern Problems in Condensed Matter Sciences: Optical Orientation ed F Meier
and B P Zakharchenya (Amsterdam: North-Holland)
Ridley B K 1988 Quantum Processes in Semiconductors (Oxford: Clarendon)
Roussignol Ph, Vinattieri A, Carraresi L, Colocci M and Fasolino A 1991 Phys. Rev. B 44 8873
Sauer R, Thonke K and Tsang W T 1988 Phys. Rev. Lett. 61 609
Sawaki N, Suzuki M, Okuno E, Goto H, Akasaki I, Kano H, Tanaka Y and Hashimoto M 1988 Solid-State Electron.
31 351
Schiff L I 1955 Quantum Mechanics (New York: McGraw-Hill)
Schmitt-Rink K, Chemla D S and Miller D A B 1989 Adv. Phys. 38 89
Shah J 1992 Hot Carriers in Semiconductor Nanostructures. Physics and Applications (London: Academic)
Smith D L and Mailhiot C 1990 Rev. Mod. Phys. 62 173
Tada T, Yamaguchi A, Ninomiya T, Uchiki H, Kobayashi T and Yao T 1988 J. Appl. Phys. 63 5491
Tsu R and Döhler G 1975 Phys. Rev. B 12 680
Twardowski A and Hermann C 1987 Phys. Rev. B 35 8144
Vas’ko F T and Raichev O E 1995 Sov. Phys.–JETP 81 1146
Waschke C, Roskos H G, Schwedler R, Leo K, Jurz H and Köhler K 1993 Phys. Rev. Lett. 70 3319
White S R and Sham L J 1981 Phys. Rev. Lett. 47 879
Winkler R 1995 Phys. Rev. B 51 14 395

You might also like