Microscopic Theory of Semiconductor-Based Optoelectronic Devices

Microscopic theory of semiconductor-based optoelectronic devices
This article has been downloaded from IOPscience. Please scroll down to see the full text article.
2005 Rep. Prog. Phys. 68 2533
(http://iopscience.iop.org/0034-4885/68/11/R02)
Download details:
IP Address: 129.217.188.233
The article was downloaded on 18/04/2013 at 13:26
Please note that terms and conditions apply.
View the table of contents for this issue, or go to the journal homepage for more
Home Search Collections Journals About Contact us My IOPscience
INSTITUTE OF PHYSICS PUBLISHING REPORTS ON PROGRESS IN PHYSICS
Rep. Prog. Phys. 68 (2005) 25332571 doi:10.1088/0034-4885/68/11/R02
Microscopic theory of semiconductor-based
optoelectronic devices
Rita C Iotti and Fausto Rossi
Istituto Nazionale per la Fisica della Materia (INFM) and Dipartimento di Fisica,
Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Torino, Italy
Received 19 May 2005, in nal form 2 August 2005
Published 7 September 2005
Online at stacks.iop.org/RoPP/68/2533
Abstract
Since the seminal paper by Esaki and Tsu, semiconductor-based nanometric heterostructures
have been the subject of impressive theoretical and experimental activity due to their high
potential impact in both fundamental research and device technology. The steady scaling
down of typical space and time scales in quantum optoelectronic systems inevitably leads to a
regime in which the validity of the traditional Boltzmann transport theory cannot be taken for
granted and a more general quantum-transport description is imperative.
In this paper, we shall reviewstate-of-the-art approaches used in the theoretical modelling,
design and optimization of optoelectronic quantum devices. The primary goal is to provide
a cohesive treatment of basic quantum-transport effects, able to explain and predict the
performances of new-generation semiconductor devices. With this aim, we shall review
and discuss a fully three-dimensional microscopic treatment of time-dependent as well as
steady-state quantum-transport phenomena, based on the density matrix formalism. This will
allow us to introduce in a quite natural way the separation between coherent and incoherent
processes. Starting with this general theoretical framework, we shall analyse two different
types of quantum devices, namely periodically repeated structures and quantum systems with
open boundaries.
For devices within the rst class, we will show how a proper use of periodic boundary
conditions allows us to reproduce and predict their currentvoltage characteristics without
resorting to phenomenological parameters. For the second class of devices, we will address
the relevant issue of a quantum treatment of charge transport in systems with open boundaries
(electrical contacts) when studying and simulating an at least two-terminal device.
0034-4885/05/112533+39$90.00 2005 IOP Publishing Ltd Printed in the UK 2533
2534 R C Iotti and F Rossi
Contents
Page
1. Introduction 2535
2. Fundamentals 2536
3. Theoretical background 2538
3.1. Physical system 2538
3.2. Density-matrix formalism 2539
3.3. Semiclassical limit 2541
4. Modelling of periodically-repeated optoelectronic devices 2542
4.1. Quantum-cascade devices 2542
4.2. Three-dimensional microscopic description 2544
4.3. Quantum-transport approach 2545
4.4. Simulation strategy 2546
4.5. Simulation of mid-infrared quantum-cascade laser 2547
4.5.1. Partially phenomenological approach 2547
4.5.2. Global-simulation scheme 2549
4.5.3. Quantum-transport approach 2553
4.5.4. Active-region cavity-mode coupling 2555
4.6. Towards terahertz sources 2557
5. Modelling of quantum devices with open boundaries 2558
5.1. Wigner-function formalism 2558
5.1.1. Conventional Wigner-function description 2560
5.1.2. Generalized Wigner-function approach 2562
5.2. A closed-system paradigm 2564
5.2.1. Semiclassical case 2565
5.2.2. Generalization to the quantum-mechanical case 2568
6. Summary 2569
Acknowledgments 2570
References 2570
Microscopic theory of optoelectronic devices 2535
1. Introduction
Following the path started by the early proposal by Esaki and Tsu [1], articially tailored
semiconducting heterostructures at the nanoscale now form the leading edge of device
technology. The design of modern optoelectronic devices, in fact, heavily exploits the
principles of band-gap engineering [2], achieved by conning charge carriers in spatial
regions comparable to their de Broglie wavelengths. This, together with the continuous
progress in microelectronics technology [3], pushes device miniaturization towards limits
where the traditional Boltzmann transport theory [4] can no longer be applied, and more
rigorous quantum-transport approaches are imperative [5]. However, in spite of the quantum-
mechanical nature of carrier dynamics in the core region of typical nanostructured devices
like semiconductor superlattices, double-barrier structures and quantum dotsthe overall
behaviour of such quantum systems is often the result of a non-trivial interplay between
phase coherence and energy relaxation/dephasing [6]. It follows that a proper treatment of
such novel devices would require a theoretical modelling able to account for both coherent
and incoherenti.e. phase-breakingprocesses on an equal footing, within a many-body
picture.
The idealized behaviour of a so-called quantum device [7] is usually described in terms
of the elementary physical picture of the square-well potential and/or by analogy with a simple
quantum-mechanical n-level system. For a quantitative investigation of realistic state-of-the-
art quantumoptoelectronic devices, however, two features strongly inuence and modify such a
simplied scenario: (i) the intrinsic many-body nature of the carrier systemunder investigation
and (ii) the potential coupling of the electronic subsystemof interest with a variety of interaction
mechanisms, including the presence of spatial boundaries [810].
The wide family of quantum devices can be divided into two main classes: a rst one
comprising semiconductor devices characterized by a genuine quantum-mechanical behaviour
of their carrier subsystemand a second one made up of low-dimensional nanostructures whose
transport dynamics may be safely treated within the semiclassical picture.
The devices within the rst classcharacterized by a weak coupling of the carrier
subsystem with the host materialare natural candidates for the implementation of quantum
information/computation processing [11]. These include, in particular, semiconductor
quantum-dot structures [12], for which all-optical implementations have recently been
proposed [1315]. In this case, the typical quantum-mechanical carrier dynamics is only
weakly disturbed by decoherence processes; therefore, the latter are usually described in terms
of extremely simplied models.
In contrast, the second class of quantum devicesin spite of the partially discrete
energy spectrum due to spatial carrier connementexhibit a carrier dynamics which
can still be described via a semiclassical scattering picture. Such optoelectronic devices
include multi-quantum-well (MQW) and superlattice structures, like quantum-cascade lasers
(QCLs) [16, 17]. In this case, we deal with a strong interplay between coherent dynamics and
energy-relaxation/dephasing processes; it follows that for a quantitative description of such
non-trivial coherence/dissipation coupling the latter need to be treated via fully microscopic
models.
In the present review we shall primarily focus on quantum devices within the
second class described above, providing a comprehensive microscopic theory of charge
transport in semiconductor nanostructures based on the well-known density-matrix approach.
Within this formalism two different strategies are commonly employed: (i) the quantum-
kinetic treatment [6] employed in this paper and (ii) the Liouvillevon Neumann-equation
description [18].
It is worth mentioning that an equivalent and alternative approach to the density-matrix
formalism is given by the non-equilibrium Greens function technique; the latter can be
regarded as an extension of the well-known equilibrium or zero-temperature Greens function
theory to non-equilibrium regimes, introduced in the sixties by Kadanoff and Baym [19]
and Keldysh [20]. An introduction to the theory of non-equilibrium Greens functions with
applications to many problems in transport and optics of semiconductors can be found in the
book by Haug and Jauho [21]. In spite of the fact that the density matrix can be regarded as
an equal-time Greens function, their equations of motionand thus the two corresponding
perturbative expansionsare totally different. The density-matrix formalism is particularly
suited to the study of single-time quantities (e.g. current density and mean kinetic energy)
and provides a more direct physical interpretation; in contrast, due to its two-time nature,
the Greens function formalism is particularly useful for the investigation of time correlation
phenomena (e.g. density and velocity autocorrelation functions).
As anticipated, the aim of this paper is to review a general theoretical framework able
to explain and predict the performances of different classes of state-of-the-art quantum
optoelectronic devices. To this end, we shall present and discuss a fully three-dimensional
microscopic treatment of time-dependent as well as steady-state quantum-transport phenomena
based on the density-matrix formalism. This will allowus to introduce in a quite natural way the
separation between coherent and incoherenti.e. phase-breakingprocesses. Moreover,
the conventional Boltzmann-transport theory is simply obtained as the semiclassical limit
of the proposed density-matrix formalism.
Based on such a general theoretical framework, we shall discuss two different types of
semiconductor-based quantum devices, namely periodically repeated structures (like QCLs)
and quantum systems with open boundaries (like resonant-tunnelling diodes). For the rst
class of devices we shall show how a proper use of periodic boundary conditions allows
us to reproduce and predict the currentvoltage characteristics of state-of-the-art mid- and
far-infrared QCLs, without resorting to any phenomenological parameter. For the second
class of devices we shall discuss the intrinsic limitations of the conventional Wigner-function
formalism used to describe in a microscopic fashion injection/loss processes in systems with
open boundaries, like double-barrier structures.
This paper is organized as follows. We start in section 2 with a summary of some
fundamental concepts and methods in opto-electronic device modelling. In section 3 we
shall review and discuss the general theoretical framework used for the microscopic modelling
of various classes of semiconductor-based quantum devices. In particular, we shall present
our fully three-dimensional density-matrix formalism as well as its semiclassical counterpart,
i.e. the conventional Boltzmann transport scenario. Sections 4 and 5 will present applications
of this density-matrix formalism to two different classes of optoelectronic quantum devices:
periodically repeated structures (e.g. mid- and far-infrared QCLs) and quantum systems with
open boundaries (e.g. multi-barrier structures), respectively. In section 6 we shall nally
summarize and draw some conclusions.
2. Fundamentals
The modelling of electron transport in semiconductor quantum devices may be
performed in terms of various approaches, each corresponding to a different degree of
accuracy/approximation. A rst level of description is by analogy with n-level atomic
systems and therefore in terms of global quantities. This approach is often grounded on
purely macroscopic models, the only relevant physical quantities being the various carrier
concentrations n
, within each subband .

1
The time evolution of n
can be described by a set

of phenomenological rate equations of the form
dn
dt
=
_
g
_
i/l
+
(w
)
s
. (2.1)
This is the sum of the injection/loss (i/l) contribution in the generic stage of the deviceg
and
1
being, respectively, the carrier injection and loss rates for level and the interlevel-
scattering term (s)w
denoting phenomenological scattering rates connecting levels

and
.
Amore rened(i.e. kinetic) level of descriptionis interms of the single-particle distribution
function f
over the electronic states of the device. The equation of motion for f
, governing
carrier transport and relaxation phenomena, may be schematically written as:
d
dt
f
=
d
dt
f
scat
+
d
dt
f
res
. (2.2)
Here, the rst termdescribes scattering dynamics within the device active region and is usually
treated at a kinetic level via a Boltzmann-like collision operator of the form:
d
dt
f
scat
=
[P
], (2.3)
where P
is the total (i.e. summed over all relevant interaction mechanisms) scattering rate
from state
to state .
The second term in (2.2) accounts for the open character of the system and describes
injection/loss contributions from/to the (at least two) external carrier reservoirs. These
processes are usually modelled by a relaxation-time-like term of the form [22]:
d
dt
f
res
=
_
f
_
= G
, (2.4)
where
1
may be interpreted as the device transit time for an electron in state , while f
is the single-particle carrier distribution in the external reservoirs. The latter may correspond
to the distinct quasi-equilibrium distributions in the left and right chemical potentials or may
describe a generic non-equilibrium distribution within the external reservoirs.
The kinetic approach conrms that the physical parameters entering equation (2.1)
are unavoidably global or macroscopic quantities, i.e. their microscopic evaluation would
require the knowledge of the carrier distribution function over the whole three-dimensional
k-space. Indeed, from a microscopic point of view and for a quasi-two-dimensional structure,
the carrier concentration/density in a given subband is the sum/integral over the in-plane
wavevector k of the single particle distribution function f
= f
k
:
n
k
f
k
. (2.5)
Let us reconsider the Boltzmann equation (2.3) written in our specic representation = k,
i.e.
d
dt
f
k
=
[P
k,
,k
f
k
]. (2.6)
1
Such macroscopic description is analogous to the well-established hydrodynamic and driftdiffusion models for
conventional microelectronic device simulations [4]. The key ingredients, i.e. total carrier density, average drift
velocity and average kinetic energy, correspond to the various momenta of the single-particle distribution function.
By summing the above three-dimensional transport equation over the in-plane wavevector k,
we get
dn
dt
=
(w
) (2.7)
with
w
k,k
P
k,
k
f
. (2.8)
The effective Boltzmann collision term in (2.7) has indeed the same form as the one in
equation (2.1); however, as shown by equation (2.8), the effective interlevel scattering rates
w
are obtained by averaging the microscopic transition probabilities P

k,
over the initial

and nal in-plane carrier wavevectors k
and k. It follows that their values are strongly affected

by the actual nature and shape of the unknown in-plane carrier distribution.
For this reason, the rate-equation scheme in (2.1) can only operate as an a posteriori
tting procedure. Indeed, rather than within a multi-level system, the electron dynamics in
semiconductor-based quantum devices generally occurs within a multi-subband structure, and
the existence of transverse, i.e. in-plane, degrees of freedom should be properly taken into
account.
In view of their Boltzmann-like structures, the kinetic equations for f
are commonly
sampled by means of the Monte Carlo method (MC) [4]. One of the main advantages of
this technique is that it allows us to include, on an equal footing, a large variety of scattering
mechanisms. As a general remark, we stress that, in spite of the kinetic nature of the scattering
dynamics in (2.3), the contribution in (2.4) still treats carrier injection and loss processes on a
partially phenomenological level.
State-of-the-art semiconductor-based quantum devices, however, often present an active
region which consists of a periodic repetition of N identical stages. This is the case, for
example, of QCLs [16] and quantum-well infrared photodetectors (QWIPs) [23]. The ending
stages of the sequence are the only ones coupled to the external carrier reservoirs (the
contacts), while each internal stage actually behaves as an injector/collector of carriers for
the preceding/following one. When N is large (typical values are in the range 1070) the
charge dynamics, and therefore the device operation, is mainly determined by inter-stage
coupling rather than by nite-size effects. In this case, an approach alternative to the one of
equation (2.2) may be adopted. It consists of imposing proper boundary conditions that allow
us to close the circuit and therefore to employ a particle-conserving simulation strategy.
3. Theoretical background
3.1. Physical system
To provide a general formulation of quantumcharge transport in semiconductor nanostructures,
let us consider a generic carrier gas within a semiconductor crystal in the presence of
electromagnetic elds. The corresponding Hamiltonian can be schematically written as
H = H
+ H
= (H
c
+ H
p
) +
_
s
H
s
_
. (3.1)
The single-particle term H
is the sum of the non-interacting carrier and phonon Hamiltonians

H
c
and H
p
. The many-body contribution H
may include various interaction mechanisms, e.g.

carriercarrier, carrierphonon, carrierlight, etc.
More specically, the single-particle Hamiltonian H
c
describes the non-interacting
carrier system within the effective three-dimensional potential prole of our quantum device.
Denoting by
(r) = r| (3.2)
the wavefunction of the single-particle state and by
the corresponding energy, the non-

interacting-carrier Hamiltonian can be written as
H
c
=
. (3.3)
Here, the usual second-quantization picture in terms of creation (c
) and destruction (c
)
operators has been employed. The generic label denotes, in general, a suitable set of
discrete and/or continuous quantum numbers; for the case of two-dimensional semiconductor
nanostructures considered in this paper this corresponds to a partially discrete index along the
so-called growth direction, where the electron motion is affected by the effective nanostructure
potential prole
2
, plus an in-plane two-dimensional wavevector.
The explicit form of the generic many-body contribution H
s
in (3.1) can be found in [6]; it
depends on the particular interaction mechanisms considered and typically involves additional
degrees of freedom like phonons, photons, plasmons, etc. From the point of view of the
carrier subsystem, such many-body contributions can be divided into two different classes:
single-particle termsincluding carrierphonon and carrierlight interactionsand two-body
Coulomb interactions.
3.2. Density-matrix formalism
In order to investigate transport as well as energy-relaxation phenomena, our aim is to study
the time evolution of single-particle quantities like total carrier density, mean kinetic energy,
charge current, etc. In general, such quantities correspond to a suitable ensemble average of a
generic single-particle operator A, i.e.
A =
_
2
A
2
c
1
c
2
_
=
2
A
2
c
1
c
2
. (3.4)
By introducing the single-particle quantity
2
=
_
c
2
c
1
_
, (3.5)
we get
A =
2
A
1
= tr(A). (3.6)
The quantity in equation (3.5) is called single-particle density matrix [6]; this is the key
ingredient for the study of single-particle properties. Its diagonal elements (f
=
c
) correspond to the usual distribution functions of the semiclassical Boltzmann theory [4]
(see below), while the non-diagonal terms (
1
=
2
) describe the degree of quantumcoherence
(or electronic polarization) between states
1
and
2
.
2
The effective potential along the growth direction is, in general, the sum of the band discontinuity prole plus
the Hartree contribution due to the charge spatial distribution; it follows that a proper treatment would require a
self-consistent solution of the Schr odinger and Poisson equations. However, when the doping level and carrier
concentrations in the quantum devices are relatively low, Hartree corrections to the potential prole may be neglected.
In the Heisenberg picture, the time evolution of A is fully dictated by the time evolution
of the density matrix . Starting from the Heisenberg equations of motion for the destruction
operators c
, i.e.
d
dt
c
=
1
i h
[c
, H] =
1
i h
[c
, H
] +
1
i h
[c
, H
] =
d
dt
c
+
d
dt
c
, (3.7)
it is possible to derive a set of equations of motion for the density-matrix elements
2
often
called semiconductor Bloch equations (SBE) [6, 24]whose general structure is given by
d
dt
2
=
d
dt
+
d
dt
. (3.8)
The time evolution induced by the single-particle Hamiltonian H
can be evaluated exactly

yielding
d
dt
=
1
i h
(
2
)
2
. (3.9)
The resulting dynamics is diagonal within our
1
2
representation, i.e. each density-matrix
element evolves independently according to
2
(t ) =
2
(t
) exp
_
(
2
)(t t
)
i h
_
. (3.10)
In particular, the diagonal elements are not affected by the free dynamics, while the non-
diagonal density-matrix elements rotate with different frequencies, given by the energy
difference between states
1
and
2
. As we shall see, this non-diagonal dynamicsabsent in
the semiclassical (diagonal) limitgives rise to a variety of coherent phenomena in quantum
optoelectronic devices.
In contrast, the time evolution due to the many-body Hamiltonian H
=

s
H
s
involves,
in general, phonon-assisted as well as higher-order density matrices. Thus, in order to close
our set of equations of motion (with respect to the single-particle density matrix in (3.5))
approximations are needed. A detailed discussion of the various approximation schemes
based on a dynamical expansion in powers of the interaction Hamiltonian H
is given in [6].
In particular, the mean-eld approximation together with the Markov limit allows us to derive
a set of closed equations of motion still local in time. For a number of interaction mechanisms,
including carrierphonon as well as carriercarrier scattering, their second-order contributions
to the time evolution can be cast into the general form:
d
dt
= F[]
2
. (3.11)
Although F[] is, in general, a non-linear functional of , it can be written as
F[]
2
=
2
,
2
, (3.12)
where is itself a function of . More specically
F[]
2
=
1
2
[(
)
in
2
+ (
)
in
1
]
1
2
[

out
2
+

out
1
]. (3.13)
Here, the matrices can be regarded as generalized in- and out-scattering rates, in analogy
with the Boltzmann collision term of the semiclassical theory [4]. They are, in general,
-dependent complex quantities; their real parts describe incoherent phenomena, like energy-
relaxation and dephasing processes, while their imaginary parts lead to modications of the
free coherent evolution in (3.10) through non-trivial energy-renormalization effects. Their
explicit form for the case of carrierphonon as well as carriercarrier interactions are given
in [25].
By combining equations (3.9), (3.11) and (3.12), the SBE in (3.8) can be written as
d
dt
2
=
2
L
2
,
2
, (3.14)
where the effective Liouville operator
L
2
,
2
=
1
i h
(
2
)
2
,
2
+
2
,
2
(3.15)
is the sum of two terms: free single-particle evolution plus scattering-induced energy-
relaxation/dephasing dynamics.
In the low-carrier-density limit (i.e. |
2
| 1), carriercarrier scattering as well as Pauli-
blocking phenomena vanish, and the effective Liouville superoperator L becomes linear, i.e.
-independent. In this case, the steady-state behaviour of the system can be readily obtained
by solving the linear homogeneous problem:
2
L
2
,
2
= 0. (3.16)
3.3. Semiclassical limit
As anticipated, in the semiclassical limit the non-diagonal elements of the density matrix
are neglected, i.e.
2
= f
2
. (3.17)
In this limit the coherent free evolution described by equation (3.9) reduces to
d
dt
f
=
d
dt
= 0, (3.18)
i.e. in the absence of many-body scattering dynamics the semiclassical distribution function
f
is constant in time.
Let us now come to the many-body contribution in equation (3.13); in the semiclassical
limit (3.17) it reduces to
d
dt
f
= (1 f
)
in
out
. (3.19)
The explicit form of the scattering matrices given in [25] in the semiclassical limit reduces
to
in
,
out
(1 f
), (3.20)
where
P
s
P
s
(3.21)
are total (summed over all scattering mechanisms s) transition probabilities from state
to
state as given by Fermis golden rule:
P
s
=
_
2
h
|, |H
s
|
|
2
(
)
_
,
. (3.22)
Here . . .
,
denotes a suitable statistical average over the initial and nal set of auxiliary
coordinates and
, e.g. phonons, photons, partner carriers, etc.

By combining equations (3.18), (3.19) and (3.20) we then recover the usual Boltzmann
transport equation (2.3):
d
dt
f
[P
], (3.23)
where
P
= (1 f
)P
(3.24)
is the total transition probability P
in (3.21) weighted by the Pauli factor of the nal state.

The Boltzmann equation in (3.23) can then be regarded as the semiclassical limit of the
SBE in (3.14). This clearly shows that in the semiclassical limit the above quantum scattering
model reduces to the conventional Boltzmann theory. As we can see from equation (3.13),
also in the quantumcase the total effect is the result of a balance between in- and out-scattering
contributions. In particular, for the case of quasi-elastic processes it is possible to show [6]
that, in analogy with the semiclassical case, the two in and out contributions strongly cancel,
thus giving rise to long dephasing times.
As for the quantum case, in the low-density limit (f
1) the Boltzmann transport

equation in (3.23) becomes linear; this allows us to obtain the steady-state solution by solving
the following homogeneous linear problem:
= 0 (3.25)
with
P
= P
. (3.26)
The above Boltzmann scattering operator P can be regarded as the semiclassical counterpart
of the effective Liouville superoperator L in equation (3.15).
4. Modelling of periodically-repeated optoelectronic devices
State-of-the-art semiconductor-based optoelectronic devices at the nanoscale are often based
on periodically-repeated core regions. This strategy is exploited, for example, to increase
either the photo-current in detectors such as QWIPs or the emitted power in coherent sources
like QCLs. As shown in this section, this feature turns out to be crucial also in the proper
modelling of these devices, since it allows us to consistently close the circuit without the need
of any phenomenological parameter accounting for the coupling with the external contacts.
4.1. Quantum-cascade devices
Unipolar coherent-light sources, such as QCLs [16, 17], are complex devices whose core is a
MQWstructure made upof a periodic repetitionof identical stages of active regions sandwiched
between electron-injecting and -collecting regions. When a proper bias is applied, an electron
cascade along the subsequent quantized-level energy staircase takes place. Due to the potential
of band-gap engineering, several successful designs have been proposed up to nowfor both the
active region and the injector part of the structure. To better present the theoretical approach
reviewed in this paper, it can be of some help just to recall the main features which are common
to most of these devices, regardless of the specicities of each design.
active region injector/collector
= 1
= -1
= 0
(+1,3)
E
D
B,C
A
1
2
3
Figure 1. Schematic representation of the conduction-band prole along the growth direction
for the diagonal-conguration QCL structure of [26]. The MQW is biased by electric eld of
48 kVcm
1
. The levels = 1, 2, 3 and = A, B, C, D, E in the active and collector regions
of the simulated stage ( = 0) are also plotted together with the corresponding charge densities.
The replica of level 3 in the following stage = +1 is shown for clarity. The separation between
subbands 1 and 2 matches the optical phonon energy ( h
LO
36 meV) and the stage length is
453 (the gure is in scale). Adapted from [27].
As an example, gure 1 shows the conduction band prole along the growth direction
for a prototypical QCL emitting in the mid-infrared (mid-IR) portion of the electromagnetic
spectrum [26, 27]. The gure presents the generic stage = 0, plus a portion of both the
neighbouring left and right stages. Also shown are the square moduli of the wavefunctions
of the conned states in the active region ( = 1, 2, 3) and in the collector/injector ( = A,
B, C, D, E), evaluated within an effective-mass framework. Indeed, taking into account the
fully three-dimensional nature of the single-particle Hamiltonian in (3.3) as well as of the
corresponding eigenfunctions
(r) r|, (4.1)

the single-particle states
(r) in this structure are also labelled by the in-plane wavevector k.

In the device of gure 1, electrons are injected by resonant tunnelling into subband 3 of
the generic stage ( = 0) and can then relax into lower-energy states in subband 2 and 1. By a
proper choice of the geometry, optical matrix elements and scattering rates can be optimized so
that a population-inversion condition is established between subbands 3 and 2, corresponding
to the lasing transition. Once a photon has been emitted in this stage, the electron is drifted off
into the collector whence it can be recycled into the downstream neighbouring stage = 1 to
emit a second photon and so on. It follows that these are intrinsicallyhigh-power devices, which
can be designed to emit over a wide range of frequencies, just by modifying the geometrical
parameters while keeping the same constituent materials.
Since their appearance, QCLs have been the subject of a rapid experimental development
in emission wavelength, lasing threshold, output power and operating temperature [17]. This
has stimulated considerable theoretical interest, mainly motivated by the desire to improve
the device performances by optimizing the semiconductor-heterostructure design. Various
theoretical models [28] have been proposed to describe gain spectra and characteristics of
QCLs. However, these approaches generally resort to macroscopic rate-equations to describe
the carrier dynamics, whose application does not have a true justication [29]. In contrast,
for a detailed understanding of the basic physical processes involved in the operation of these
devices, a fully three-dimensional description is needed. Besides, a completely microscopic
analysis would indeed serve as a predictive tool for the evaluation of newdesigns and strategies.
In the following we shall review and discuss such a kinetic three-dimensional Boltzmann-like
treatment and its extension into a fully quantum-mechanical approach.
4.2. Three-dimensional microscopic description
To provide a quantitative insight into the details of the carrier dynamics in these unipolar
devicesenergy-relaxation as well as dephasing processesa proper treatment of the fully
three-dimensional nature of the problem is imperative. As originally proposed in [30] and
generalized in [31] and [32], this can be done within the kinetic Boltzmann-like approach
reviewed in section 3. In this scheme, the fundamental ingredient is the carrier distribution
function f
. In our MQW structure, is a global quantum number labelling single-particle

states and comprising the stage index , the subband index within each stage and the in-plane
wavevector k. The time evolution of the carrier distribution function f
is then governed by
the Boltzmann transport equation (3.23).
Although equation (3.23) corresponds to the more complete description of the problemand
allows us to kinetically study electron dynamics within the full core region of the device, semi-
phenomenological models, inbetweenequations (2.1) and(3.23), have alsobeenproposed[30].
They can still provide valuable information on the microscopic carrier dynamics with the
advantage of being much less computer-time consuming. This is valid, in particular, when the
region of interest corresponds to a portion of the whole structure, e.g. the bare active region
of the device. Limiting the kinetic treatment to this subset = { = 0; = 1, 2, 3; k} of the
multisubband structure of gure 1, the kinetic equation in (3.23) can be rewritten as
d
dt
f

=
_
g

f

_
i/l
+
[P

f

P

f

]. (4.2)
Here, the rst two terms describe injection and loss of carriers with in-plane wavevector k in
miniband , while the last one accounts for in- and out-scattering processes between states k
and
. Contrary to the macroscopic model in equation (2.1), this formulation provides a

description of the intra- and inter-miniband scattering, in the subset, in terms of microscopic
ingredients only. However, the injection/loss contributions in equation (4.2), coupling the
active region with the injector/collector, are treated on a partially phenomenological level.
The g and functions are dened within the same kinetic picture k and are adjusted to
reproduce the experimentally measured current density across the device.
The kinetic description proposed in equation (4.2) can be quite useful to address the
microscopic nature of the hot-carrier relaxation within a portion of the structure, e.g. the
device active region [30]. However, due to the presence of free-parameters coupling this region
of interest to the rest of the device, it does not allow us to address the nature of the physical
mechanisms governing charge transport through injector/active-region/collector interfaces. To
this end, the partially phenomenological model has to be replaced by the fully microscopic
description of the whole MQW core structure in equation (3.23).
Since real devices are made up of several (generally 3070) identical stages, the number
of coupled equations one has to deal with may be prohibitively large. In this respect, the
periodicity of the real structure allows us to reasonably assume that
f
k
= f
k
, (4.3)
i.e. the carrier distribution function does not depend on the stage index. Under this assumption,
equation (3.23) can then be rewritten as
d
dt
f
k
=
[P
k,
,k
f
k
] +
=
P
k,
=
P
,k
f
k
.
(4.4)
Here, the rst term accounts for all intra-stage scattering dynamics, while the second and third
ones correspond, respectively, to injection and loss contributions, coupling this stage with
the neighbouring ones. They are the microscopic equivalent of the g and functions of the
partially phenomenological model in equation (4.2). The periodic boundary conditions (4.3)
then allow us to close the circuit without resorting to phenomenological parameters. To
properly model charge transport in realistic devices, the interstage ( =
) scattering can be
safely limited to nearest-neighbour coupling (
= 1).
4.3. Quantum-transport approach
So far no quantum mechanical effects, like coherent resonant tunnelling between adjacent
states, have been considered. In order to see how such coherent phenomena can change
the semiclassical scenario presented thus far, we have extended our semiclassical simulation
scheme in terms of the density-matrix formalism introduced in section 3, which allows us
to describe on an equal footing phase coherence effects as well as energy relaxation and
dephasing phenomena. In the proposed quantum-transport approach, the basic ingredient is
the single-particle density matrix in equation (3.5) written in our k representation:
1
k
1
,
2
2
k
2
= tr{ c
2
k
2
c
1
k
1
}. (4.5)
Inviewof the in-plane translational symmetryof the carrier system, k
1
= k
2
density-matrix
elements are equal to zero; moreover, we shall neglect interstage (
1
=
2
) phase-coherence.
This amounts to saying that the electron coherence length is signicantly shorter than the period
of our nanostructure, which is conrmed a posteriori by our quantum-transport calculations.
Within such an approximation scheme we get
1
k
1
,
2
2
k
2
=
1
,
1
2
,k
1
k
1
k
2
. (4.6)
Analogous to our semiclassical treatment (see equation (4.3)), also for the new quantum-
transport formalism we can adopt the same periodic conditions to close the circuit:
,
1
2
,k
=
,
1
2
,k
=
2
,k
. (4.7)
This allows us to study the time evolution of the density matrix in the = 0 periodicity region
only, accounting for both diagonal and non-diagonal scattering processes within region = 0
as well as from and to regions = 1. Similarly to the semiclassical scheme, this provides
the currentvoltage characteristics without resorting to any phenomenological parameter.
As discussed in section 3, the time evolution of the single-particle density matrix is
governed by the following equation:
d
dt
2
,k
=

1
k

2
k
i h
2
,k
+ F[]
2
,k
. (4.8)
Anumerical solution of equation (4.8) is indeed prohibitive, due to its three-dimensional nature
as well as to the non-linearities in the functional F. Auseful simplication of the problemmay
be achieved if a time-step dynamical evolution of the density matrix is considered. Given
the at the generic timestep, corresponding to time t , the functional F can be easily evaluated
and cast into the following locally linear form:
F[]
2
,k
=
2
,k
2
,k;
2
,k
2
,k
. (4.9)
Moreover, since for the considered QCL designs in-plane and along-z carrier dynamics
are strongly decoupled, it is possible to adopt a factorization of the density matrix in (4.7)
according to
2
,k
=
2
f
k
, (4.10)
where f
denotes the parallel or in-plane carrier probability distribution. The latter should, in
principle, be -dependent. However, as demonstrated by the numerical results for state-of-the-
art devices presented in the following subsection, electrons within the various subbands of a
QCL structure are described by thermalized distribution functions, characterized by the same
effective temperature. Taking into account thatfor any probability distribution
k
f
k
= 1,
we get
2
=

k

2
,k
.
The above factorization scheme allows us to obtain an effectivei.e. one-dimensional
equation of motion for
2
of the form
d
dt
2
=

2
i h
2
+
1D
2
,
2
(4.11)
with
1D
2
,
2
=
k,k
2
,k;
2
,k
k
. (4.12)
Here, for each subband a parabolic dispersion with the same effective mass m
has been
assumed:
i
k
=
i
+
h
2
k
2
2m
. (4.13)
Equation (4.11), which has the structure of equations (3.14) and (3.15), involves the one-
dimensional effective superoperator
1D
, given by an in-plane average of the scattering tensor
in equation (4.9). As we shall see, such in-plane average can be readily performed within the
proposed simulation strategy via a MC sampling of the double k-space sum in equation (4.12).
Indeed, following the generic prescription [33], any sum of the form

i
g
i
f
i
, where f
i
is a
probability distribution, can be obtained by evaluating the arithmetic average of g
i
, where i is
randomly selected according to the probability distribution f
i
.
4.4. Simulation strategy
In view of their Boltzmann-like structures, equation (3.23) as well as (4.2) can be sampled
by means of a proper MC simulation scheme [4]. Indeed, the latter has proved to be a very
powerful technique allowing for the inclusion, at a kinetic level, of a large variety of scattering
mechanisms.
More specically, benetting from the translational symmetry previously discussed, we
can then simulate carrier transport over the central = 0stage only. Every time a carrier
in state undergoes an inter-stage scattering process (i.e. 0, 1,
), it is properly
reinjected into the central region (0, 0,
) and the corresponding electron charge e

will contribute to the current through the device. This charge-conserving scheme allows for
a purely microscopic evaluation of the device performances such as the gain spectrum or the
currentvoltage characteristics. The current density j across the whole structure, for example,
can be obtained as a pure output of the simulation just by a proper counting of the in- and
out-stage scattering processes:
j
kk
>
P
,k
f
k

<
P
,k
f
k
_
. (4.14)
Within such fully three-dimensional MC simulation scheme it is easy to compute the in-
plane average of the generalized scattering operator in (4.9) previously discussed. This
is performed by sampling the integral on the in-plane k coordinate over the usual free
ight + scattering event sequence of our many-particle simulation by means of a generalized
MC sampling strategy [33].
Contrary to equations (3.23) and (4.2), the one-dimensional effective quantum transport
equation in (4.11) is not Boltzmann-like, i.e. the matrix elements of the superoperator
1D
are not positive-denite. Therefore, equation (4.11) cannot be solved by means of a standard
particle-like MC simulation, as in the semiclassical case. In contrast, this quantum-transport
equation is solved by means of a direct numerical integration based on a time-step discretization
strategy.
In order to properly describe the complex interplay between carrier thermalization and
energy-relaxation/dephasing in state-of-the-art QCL structures, we have included in our
simulation all relevant interaction mechanisms in a fully three-dimensional fashion: intra- and
inter-subband carrierphonon as well as carriercarrier scattering processes. In the following
we shall present and discuss simulated experiments based on the semiclassical as well as
quantum-transport approaches previously introduced. They are aimed at understanding the
non-equilibrium electro-optical response of a variety of QCL designs.
4.5. Simulation of mid-infrared quantum-cascade laser
To achieve and maintain a population-inversion regime between the two subbands involved
in the optical transition, some clever design of the structure is required. In particular, the
inter-subband scattering rates have to be properly tailored to efciently deplete the lower laser
states, while preventing electrons from a too fast relaxation out of the upper laser states. This
does strongly depend on the frequency
phot
of the emitted light. For QCLs designed to emit
in the mid-IRregion of the electromagnetic spectrum, such a tailoring of the various relaxation
times can be achieved benetting from an energy-selective mechanism by which electrons
candissipate their energythe polar-optical phononemission. The Fr olichinteractionbetween
electrons and longitudinal optical (LO) phonon modes in polar materials (such as the IIIV
alloys which are usually employed in state-of-the-art unipolar devices) is characterized by
scattering rates that fall as the reciprocal of the squared transferred momentum, k. Since
LO-phonons are characterized by a well-dened energy h
LO
(their bands can be regarded as
dispersionless), an electronic inter-subband relaxation process via LO-emission is vertical in
momentumspaceand therefore highly probablewhen the difference in energy between the
two subbands matches the phonon energy. On the contrary, when the subbands are more than
one phonon far apart, i.e. h
phot
> h
LO
, the electronic transition comes out to be diagonal in
k space and thus less probable than in the previous case.
4.5.1. Partially phenomenological approach. Let us start our analysis by applying the
semiclassical partially phenomenological simulation scheme in equation (4.2) to state-of-
the-art mid-IR QCL structures. In particular, as prototypical structure, we have considered
the GaAs/(Al,Ga)As-based diagonal-conguration QCL proposed in [26], whose potential
prole is schematically depicted in gure 1. We shall focus on the hot-carrier dynamics
within the active region of the device, and we will therefore limit our kinetic treatment to
the subset = { = 0; = 1, 2, 3; k}. The formulation based on equation (4.2) provides
a fully three-dimensional description of energy relaxation based on microscopic intra- as
well as inter-miniband scattering mechanisms within the bare active region. Contrarily, the
injection/loss contributions (i.e. the coupling with the injector and the collector) are still treated
on a partially phenomenological level. However, as discussed previously, the functions g and
are now dened within the same kinetic picture k.
In view of their Boltzmann-like structures, the kinetic equations (4.2) are sampled by
means of a proper MC simulation scheme. In particular, the actual problem requires a MC
simulation where the total number of particles (in the active region of the device) is not constant.
With this aim, a combineddirect plus MCtime-dependent solution has been implemented,
which is based on a time-step separation between injection/loss and scattering contributions.
The only interaction mechanismconsidered in our simulated experiments is carrier-LOphonon
scattering; as previously recalled, this is indeed the key effect inspiring the device design at
these frequencies. A more detailed discussion on the role of other scattering mechanisms and
carriercarrier interaction is presented in the following subsection.
Figure 2 shows the three global carrier populations of the active-region subbands (n
k
f
k
) as a function of time, for an injection current of 10 kAcm
2
at 77 K. At the initial time
t = 0 the QCL active region is empty; then electrons are continuously injected into level 3,
according to a FermiDirac distribution corresponding to the lattice temperature and simulated
current density, andthe cascade3 2, 2 1occurs ontime-scales correspondingtothe
different values of the interminiband scattering rates. Eventually, the steady-state condition
is reached, leading to the desired 32 population inversion. Figure 2(a) has been obtained
using a typical value for the carrier escape time from level 1:
1
= 1 ps. Here, the results
of our MC simulated experiments (solid curves) are compared with the corresponding results
obtained by neglecting various Pauli-blocking factors (1 f ) (dotted curves) entering the
Boltzmann collision term in equation (3.24). Since in this simulation carrier escape from
level 1 is relatively fast (
1
= 1 ps), we have f
1k
1 and thus Pauli-blocking effects are not
very important. In contrast, by increasing the escape time from level 1 non-linear effects due
to Pauli blocking may play a crucial role. This can be clearly seen in gure 2(b), where the
same simulated experiments are shown for
1
= 10 ps. In this case, the strong Pauli blocking
leads to a signicant reduction of the interminiband relaxation 2 1, thus giving rise to a
phonon bottleneck and preventing the 3 2 population inversion
3
. Although a quantitative
evaluation of carrier dynamics in this regime of slow depletion of level 1 would require to
take into account the competing carriercarrier scattering channel, gure 2(b) shows that, as a
general prescription, it is important to limit n
1
also if this level is not directly involved in the
lasing process. Such Pauli-blocking effects are intrinsically time-dependent: during the rst
stage of the simulation n
1
is still small and the two results (solid and dotted) basically coincide;
at later times n
1
increases and signicant deviations come into play. As anticipated, such a
non-linear behaviour cannot be described by the rate-equation model in (2.1). The same is
true for the so-called in-plane (or intrasubband) relaxation. This is a purely three-dimensional
feature, whose analysis is straightforward within our MC simulation scheme.
Figure 3 presents a direct energy versus time two-dimensional map of the simulated
carriers corresponding to the run in gure 2(a). We can clearly see that, in addition to the
3 2 1 intersubband relaxation which is the only one considered in a purely macroscopic
model, the quantum-cascade process is also characterized by a signicant intrasubband energy
3
Note the different time scales in gures 2(a) and 2(b), corresponding to the different values of the escape time
1
.
Figure 2. Various carrier concentrations (n
, = 1, 2, 3) as a function of time with () and

without ( ) Pauli-blocking effects. Simulated experiments in (a) and (b) correspond to an
escape time
1
of 1 ps and 10 ps, respectively. From [30].
relaxation, resulting in the typical phonon-replica scenario. Since the overall intersubband
carrier transfer depends on the scattering rates P
k,
as well as on the carrier distributions f

k
in the various subbands, any phenomenological interlevel transition rate w
is unavoidably
time-dependent, thus giving rise to non-linear effects. In particular, for steady-state conditions
the energy position of the lowest phonon replica in subband 2 is crucial in determining the
2 1 phonon-induced carrier transfer. Therefore, in order to improve the QCL performance
it is important to locate the phonon replicas in phase-space regions with maximum 2 1
phonon coupling.
4.5.2. Global-simulation scheme. The kinetic description proposed in the previous
subsection can be quite useful to address the microscopic nature of the hot-carrier relaxation
within a portion of the structure which is the bare active region of the generic stage. However,
due to the presence of free-parameters coupling the region of interest to the rest of the device, it
does not allow us to address the nature of the physical mechanisms governing charge transport
through injector/active-region/collector interfaces. To this end, the partially phenomenological
model has to be replaced by the fully microscopic description of the whole MQWcore structure
in equation (3.23). Let us then apply such a global analysis again to the device schematically
depicted in gure 1.
Figure 3. Energy versus time two-dimensional map of the simulated carriers involved in the
3 2 1 cascade process, for the case
1
= 1 ps. From [30].
To model properly phase-breaking hopping processes, all various intra- as well as
inter-subband carrier-optical phonon and carriercarrier scattering mechanisms have been
considered. As far as the rst interaction is concerned, a bulk phonon bath is assumed,
neglecting both quantized/localized phonon states and hot-phonon effects, which should
give only minor corrections for the considered device designs [34]. For the carriercarrier
interaction, we employ the well-established time-dependent static-screening model commonly
used in MC simulation of energy relaxation and transport phenomena in two-dimensional
systems [35]. Other intrinsic scattering mechanisms, not included in the simulation, are
expected to play a minor role [4]. In particular, the interaction with acoustic phonons, in
spite of its quasi-elastic nature, does not affect charge transport signicantly due to its small
coupling constant.
The role of extrinsic mechanisms, such as interface roughness and carrier-impurity
scattering, does considerably vary according to material growth and processing, with the
result that the former are strongly material/device-dependent. In this respect, a modelling
of their effect would unavoidably require some a posteriori adjustment of free-parameters to
reproduce the diverse experimental data. For this reason, although they might be relevant to
properlyreproduce specic features of the behaviour of realistic devices at the various operating
conditions, extrinsic scattering mechanisms are not included in our microscopic simulations.
As a starting point, we have investigated the relative weight of the carriercarrier and
carrierphonon competing energy-relaxation channels. The time evolution of the carrier
population in the various subbands as well as of the total current density in the presence
of carrierphonon scattering are depicted in gure 4. Parts (a) and (b) refer, respectively, to
simulated experiments without and with two-body carriercarrier scattering. In our charge
conserving scheme, we start the simulation assuming the total number of carriers to be equally
distributed among the different subbands; then the electron distribution functions evolve
according to equation (3.23) and a steady-state condition is eventually reached, leading to the
desired 3 2 population inversion. As we can see, the inclusion of intercarrier scattering has
0 1 2 3 4 5
10
9
10
10
10
11
0
25
50
75
0 2 4 6 8 10
0 5 10 15 20 25 30
10
9
10
10
10
11
E
D
C
B
c
a
r
r
i
e
r

d
e
n
s
i
t
y

(
c
m
-
2
)
time (ps)
2
3
1
2
1
3
current density (kA/cm
2
)
e
l
e
c
t
r
i
c

f
i
e
l
d

(
k
V
/
c
m
)
carrier-phonon only
carrier-phonon + carrier-carrier
T = 77 K
A
E
D
C
B
A
c
a
r
r
i
e
r

d
e
n
s
i
t
y

(
c
m
-
2
)
time (ps)
(a)
(b)
(c)
Figure 4. Time evolution of simulated carrier densities in the various subbands of the MQW
structure of gure 1, without (a) and with (b) intercarrier scattering. (c) Simulated applied-eld
versus current-density characteristics of the whole structure with only carrierphonon ( ) and with
both carriercarrier and carrierphonon scattering (). Lines are a guide to the eye. Adapted
from [27].
signicant effects; it strongly increases inter-subband carrier redistribution, thus reducing the
electron accumulation in the lowest energy level Aand optimizing the coupling between active
region and injector/collector (the populations of subbands 3 and B become comparable). This
effect is crucial in determining the electron ux through the MQWstructure. Figure 4(c) shows
0 50 100 150 200 250 300
10
-4
10
-3
10
-2
10
-1
E
2
D
1
C
B
3 A
d
i
s
t
r
i
b
u
t
i
o
n

f
u
n
c
t
i
o
n

in-plane energy (meV)
Figure 5. Electron distribution function versus in-plane energy for carriers in subbands = 1, 2, 3
and = A, B, C, D, E of the MQW structure of gure 1. Sheet density = 3.9 10
11
cm
2
. The
slope of each curve provides an estimate of the effective carrier temperature T
in the various
subbands, which ranges from 600 to 750 K. Adapted from [31].
the simulated currentvoltage characteristics of our QCL device, obtained with (discs) and
without (lled squares) carriercarrier interaction. At the threshold operating parameters
estimated in [26] (marked by an arrow in gure 4(c)) the current density in the presence
of both electronphonon and electronelectron scattering mechanisms is about 5.6 kAcm
2
.
This value is about a factor of three higher than the result obtained with only carrierphonon
scattering and is in good agreement with the experimental ndings (7 kAcm
2
) [26].
The results plotted in gure 4 clearly demonstrate that within a purely semiclassical
picture the electronphonon interaction alone is not able to efciently couple the injector
subbands to the active region ones. While carrierphonon relaxation well describes the
electronic quantumcascade withinthe bare active region[30], carriercarrier scatteringplays an
essential role in determining charge transport through the full core region. This can be ascribed
to two typical features of carriercarrier interaction as compared with the carrierphonon
interaction: (i) this is a long-range two-body interaction mechanism, which also couples non-
overlapping single-particle states (see gure 1) and (ii) the corresponding scattering process at
relatively low carrier density is quasi-elastic, thus coupling nearly resonant energy levels, like
states 3 and B.
Tofocus onthe relative weight of the carriercarrier andcarrierphononcompetingenergy-
relaxation channels, we have investigated the steady-state electron distribution as a function of
the in-plane energy for different carrier densities [31]. Indeed, for low carrier concentrations,
intercarrier scattering plays a very minor role and electrons relax their energy via a cascade
of successive optical-phonon emissions; for high carrier densities, carriercarrier scattering is
more effective in setting up a heated Maxwellian distribution, with a given temperature T
.
This is a well-known trend which results from the screened-Coulomb nature of the carrier
carrier interaction potential [4, 6]. The absolute density value corresponding to the transition
between these two different regimes, however, depends on the heterostructure details and can
be obtained only after performing a direct simulation. Our results clearly showthat, for typical
operating conditions (sheet density = 3.9 10
11
cm
2
), electrons in QCLs thermalize within
each subband.
Temperatures corresponding to the diverse subbands vary in a quite narrow range (T
=
600750 K), as shown in gure 5 [31]. Again, the role of electronelectron interaction is crucial
in setting up this behaviour. This effect can be mainly ascribed to bi-intrasubband scattering
processes, i.e. intercarrier scattering processes coupling electrons in different subbands,
without changing their principal quantumnumber . They provide, in fact, a very efcient way
of redistributing excess kinetic energy in order to achieve a common effective temperature.
Such temperature results to be about 700 K, to be compared with a lattice temperature of 77 K
assumed in the simulation. This heating is a clear ngerprint of a strong hot-carrier regime:
the carrier system is not able to dissipatevia optical-phonon emissionthe relatively large
amount of energy provided by the applied bias. Recent experimental investigations appear to
conrm the theoretical ndings [36].
4.5.3. Quantum-transport approach. So far, due to our semiclassical description, no quantum
mechanical effects have been considered. This remark addresses, in particular, the nature
(coherent versus sequential) of the resonant-tunnelling injection of electrons into the active
region of the device, which has been a long-standing open issue [37]. To include coherent
phenomena, we have to employ the quantum-transport description presented in section 4.3,
which allows us to describe on the same footing phase coherence as well as energy-relaxation
and dephasing processes [6]. Comparing the results obtained with the latter approach to those
of the semiclassical global simulation scheme, we nd negligible quantum corrections (of a
few per cent) to the stationary current density. In the absence of carriercarrier scattering, we
get, e.g. a 2% quantum correction to the result at the threshold reported in gure 4. This is due
to the extremely small value of non-diagonal density-matrix elements
=
(compared with
the diagonal ones). The physical interpretation of such a behaviour proceeds as follows: non-
diagonal terms of the generalized carrierphonon scattering operators
in/out
tend to maintain
a non-diagonal density matrix also in stationary conditions. On the other hand, diagonal
energy-relaxation and dephasing processes tend to suppress non-diagonal elements on the
sub-picosecond time-scale. Since the average transit timethe time needed by an electron
to travel through the device period and thus to be re-injected into the simulation regionis
of the order of several picoseconds, the degree of coherence, i.e. the weight of non-diagonal
density-matrix elements, in stationary conditions is very small.
However, this does not mean that wavefunction-interference phenomena, like resonant-
tunnelling processes, are not present. Figure 6 presents the rst simulation of the ultrafast
dynamics of a properly tailored electron wavepacket, within the MQW core region of the
QCL sketched in gure 1. Here, the aim is to focus on the injectoractive-region tunnelling
mechanisms. For this purpose, the electronsystemhas beenpreparedat time t = 0 as a coherent
superposition of the two quasi-resonant states (0,B)(+1,3) shown in gure 1, to reproduce a
charge-density distribution fully localized in the = 0 injector. As clearly shown, the transient
dynamics is characterized by a strong interplay between phase-coherence and relaxation; only
after several picoseconds will it eventually reach the stationary transport solution, in which
incoherent sequential tunnelling is the dominant interwell mechanism.
Figure 6 refers to the case where only carrierphonon interaction is taken into account.
When carriercarrier contributions are also included the transient evolution (see gure 7) gets
much shorter and the systemreaches the steady state on a sub-picosecond time-scale. This time
is much shorter than the average transit-time across one period, which is the typical time-scale
needed to reach the steady-state condition. During the initial transient, a clear gain overshoot
appears, as shown in gure 7 [38]. These results suggest that ultrafast optical experiments,
like pump-and-probe or four-wave-mixing measurements, should provide a clear ngerprint
of such a coherent versus energy-relaxation carrier dynamics. This is conrmed by recent
ultrafast experiments by Eickemeyer and co-workers [39].
In spite of the fact that in steady-state conditions the currentvoltage characteristics
obtained from the quantum simulation do not differ signicantly from their classical
(c)
Figure 6. Time evolution of the charge density for an electron wavepacket properly tailored to
study the carrier tunnelling dynamics across the injection barrier for the QCL design of gure 1.
At t = 0 (c) the wavepacket is fully localized in the injector. The shaded regions correspond to the
(Al,Ga)As barriers in the heterostructure design. The transient dynamics (b) is characterized by a
strong interplay between phase-coherence and relaxation processes. At much longer time (a) the
system will eventually evolve into the stationary transport solution. From [32].
counterpart, a closer inspection of the steady-state density matrix
2
reveals the presence of
non-diagonal matrix elements. Such a feature, not present within a T
1
T
2
model [24, 40], is due
to the scattering-induced coupling between population and polarization discussed previously.
On the basis of the previous results, we are able to answer the long-standing controversial
question on the naturecoherent versus incoherentof charge transport in QCLs. For the
typical structures considered, energy-relaxation and dephasing processes are so strong as to
0.0 0.5 1.0 1.5 2. .5
0
1
2
3
4
5
6
n
u
p
p
e
r
-
n
l
o
w
e
r

(
1
0
1
6

c
m
-
3
)
time (ps)
Figure 7. Time evolution of the population difference between the upper and the lower laser state
for the QCL structure of gure 1. The system is prepared so that the charge distribution is fully
localized in the injector at time t = 0. The dotted line is set to the steady-state value. From [38].
destroy any phase-coherence effect on a sub-picosecond time-scale; as a result, the usual
semiclassical or incoherent description of stationary charge transport is found to be in excellent
agreement with the experimental results.
4.5.4. Active-regioncavity-mode coupling. Uptonow, our theoretical approachhas simulated
the behaviour of the bare electron system, without considering any optical-cavity effect. That
is, we have modelled the electroluminescent device and not the laser. A description of the
coupled carrier-photon systemmay of course be given at various levels of accuracy [41]. Within
the usual Fermis golden rule approximation, the cavity feedback on the electronic dynamics
is treated in terms of an additional Boltzmann-like contribution to equation (3.23) [42]. In
particular, for the case of a single-mode optical cavity, the latter has the form:
d
dt
f
k
light
=
[P
light
k,
k
f
P
light
,k
f
k
]. (4.15)
In equation (4.15), the two contributions corresponding to absorption and emission processes
can be identied
P
light
k,
k
= P
abs
k,
k
+ P
em
k,
k
(4.16)
with
P
abs
k,
k
= g
k,
N(1 f
k
) (4.17)
and
P
em
k,
k
= g
k,
(N + 1)(1 f
k
) (4.18)
here, N is the photon population of the mode and g
k,
is the carrier-photon coupling term.

The latter is dened as follows
g
k,
=
2
h
_
e h
m
_
2
|p|
2
0
V
1
h(
2
+
2
)
1
h
, (4.19)
where e and m are the electron charge and mass, respectively, V is the cavity volume, |p| the
momentum matrix element, the mode frequency, the spontaneous emission rate and the
frequency detuning.
Equation (4.15) goes together with the following photon counterpart:
d
dt
N = (A )N + B, (4.20)
where
A =
k,
g
k,
(f
k
f
), (4.21)
B =
k,
g
k,
f
k
(4.22)
and the N term describes loss processes. In state-of-the-art QCLs, the latter are mainly due
to quality of the mirrors and due to free-carrier absorption in the doped semiconductor regions
and in the metallic contact layers [17].
The main advantage of treating the carrierphoton systemvia the coupled equations (4.15)
and (4.20) is that, due to their structure, they can be sampled by means of a MC technique
and therefore implemented in the above described simulation code. Two aspects, however,
make this implementation not so straightforward. The rst is represented by the extremely
different time scales characterizing the carrier and photon dynamics. While the electronic
ensemble reorganizes itself within picoseconds, the build-up of the photon population in an
empty device might take several hundreds of picoseconds. The second reason is linked to
the control of the statistical uctuations, which are unavoidable when dealing with rare events.
These considerations suggest that a proper strategy for evaluating the evolution of the coupled
electronphoton system is by means of a so-called weighted MC simulation [33].
In its general formulation [43], the MC method is a powerful stochastic numerical
technique for the sampling of integro-differential equations. In this respect, the natural
probabilities occurring in our transport equations may be adjusted, or weighted, to optimize the
simulation efciency and reduce the statistical uctuations. This is the essence of a weighted
MCapproach and allows us to guide the carriers to sample particular phase-space regions [33].
In this way the simulation is able to yield more information about phenomena related to the
ne details of the distribution function, and a larger fraction of computer time is devoted to
obtain a better knowledge of the special features of interest. The traditional MC technique
is a special case of the weighted scheme, in which the arbitrary probabilities assigned to the
various processes coincide with the natural ones.
In the spirit of the weighted MC scheme, the carrier-photon scattering probabilities in
equation (4.15) can be articially enhanced to effectively control the statistical uctuations.
This artefact is then compensated by a proper renormalization of the carrier and photon
counters, so that the steady-state result does not depend on the weighting factor. Due to
the rare-event nature of the problem, however, the transient dynamics may show signicant
variations, especially when simulating the photon build-up in an initially empty cavity. This
is shown in gure 8 where the photon populations are reported as a function of the simulated
time, for the initial condition of ten photons in the cavity at time t = 0 and for two different
sequences of (pseudo)random numbers. Such instability is a mathematical feature of the
photon equation (4.20) and not an anomaly of the weighting procedure. To reduce computing
resources, the simulations of gure 8 have been performed in the presence of only carrier
phonon scattering. The relevant cavity parameters in equations (4.15) and (4.20) are taken
from [26], while the spontaneous emission linewidth is adjusted to reproduce the emitted
power of the operating device. Such a free parameter (varying in the range 912 meV, in
extremely good agreement with the experimental ndings [26]) includes the contributions of
400 800 1200 1600
10
1
10
3
10
5
10
7
10
9
n
u
m
b
e
r

o
f

p
h
o
t
o
n
s
time (ps)
Figure 8. Microscopic Monte Carlo simulation of the lasing build-up in the device of [26]. The
two curves refer to simulated experiments differing in the (pseudo)random number sequence; the
initial transient shows the typical oscillations of a rare-event problem.
all intrinsic (carriercarrier, carrierphonon, etc) as well as extrinsic (carrier-impurity, interface
roughness, etc) scattering mechanisms and is therefore strongly device-dependent.
Analogously to the present case, the MCsimulation of other semiconductor-based devices,
such as QWIPs, would also greatly benet from a weighting strategy. In these systems, in
fact, the carriers involved in the transport processes of interest (i.e. those giving rise to the
photo-current) are very few; so few that the statistics is insufcient to rely on the central-limit
theorem. Being interested in the current change on the biased device, this system is actually
the right candidate for the alternative steady-state weighted MC scheme that we have recently
proposed [44, 45].
4.6. Towards terahertz sources
Since their rst demonstration in 1994 [16], the performances of QCLs have experienced
tremendous improvements and the range of emission wavelength has been continuously
extended [17]. In principle, QCLs can be designed to emit at any frequency over an extremely
wide range, using the same combination of materials and only varying the heterostructure
design. However, the translation of the cascade scheme into devices operating at photon
energies below the LO-phonon threshold of the host material ( h
LO
36 meV in GaAs) is
denitely not straightforward. In this conguration, in fact, the main difculty in achieving
a population inversion regime in the active region is that optical-phonon emission cannot be
used to selectively depopulate the MQW excited states, since it acts with equal efciency on
both the upper and the lower laser subbands. On the contrary, the complex and non-intuitive
interplay between various competing non-radiative relaxation channels has to be taken into
account to properly evaluate the performances of newdesigns and proposals. In particular, two
main targets have to be fullled: efcient depletion of the lower laser states and long lifetime
in the upper ones.
Although electroluminescence in the terahertz region of the electromagnetic spectrumhas
been detected from a variety of QC structures in the last years, no evidence of population
inversion was achieved, mainly due to slow extraction of electrons from the lower laser
states. The parameter-free simulation scheme previously introduced has proved to be an
ideal tool to have some insight into the main limitations of these early structures and for the
characterization and optimization of novel QCL prototypes [46, 47]. Indeed, our simulated
experiments allowed us to propose promising designs for which the predicted lasing action
was later demonstrated [48].
Our prototypical QCL structure is designed to emit at 4.4 THz ( 67 m) [46]. It is
based on a vertical-transition conguration, which is known to lead to larger dipole matrix
elements and narrower linewidths, and employs a conventional chirped-superlattice design.
Its operating strategy exploits the same concepts successfully implemented for the shortest-
wavelength ( 1724 m) mid-IR QC lasers [49]. To minimize the occupation of the lower
laser subband, we employ a dense miniband with seven subbands, which provides a large
phase space where electrons scattered either from the upper laser subband or directly from
the injector can spread. The miniband dispersion is chosen as large as possible compatibly
with the need of avoiding cross absorption. This suppresses thermal backlling and provides
a large operating range of currents and voltages. Energy relaxation within the rst miniband
would appear to be hindered by the lack of nal states with appropriate energy to allow for
LO-phonon emission. Nevertheless, carriercarrier interaction does benecially operate as
an activation mechanism since it provides sufcient in-plane momentum for the electrons to
open a scattering path via optical phonon to the lower subbands. In the presence of carrier
carrier interaction, electrons efciently relax into the lower states of the injector from where
they are readily trasferred into the upper laser state of the following period. A signicant
population inversion then results between the subbands of the laser transition [46]. Apart from
its strong technological implications, this nanostructured device came out to be an ideal context
to investigate the complex synergy between carrier thermalization and phonon-assisted energy
relaxation. This is indeed the key ingredient for a proper operation of these new light sources.
5. Modelling of quantum devices with open boundaries
As anticipated in the introductory part of the paper, in spite of the quantum-mechanical nature
of carrier dynamics in the active region of typical nanostructured devices, the overall behaviour
of such quantum systems is often the result of a complex interplay between phase coherence
and energy relaxation/dephasing, the latter being primarily induced by the presence of spatial
boundaries [8]. In spite of the kinetic nature of the scattering dynamics in (2.3), the contribution
in (2.4) describes carrier injection and loss processes on a partially phenomenological level.
This has two important consequences. First of all, no dependency on the real position of the
device spatial boundaries is included. Secondly, hybrid simulation strategies are required [30],
combining a MC sampling of the scattering dynamics with a direct numerical integration of
injection/loss terms. The former aspect is the subject of section 5.1, where a fully-microscopic
real-space description of the carrier injection process is reviewed. The latter is addressed in
section 5.2, where a kinetic description of the system-reservoir thermalization processes is
revisited, which allows us to model open quantum devices within a closed-system paradigm.
5.1. Wigner-function formalism
A proper treatment of quantum devices with open boundaries requires a theoretical approach
able to account for both coherent and incoherenti.e. phase-breakingprocesses on an equal
footing. Towards this end, a commonly used theoretical instrument is the single-particle
Figure 9. Schematic representation of the device active region sandwiched between its electrical
contacts (a) and of the corresponding U boundary-condition scheme for a one-dimensional
system (b). The latter implies, in particular, the knowledge of the incoming Wigner function
f (z
b
, k), i.e. f (z
left
, k > 0) and f (z
right
, k < 0). From [10].
density matrix [24];
4
however, while the latter is the ideal tool for the description of ultrafast
phenomena in innitely-extended/periodically-repeated nanostructures, the density-matrix
approach cannot be directly applied to quantum systems with open boundaries, for which
a real-space treatment is imperative.
Such a real-space description is naturally provided by the Wigner-function formalism [8];
within this approach the statistical quantum state of the electronic subsystem is fully described
in terms of the so-called Wigner function, a function dened over the conventional phase-space
as the WeylWigner transformof the single-particle density matrix [6,9]. Different approaches
to the study of quantum-transport phenomena in semiconductor nanostructures based on the
Wigner-function formalismhave been proposed. On the one hand, starting fromthe pioneering
work by Frensley [51], few groups [52] have performed quantum-transport simulations based
on a direct numerical solution of the Wigner transport equation via nite-difference approaches
by imposing the standard boundary-condition scheme of semiclassical device modelling, also
called U scheme (see gure 9). On the other hand, a generalization to systems with open
boundaries of the SBE in (3.14) has been proposed [9, 10]. In the latter approach, spatial
boundaries are incorporated through the following three-step procedure: (i) application of
the WeylWigner transform to the standard SBE; (ii) implementation of spatial boundary
conditions to the Wigner function via the U scheme previously mentioned and (iii) application
4
An equivalent and alternative approach to the density-matrix formalism is provided by the non-equilibrium Greens
function technique. By using the latter, Lake, Datta and co-workers have proposed an efcient quantum-transport
simulation scheme for the study of open semiconductor devices [50].
to the resulting Wigner equation of the WeylWigner anti-transform. As discussed below, such
generalized SBE describe the open nature of the problem via a boundary source term and a
corresponding renormalization of the Liouville superoperator.
In addition to the two alternative simulation strategies previously recalledboth based
on effective treatments of relevant interaction mechanismsJacoboni and co-workers have
proposed a fully quantum-mechanical simulation scheme for the study of electronphonon
interaction based on the so-called Wigner paths [53]. This approach is intrinsically able to
overcome the standard approximations of the semiclassical scattering theory; however, due to
the huge amount of computation required, its applicability is often limited to short time-scales
and extremely simplied situations.
In this subsection we shall mainly review the microscopic analysis presented in [10],
which provides a fully quantum-mechanical description of the coupling dynamics between
the device active region and external charge reservoirs. With this aim, we start revisiting the
theoretical approach proposed in [9] and reviewed in more detail in [6].
5.1.1. Conventional Wigner-function description. The starting point of the microscopic
treatment of open quantum systems proposed in [9] is the SBE for a closed system in
equation (3.14) where the effective Liouville operator given in equation (3.15) is the
sum of two terms: coherent (i.e. scattering-free) single-particle evolution plus energy-
relaxation/dephasing dynamics. The latter is described in terms of the scattering tensor ,
whose explicit forminvolves the microscopic in- and out-scattering rates for the various carrier-
quasi-particle interaction mechanisms considered. The key idea proposed in [9] is to apply the
usual WeylWigner transform
u
2
(r, k) =
_
dr
1
_
r +
r
2
_
e
ikr
(2)
3/2
2
_
r
r
2
_
(5.1)
(
(r) r| denoting the single-particle wavefunction of state ) to the SBE in (3.14). In

this way the latter is translated into its phase-space representation r, k, which allows us to
impose on the Wigner function [8]
f (r, k) =
2
u
2
(r, k)
2
(5.2)
the desired values at the device spatial boundaries according to the well-known U scheme
depicted in gure 9. More specically, in order to properly impose the desired spatial boundary
conditions on the equation of motion for f , we add and subtract a source term
S(r, k) = v(k)f
b
(k)(r r
b
), (5.3)
where v(k) denotes the negative or incoming part of the carrier group velocity normal to the
boundary surface and f
b
(k) is the Wigner function describing the distribution of the injected
carriers. By applying the inverse of the WeylWigner transform in (5.1) to the new equation
of motion for f , we nally get
d
dt
2
=
2
L
2
,
2
+ S
2
L
2
,
2
, (5.4)
where S
2
is the WeylWigner antitransform of the source term in (5.3) and the effective
Liouville operator L in (3.15) is renormalized by
L
2
,
2
=
_
dr
b
dk u
2
(r
b
, k)v(k)u
2
(r
b
, k). (5.5)
Figure 10. Comparison between the real-space charge distribution obtained from the
phenomenological injection model in (2.4) [n(r) =

(r)|
2
dashed curve] and the
microscopic model in (5.8) [n(r) =

1
(r)
2
(r)solid curve] for a GaAs-based
single-barrier structure (with height of 0.5 eV and width of 4 nm) equidistant from the electrical
contacts. In this room-temperature simulation carriers are injected primarily from the left, due to
a misalignment = 0.2 eV between the left and right chemical potentials. The corresponding
charge distribution in momentum space is also reported in the inset. From [10].
Equation (5.4) is the desired generalization to open systems of the SBE in (3.14)
5
and
may be schematically written as
d
dt
2
=
d
dt
L
+
d
dt
res
(5.6)
with
d
dt
res
= S
2
L
2
,
2
. (5.7)
As for the semiclassical case (see equation (2.2)), the global system dynamics is the sum
of the dynamics induced by the Liouville superoperator L inside the device active region
plus the one induced by the presence of the external reservoirs. While the former is trace-
preserving, the latter leads, in general, to a variation of the total number of carriers within
the spatial region of interest, exactly as for the semiclassical model in equations (2.2)(2.4).
Moreover, equation (5.7) exhibits the same injection-minus-loss structure of the relaxation-
time-approximation model in (2.4): the quantum-mechanical source termS corresponds to the
semiclassical generation G, while the superoperator L is a non-diagonal generalization of
the loss rate .
To validate the theoretical approach presented so far, we shall focus on a very simple
semiconductor nanostructure: a single-barrier equidistant from the device contacts (see
gure 10). As basis states we adopt the scattering states of the device potential prole;
moreover, to better identify the role played by carrier injection, we shall neglect all other
sources of energy relaxation/dephasing in the device active region due to carrier-quasi-particle
5
The open character of the systemresults in a non-Hermitian correction Lto the Liouville operator L, whose effect
is equivalent to a purely dissipative process within the simulated region, as originally pointed out in [8].
scattering:
2
,
2
= 0. Under these assumptions, equation (5.4) in steady-state conditions
reduces to
i
h
(
2
)
2
+
2
L
2
,
2
= S
2
. (5.8)
Figure 10 shows results for the single-barrier potential prole when carriers are primarily
injected from the left. Here, the simulated real-space charge distribution obtained from the
phenomenological injection model in equation (2.4) (dashed curve) is compared with that of
the microscopic model in (5.8) (solid curves). As we can see, the two models give completely
different results. The phenomenological model gives basically what we expect: since we have
signicant carrier injection from left only and since the potential barrier is relatively high, the
carrier distribution is mainly located on the left side. In contrast, the microscopic model gives
an almost symmetric charge distribution.
In order to understand the origin of this non-physical result, let us focus on the nature of
the source term in (5.4). Contrary to the phenomenological injection/loss term of (2.4), the
latter is intrinsically non-diagonal, i.e. the injection of a carrier with well-dened wavevector
k (see equation (5.3)) is described by a non-diagonal source contribution S
2
. In other words,
we inject into the device active region a coherent superposition of states
1
and
2
, in clear
contrast with the idea of injection from a thermali.e. diagonalcharge reservoir. More
specically, in this case the generic scattering state on the left comes out to be an almost
equally weighted superposition of +k and k:
k
(z) = a
k
e
ikz
+ b
k
e
ikz
. This, in turn, tells
us that the generic plane-wave state k injected from the left contact is also an almost equally
weighted superposition of the left and right scattering states. This is why the charge distribution
(solid curve in gure 10) is almost symmetric; any electron injected from left couples to left
as well as to right scattering states.
The anomaly of the microscopic model is even more pronounced if we look at the carrier
distribution in momentumspace (see inset in gure 10). While for the phenomenological model
(dashed curve) we get a positive-denite distribution showing, as expected, the two symmetric
wavevector components of the scattering state, the microscopic result is not positive denite;
this tells us that the boundary-condition scheme considered so far does not provide a good
Wigner function.
The scenario previously discussed is highly non-physical; it can be mainly ascribed to the
boundary-condition scheme employed so far, which implies injection of plane-wave electrons
(see source term in equation (5.3)), regardless of the shape of the device potential prole. This
is an intrinsic limitation of the conventional Wigner-function representation r, k. It is then
clear that, in order to overcome this limitation, what we need is a boundary-condition scheme
realizing diagonal injection over the scattering states of the device potential prole.
5.1.2. Generalized Wigner-function approach. To overcome the intrinsic limitations of
the conventional Wigner-function formulation previously discussed, the key idea proposed
in [10] is to extend the WeylWigner transform in (5.1) from the k to a generic basis set {|}
according to
6
u
2
(r) =
_
dr
1
_
r +
r
2
_
1
_
r +
r
2
_
2
_
r
r
2
_
2
_
r
r
2
_
, (5.9)
6
Here, the set of basis functions
(r) r| is, in general, different from the basis set
. It is easy to show that if

we consider as basis states | conventional plane waves, the standard WeylWigner transform in (5.1) is recovered.
We stress thatcontrary to (5.1)this new WeylWigner transform is not a unitary transformation corresponding to
a simple basis change; it amounts to a non-trivial projection operation involving the real-space Wigner coordinate r:
u
2
(r) =
1
|
1
|rr|
2
|
2
.
where denotes the volume of the simulated region. In analogy with (5.2), our generalized
Wigner function is given by
7
:
2
(r) =
2
u
2
(r)
2
. (5.10)
By combining equations (5.2) and (5.10), the new Wigner function

f can be easily expressed
in terms of the standard one as
2
(r) =
_
dr
dk
2
(r; r
, k
)f (r
, k
) (5.11)
with
K
2
(r; r
, k
) =
2
u
2
(r)u
2
(r
, k
). (5.12)
The new Wigner function can then be regarded as a sort of convolution of the original one
with the kernel K in (5.12). This may recall a well-established procedure used to obtain
positive-denite phase-space quantumdistributions, the so-called smoothing procedure [54].
However, we stress that this is not the case: (i) here there is no need for a positive-denite
function and (ii) contrary to the standard smoothing procedure, the initial and nal phase spaces
do not coincide (r
, k
r,
1
2
).
By adopting as basis states | again the scattering states of the device potential prole
| and assuming a diagonal source term of the form
2
(r) = v
1
f
b
2
(r r
b
), (5.13)
the equation of motion for the new Wigner function

f in (5.10) will be given by
d
dt
2
(r) =
2
_
dr
[L
2
,
2
(r, r
) + L
2
,
2
(r, r
)]

f
2
(r) +

S
2
(r) (5.14)
with
L
2
,
2
(r, r
) = v
1
,
2
2
,
2
(r r
b
)(r r
). (5.15)
We stress that now the source term

S in (5.13) describes diagonal injection over the scattering
states (with velocity v
), as requested. Indeed, if we nowintegrate equation (5.14) over the real-

space coordinate r, we get again the density-matrix equation in (5.4), but now with a diagonal
source termS
2
= v
1
f
b
2
and a much simpleri.e. partially diagonalrenormalization
term L
2
,
2
= v
2
u
2
(r
b
). In the scattering-free case, the stationary solution is
again described by equation (5.8). However, due to the diagonal nature of the new source
term as well as of the partially diagonal structure of L, equation (5.8) now has a diagonal
solution:
2
= f
2
. More specically, the diagonal density-matrix elements f
obey
the following steady-state equation:
= f
b
(5.16)
with T
= u
(r
b
). Equation (5.16) is semiclassical in nature, i.e. it involves diagonal
density-matrix terms only. However, contrary to the phenomenological injection model in
(2.4), here the distribution function in state is the result of an incoherent superposition
from all the injection channels: f
T
1
f
b
. We nally stress that, by replacing the

T with the identity operator (T
), the phenomenological injection model in (2.4) is

recovered.
7
Similarly to the conventional Wigner-function theory [8], the real-space charge distribution (see solid line in
gure 11) is now given by n(r) =

(r).
Figure 11. Same as in gure 10 but for the new microscopic model in equation (5.16). From [10].
Figure 11 shows again results for the single-barrier potential prole previously considered.
Here, the simulation based on the phenomenological injection model in equation (2.4) (dashed
curves) is compared with that of the new microscopic model in (5.16) (solid curves). As we
can see, the highly non-physical behaviours of gure 10 (solid curves) have been completely
removed. Indeed, the momentum distribution in the inset is always positive-denite and the
two models exhibit a very similar behaviour. We nd relatively small deviations close to
the device spatial boundaries, which can be ascribed to the interlevel injection coupling T
(see equation (5.16)), not present in the phenomenological injection model. This is clearly a
ngerprint of our real-space description, where the point-like carrier injection is located at the
device spatial boundaries. However, when the device active region is relatively far from the
contacts these deviations can be safely neglected, and the phenomenological model in (2.4)
provides reliable results.
At this point, a question needs to be answered: would it be possible to describe the diagonal
injection over the scattering states in (5.13) by means of an ad hoc source termS(r, k) within the
standard phase-space? The answer to this question is yes. However, a closer inspection reveals
that such an ad hoc function can never be point-like in space, which in turn does not allow
us to employ the conventional boundary-condition approach considered so far (see gure 9),
where we impose the value of the Wigner distribution only at the device spatial boundaries by
means of an effective point-like source term (see equation (5.13)). Therefore, not only is the
generalized WeylWigner representation proposed in [10] physically sound, but it also allows
us to maintain all the well-known advantages of the standard boundary condition scheme.
5.2. A closed-system paradigm
The fully microscopic treatments reviewed above are essential for the basic understanding of
the quantum phenomena involved in new-generation nanoscale devices. However, they are
often extremely computer-time consuming and cannot be employed in standard optoelectronic-
device modelling and optimization where, in contrast, partially phenomenological (and
computationally affordable) models are highly preferred. Within such phenomenological
treatments the coupling of the quantum device with external reservoirs is typically described
in terms of extremely simplied injection/loss models and requires hybrid simulation schemes
(which, e.g. do not conserve the number of simulated particles). In this subsection a different
strategy is proposed [55], which allows us to treat also the partial carrier thermalization induced
by the device spatial boundaries via a traditional (i.e. particle-conserving) MC sampling of
the Boltzmann equation. This particle-conserving (closed-system) kinetic approach is then
extended to the quantum-transport regime [56].
5.2.1. Semiclassical case. In spite of the kinetic nature of the scattering dynamics
in (2.3), the contribution in (2.4) describes carrier injection and loss processes on a
partially phenomenological level; this, in turn, requires hybrid simulation strategies [30]
combining a MC sampling of the scattering dynamics with a direct numerical integration
of injection/loss terms.
In what follows we propose to replace the conventional relaxation-time term in (2.4) with
a Boltzmann-like operator of the form
d
dt
f
res
=
_
P
r
P
r
_
. (5.17)
This contribution has indeed the same structure of the scattering operator in (2.3); however, the
new scattering rates P
r
describe electronic transitions within the simulated region induced

by the coupling to the external carrier reservoirs. It is worthwhile to stress that, contrary to
the conventional injection/loss term in (2.4), in this case there is no particle exchange between
device active region and thermal reservoirs. The total number of simulated particles is therefore
conserved.
Let us now discuss the explicit form of the rates P
r
entering equation (5.17). In the

absence of scattering processes (P
s
= 0), the steady-state solution of the conventional

injection/loss model in (2.4) is f
= f
, i.e. the carrier distribution inside the device coincides

with the distribution in the external carrier reservoirs. As a rst requirement, we therefore
impose the same steady-state solution (f
= f
) to the new collision operator in (5.17).

This, in turn, will impose conditions on the explicit form of the scattering rates P
r
. More
specically, from the detailed-balance principle [4] we get
8
:
P
r
P
r
=
f
. (5.18)
It follows that our transition rates should be of the form
P
r
= P
, (5.19)
where P can be any positive and symmetric transition matrix (P
= P
> 0). What is

important in steady-state conditions is the ratio of the scattering rates in (5.18) and not their
absolute values which are, in contrast, crucial in determining the transient non-equilibrium
response of the system. Since our aim is to replace the injection/loss term in (2.4) with
the Boltzmann-like term in (5.17), as second requirement we ask that the relaxation dynamics
induced by the newcollision termcorrespond to the phenomenological relaxation times in (2.4).
This corresponds to imposing that the total out-scattering ratesummed over all possible nal
statescoincides with the relaxation rates
P
r
. (5.20)
8
For the case of a non-equilibrium carrier distribution f
, the detailed-balance principle should be replaced by a

total-balance one. However, by inserting equation (5.19) into the effective collision operator in (5.17) it is easy to
verify that its steady-state solution is f
= f
.
By assumingas simplest form of the symmetric transition matrix in (5.19)P
= p
,
equation (5.20) reduces to the following system of equations for the unknown quantities p
. (5.21)
Since the sum on the left is -independent, we immediately get p
. Starting from this

result, we nally obtain
P
= p
. (5.22)
The explicit form of the desired system-reservoir scattering rates entering the Boltzmann-
like collision term in (5.17) may be derived by combining equations (5.19) and (5.22). As
stated above, the proposed kinetic formulation in terms of Boltzmann-like collision operators
only is particularly suited to a standard ensemble-MC simulation approach, where one deals
with a xed number of particles. In this respect, contrary to the phenomenological model in
equation (3), in the present closed-system formulation the total carrier density is not xed by
the external reservoirs and the resulting transport equation is homogeneous.
To test the proposed simulation strategy, we have developed a fully three-dimensional MC
simulator, usingas basis states the product of scatteringstates alongthe eld/growthdirection,
and two-dimensional plane waves accounting for the in-plane dynamics. In order to properly
describe phonon-induced energy and momentum relaxation within the device active region,
carrierphonon scattering in a fully three-dimensional fashion has been included, in addition to
the newscattering-like thermalization mechanismin (5.17). Since in the simulated experiments
discussed below we shall focus on low density conditions, carriercarrier scattering has not
been considered.
We start considering an extremely simple transport problem: a GaAs mesoscopic bulk
system of length l = 200 nm sandwiched between two reservoirs with different chemical
potentials (
left

right
= 50 meV). We have applied to this problem the simulation strategy
previously described (see equation (5.17)) and have compared the results with those of the
conventional simulation approach (see equation (2.4)).
Figure 12(a) presents the transient carrier dynamics resulting from the conventional
injection/loss model in (2.4). Here, we show the time evolution of the carrier distribution
in momentum space at steps of 1 ps. Since in this model we start at time t = 0 with an empty-
device conguration, the simulated experiment shows a progressive increase of the carrier
distribution, which from the very beginning exhibits a strong leftright asymmetry due to the
chemical-potential misalignment. This scenario manifests the open nature of the conventional
approach, which does not allow the direct use of a standard MC procedure.
Figure 12(b) shows again the transient evolution of the carrier distribution in momentum
space but obtained from the proposed simulation approach. In this case we deal with a xed
number of particles which at time t = 0 are arbitrarily chosen to be equally distributed in the
three-dimensional momentum space. Moreover, the total carrier density, which is now a free
parameter, has been set equal to the steady-state value in (a) (which can be directly evaluated
from the thermal distributions f
). Contrary to the time evolution in (a), here at very short

times the device region is already occupied and its charge distribution in momentum space is
almost symmetric. Only at later times, due to the effective scattering mechanism in (5.19), we
recover the asymmetric distribution of gure 12(a) (see solid curve).
Figure 13 shows the charge current density as a function of time corresponding to the two
simulated experiments in gure 12(a) (dashed curve) and gure 12(b) (solid curve). At time
t = 0 the current is in both cases equal to zero; this is however ascribed to different reasons: in
gure 12(a) at t = 0 the carrier density is equal to zero while the mean velocity is different from
0.0
0.5
1.0
1.5
2.0
2.5
k
z
(nm
-1
)
(b)
(a)
c
h
a
r
g
e

d
e
n
s
i
t
y

(
a
r
b
.

u
n
i
t
s
)
-0.4 -0.2 0.0 0.2 0.4
0.0
0.5
1.0
1.5
2.0
2.5
Figure 12. Room-temperature transport properties of a GaAs mesoscopic bulk system of length
l = 200 nm, sandwiched between two reservoirs with different chemical potentials (
left
right
=
50 meV). Transient dynamics of the carrier distribution in momentum spacefrom 1 ps (dashed
curve) to 9 ps (thick solid curve) at intervals of 1 ps (thin solid curves)as obtained (a) from the
conventional injection/loss model of equation (2.4) and (b) from the proposed simulation strategy.
Adapted from [55].
0 2 4 6 8 10 12
0.0
0.4
0.8
1.2
1.6
2.0
c
u
r
r
e
n
t

d
e
n
s
i
t
y

(
k
A
/
c
m
2
)
time (ps)
Figure 13. Charge current density as a function of time corresponding to the two simulated
experiments in gures 12(a) (- - - -) and (b) (). The scattering-free or ballistic result ( )
is also reported. Adapted from [55].
zero; in gure 12(b) the mean velocity is equal to zero while the carrier density is different from
zero. In spite of a slightly different transient, both curves reach almost the same steady-state
value, conrming the validity of the proposed simulation strategy. To compare the transient
response of the two models it is useful to assume identical initial conditions. To this end
we have also simulated the open system starting from the same initial distribution considered
in (b). The resulting current density as a function of time (not reported here) coincides within
a few per cent with the dashed curve in (c).
The steady-state regime results froma strong interplay between the thermalization induced
by the external reservoirs and the phonon-induced momentum relaxation within the device
0 50 100 150 200 250 300
0
50
100
150
200
c
u
r
r
e
n
t

d
e
n
s
i
t
y

(
A
/
c
m
2
)
voltage (mV)
Figure 14. Room-temperature currentvoltage characteristics of a GaAs/AlGaAs resonant-
tunnelling diode (with barrier height of 0.24 eV and barrier width and separation of 2.8 nm and
4.4 nm, respectively) as obtained from our MC simulation scheme with () and without (- - - -)
carrierphonon scattering. From [55].
active region. Indeed, in the phonon-free case (dotted curve in gure 13) the steady-
state currentwhich is fully ballisticreaches signicantly higher values. The momentum-
relaxation dynamics previously mentioned is clearly visible in gure 12(a), where the peaks
of the injected carrier distribution are progressively shifted to lower wavevectors.
As a second testbed, we have considered a prototypical semiconductor quantum device:
a GaAs/AlGaAs resonant-tunnelling diode with a barrier height of 0.24 eV and a barrier width
and separation of 2.8 nm and 4.4 nm, respectively. Figure 14 shows the currentvoltage
characteristics obtained from the proposed MC simulation scheme with (solid curve) and
without (dashed curve) carrierphonon scattering. The results demonstrate that we are able
to properly describe the typical resonance scenario. More specically, as expected, in the
presence of phase-breaking processes, like carrierphonon scattering, the resonance peak is
signicantly reduced. Also in this more realistic case the proposed simulation strategy properly
describes the key phenomena under investigation.
5.2.2. Generalization to the quantum-mechanical case. The aim of the present section is to
extend the theoretical framework previously introduced to the quantum-mechanical case.
The formulation in terms of the effective Liouville superoperator in (3.14), recalled in
section 3, is typical of a so-called closed system, i.e. a systemdened over the whole coordinate
space. This is conrmed by the trace-preserving character of the Liouville superoperator L,
which corresponds to say that the total number of carriers is preserved.
What we propose here is a completely different approach: following the very same strategy
introduced in section 5.2.1, the key idea is again to replace the particle-non-conserving term
in (5.7) with an ad hoc scattering superoperator
r
describing on a kinetic level the system-
reservoir thermalizationprocess. More specically, similar toequation(5.17), for the quantum-
mechanical case we may write:
d
dt
res
=
2
,
2
. (5.23)
Let us now discuss the explicit form of this new scattering superoperator
r
. As a rst
requirement, we shall ask that in the semiclassical limit (
2
= f
2
) equation (5.23) will
reduce to equation (5.17). This requires that
1
,
1
= P
r
P
r
1
. (5.24)
In addition to these semiclassical or T
1
terms, the scattering superoperator
r
should also
contain dephasing or T
2
contributions. The latter describe, in general, decoherence effects
induced by the external reservoir on the carrier subsystem and will produce a damping of the
non-diagonal density-matrix elements; for a given non-diagonal term
1
=
2
, the corresponding
dephasing rate is given by the average of the total out-scattering rates for states
1
and
2
, i.e.
2
,
1
2
=
1
2
_
P
r
1
+
P
r
2
_
. (5.25)
As anticipated, in addition to T
1
and T
2
terms, a generic scattering superoperator may also
contain additional contributions describing non-trivial couplings between diagonal and non-
diagonal density-matrix elements. Such extra terms may lead to a non-diagonal steady-
state solution. However, since in the absence of scattering mechanisms inside the simulated
region we require a quasi-thermal, i.e. diagonal, steady-state solution, these extra-terms in the
scattering superoperator
r
are set equal to zero. Combining equations (5.24) and (5.25), we
nally obtain
2
,
2
=
1
,
2
2
_
P
r
P
r
1
_

1
2
2
,
2
_
1
P
r
1
+
2
P
r
2
_
.
(5.26)
We stress that the only ingredients entering the proposed effective scattering superoperator are
the device-reservoir effective scattering rates in (5.19).
By combining equations (3.14) and (5.23), in steady-state conditions, the proposed
quantum-transport equation in (5.6) is given by
d
dt
2
=
2
L
open
2
,
2
= 0 (5.27)
with
L
open
2
,
2
= L
2
,
2
+
r
2
,
2
. (5.28)
6. Summary
In this paper, we have reviewed state-of-the-art approaches used in the design and theoretical
modelling of optoelectronic quantum devices. The primary goal was to provide a cohesive
treatment of basic quantum-transport effects, able to explain and predict the performances
of new generation semiconductor quantum devices. With this aim, we have reviewed and
discussed a three-dimensional microscopic description of time-dependent as well as steady-
state quantum-transport phenomena, based on the density matrix formalism. This has allowed
us to introduce in a quite natural way the separation between coherent and incoherent processes.
Starting from this general theoretical framework, we have analysed two different types of
quantum devices, namely periodically repeated structures (e.g. QCLs) and quantum systems
with open boundaries (e.g. multibarrier nanostructures).
For the rst class, we have shown howa proper use of periodic boundary conditions allows
us to reproduce and predict the currentvoltage characteristics of state-of-the-art mid- and far-
IR QCLs without resorting to phenomenological parameters. In particular, such analysis has
shown that in steady-state conditions quantum-transport corrections are negligible and carrier
transport is mainly semiclassical. Moreover, the role of carriercarrier interaction in providing
an efcient coupling between injector/collector and active region, as well as in improving the
efciency of the phonon cascade in far-IR structure, has been investigated.
For the second class of devices, we addressed the relevant issue of a quantum treatment of
charge transport in systems with open boundaries (contacts), when studying and simulating an
at least two-terminal device. More specically, our analysis has shown that the conventional
Wigner-function formalism may lead to unphysical results, such as the injection of coherent
superpositions of states from the device spatial boundaries, in clear contrast to the idea of a
thermal (i.e. diagonal) injection from the electrical contacts. We have shown how this basic
limitation can be removed by introducing a generalization of the standard Wigner-function
formulation able to describe the incoherent nature of carrier injection. The latter constitutes a
rigorous derivation of the phenomenological injection models commonly used in the simulation
of open quantumdevices. Less rened approaches, which are however more affordable from
the computational point of view, have also been discussed. Here the emphasis has been put on
replacing an open boundary model (injection/loss rates between the device and the reservoirs)
with a closed boundary scheme, both within a semiclassical and within a quantum-mechanical
picture.
Acknowledgments
We wish to thank Remo Proietti Zaccaria for his signicant contribution to part of the work
reviewed in this paper. We are grateful to Massimo Fischetti, Carlo Jacoboni, Tilmann Kuhn,
Salvatore Savasta and Alessandro Tredicucci for stimulating and fruitful discussions.
References
[1] Esaki L and Tsu R 1970 IBM J. Res. Dev. 14 615
[2] Capasso F 1987 Science 235 1726
Capasso F 1987 Heterojunction Band Discontinuities: Physics and Device Applications ed F Capasso and
G Margaritondo (Amsterdam: North-Holland)
[3] Luryi S, Xu J and Zaslavsky A (ed) 2004 Future Trends in Microelectronics (Hoboken: Wiley)
[4] See, e.g. Jacoboni C and Lugli P 1989 The Monte Carlo Method for Semiconductor Device Simulations (Wien:
Springer)
[5] See, e.g. Shah J (ed) 1992 Hot Carriers in Semiconductor Nanostructures: Physics and Applications (Boston:
Academic)
Sch oll E (ed) 1998 Theory of Transport Properties of Semiconductor Nanostructures (London: Chapman
and Hall)
[6] See, e.g. Rossi F and Kuhn T 2002 Rev. Mod. Phys. 74 895950 and references therein
[7] See, e.g. Capasso F (ed) 1990 Physics of Quantum Electron Devices (Berlin: Springer)
Kelly M J 1995 Low-dimensional semiconductors (New York: Oxford)
[8] See, e.g. Frensley W 1990 Rev. Mod. Phys. 62 74591 and references therein
[9] Rossi F, Di Carlo A and Lugli P 1998 Phys. Rev. Lett. 80 334851
[10] Proietti Zaccaria R and Rossi F 2003 Phys. Rev. B 67 113311-14
[11] See, e.g. DiVincenzo D P and Bennet C 2000 Nature 404 24755 and references therein
[12] See, e.g. Jacak L, Hawrylak P and Wojs A 1998 Quantum Dots (Berlin: Springer)
[13] Zanardi P and Rossi F 1998 Phys. Rev. Lett. 81 47525
[14] Biolatti E, Iotti R C, Zanardi P and Rossi F 2000 Phys. Rev. Lett. 85 564750
[15] Rossi F 2004 IEEE Trans. Nanotechnol. 3 16572
[16] Faist J, Capasso F, Sivco D L, Sirtori C and Cho A Y 1994 Science 264 5536
[17] Gmachl C, Capasso F, Sivco D L and Cho A Y 2001 Rep. Prog. Phys. 64 1533601 and references therein
[18] See, e.g. Rossi F, Brunetti R and Jacoboni C 1992 Hot Carriers in Semiconductor Nanostructures: Physics and
Applications ed J Shah (Boston: Academic) p 153
[19] See, e.g. Kadanoff L P and Baym G 1962 Quantum Statistical Mechanics (New York: Benjamin)
[20] See, e.g. Keldysh L V 1965 Sov. Phys. JETP 20 101822
[21] See, e.g. Haug H and Jauho A-P 1996 Quantum Kinetics in Transport and Optics of Semiconductors
(Berlin: Springer)
[22] Fischetti M V 1999 Phys. Rev. B 59 490117 and references therein
[23] Levine B F 1993 J. Appl. Phys. 74 R181
[24] See, e.g. Haug H and Koch S W (ed) 1990 Quantum Theory of the Optical and Electronic Properties of
Semiconductors (Singapore: World Scientic)
[25] Kuhn T 1998 Theory of Transport Properties of Semiconductor Nanostructures ed E Sch oll (London: Chapman
and Hall) p 173
[26] Sirtori C, Kruck P, Barbieri S, Collot P, Nagle J, Beck M, Faist J and Oesterle U1998 Appl. Phys. Lett. 73 34868
[27] Iotti R C and Rossi F 2003 Phys. Status Solidi (b) 238 4629
[28] Gornkel V B, Luryi S and Gelmont B 1996 IEEE J. Quantum Electron. 32 19952003
Friedman L and Soref R A 1998 J. Appl. Phys. 83 34805
[29] Harrison P 1999 Appl. Phys. Lett. 75 28002
[30] See, e.g. Iotti R C and Rossi F 2000 Appl. Phys. Lett. 76 22657
[31] Iotti R C and Rossi F 2001 Appl. Phys. Lett. 78 29024
[32] Iotti R C and Rossi F 2001 Phys. Rev. Lett. 87 146603-14
[33] Rossi F, Poli P and Jacoboni C 1992 Semicond. Sci. Technol. 7 101735
[34] Compagnone F, Manenti M, Di Carlo A and Lugli P 2002 Physica B 314 33640
[35] Goodnick S M and Lugli P 1988 Phys. Rev. B 37 257888
[36] Spagnolo V, Scamarcio G, Schrenk W and Strasser G 2004 Semicond. Sci. Technol. 19 S1102
Spagnolo V, Scamarcio G, Page H and Sirtori C 2004 Appl. Phys. Lett. 84 36902
[37] Sirtori C, Capasso F, Faist J, Hutchinson AL, Sivco DLand Cho AY1998 IEEEJ. QuantumElectron. 34 17229
[38] Iotti R C and Rossi F 2003 J. Comput. Electron. 2 1915
[39] Eickemeyer F, Reimann K, Woerner M, Elsaesser T, Barbieri S, Sirtori C, Strasser G, M uller T, Bratschitsch R
and Unterrainer K 2002 Phys. Rev. Lett. 89 47402-14
[40] See, e.g. Shah J 1996 Ultrafast Spectroscopy of Semiconductors and Semiconductor Nanostructures (Berlin:
Springer)
[41] See, e.g. Koch S W (ed) 1996 Microscopic Theory of Semiconductors: Quantum Kinetics, Connement and
Laser (Singapore: World Scientic)
[42] Sargent M III, Scully M O and Lamb W E 1974 Laser Physics (London: Addison Wesley)
[43] See, e.g. Kalos M H and Whitlock P A 1986 Monte Carlo Methods (New York: Wiley)
[44] Portolan S, Iotti R C and Rossi F 2004 Semicond. Sci. Technol. 19 S1079
[45] Portolan S, Iotti R C and Rossi F 2004 Monte Carlo Methods Appl. 10 5319
[46] K ohler R, Iotti R C, Tredicucci A and Rossi F 2001 Appl. Phys. Lett. 79 39202
[47] Iotti R C and Rossi F 2002 Physica E 13 7158
[48] K ohler R, Tredicucci A, Beltram F, Beere H E, Lineld E H, Davies A G, Ritchie D A, Iotti R C and Rossi F
2002 Nature 417 1569
[49] Tredicucci A, Gmachl C, Capasso F, Sivco D L, Hutchinson A L and Cho A Y 1999 Appl. Phys. Lett. 74 63840
Tredicucci A, Gmachl C, Wanke M C, Capasso F, Hutchinson A L, Sivco D L, Chu S N G and Cho A Y 2000
Appl. Phys. Lett. 77 22868
Colombelli R, Capasso F, Gmachl C, Hutchinson A L, Sivco D L, Tredicucci A, Wanke M C, Sergent A M and
Cho A Y 2001 Appl. Phys. Lett. 78 26202
[50] See, e.g. Lake R and Datta S 1992 Phys. Rev. B 45 667085
Rivas C et al 2001 Appl. Phys. Lett. 78 8146
[51] Frensley W R 1986 Phys. Rev. Lett. 57 28536
[52] See, e.g. Kluksdahl N C, Kriman A M, Ferry D K and Ringhofer C 1989 Phys. Rev. B 39 772035
[53] See, e.g. Pascoli M, Bordone P, Brunetti R and Jacoboni C 1998 Phys. Rev. B 58 35036
Bordone P, Pascoli M, Brunetti R, Bertoni A, Jacoboni C and Abramo A 1999 Phys. Rev. B 59 30609
Jacoboni C, Brunetti R and Monastra S 2003 Phys. Rev. B 68 125205-19
[54] See, e.g. Bertrand P, Deremus J P, Izrar B, Nguyen VT and Feix MR1983 Phys. Lett. A94 4157 and references
therein
[55] Proietti Zaccaria R, Iotti R C and Rossi F 2004 Appl. Phys. Lett. 84 13941
[56] Proietti Zaccaria R, Iotti R C and Rossi F 2004 Phys. Rev. B 70 195311-18

Microscopic Theory of Semiconductor-Based Optoelectronic Devices

Încărcat de

Informații document

Descriere originală:

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Microscopic Theory of Semiconductor-Based Optoelectronic Devices

Încărcat de

Drepturi de autor:

Formate disponibile

Microscopic theory of semiconductor-based optoelectronic devices

, within each subband .

can be described by a set

denoting phenomenological scattering rates connecting levels

are obtained by averaging the microscopic transition probabilities P

over the initial

and k. It follows that their values are strongly affected

is the sum of the non-interacting carrier and phonon Hamiltonians

may include various interaction mechanisms, e.g.

the corresponding energy, the non-

can be evaluated exactly

, e.g. phonons, photons, partner carriers, etc.

in (3.21) weighted by the Pauli factor of the nal state.

1) the Boltzmann transport

(r) r|, (4.1)

(r) in this structure are also labelled by the in-plane wavevector k.

. In our MQW structure, is a global quantum number labelling single-particle

. Contrary to the macroscopic model in equation (2.1), this formulation provides a

) and the corresponding electron charge e

, = 1, 2, 3) as a function of time with () and

as well as on the carrier distributions f

is the carrier-photon coupling term.

(r) r| denoting the single-particle wavefunction of state ) to the SBE in (3.14). In

(r) r| is, in general, different from the basis set

. It is easy to show that if

), as requested. Indeed, if we nowintegrate equation (5.14) over the real-

. We nally stress that, by replacing the

), the phenomenological injection model in (2.4) is

describe electronic transitions within the simulated region induced

entering equation (5.17). In the

= 0), the steady-state solution of the conventional

, i.e. the carrier distribution inside the device coincides

) to the new collision operator in (5.17).

> 0). What is

, the detailed-balance principle should be replaced by a

. Starting from this

). Contrary to the time evolution in (a), here at very short

S-ar putea să vă placă și