Professional Documents
Culture Documents
2
2m
2
(r) qV (r)(r) = (r) (2.1)
The eigenstates are lled according to Fermis function, and their squared
amplitudes give the charge density, taking into account the spin degeneracy:
n(r) = 2
i
|
i
(r)|
2
f((
i
E
F
)/kT) (2.2)
where:
f(x) = (1 + exp(x))
1
(2.3)
The calculated charge density is then used in Poissons equation:
.(
s
(r)) = q(n(r) + N(r)) (2.4)
and the potential is substituted back into Equation (1) until convergence.
Once the potential and charge proles have been calculated, a suitable transport
model is used.
For devices far from equilibrium, the NEGF has become the method of choice.
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 14
For closed systems, (channel) the governing equation is
[EI H] {} = 0 (2.5)
Where H is the unperturbed Hamiltonian of the system, E is the energy of
electron, I is an identity matrix, is the eigenstate or the electron wave function.
If we consider an open system, we need to modify the governing equation as
following
[EI H ] {} = {S} (2.6)
Here, is the self-energy of dierent processes such as the contacts or the
scattering centers within the device. S is the source term which tells us how
electrons are getting into the system from outside or the contact.
Figure 2.1: System Contact with Channel
So, the wave function can be calculate as
{} = [G] {S} (2.7)
And
[G] = [EI H ]
1
(2.8)
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 15
G is called Greens function To calculate , one has to formulate the Hamil-
tonian matrix, H; then insert the self-energy, ; calculate the Greens function, G
and lastly multiply G with the source term. Rigorous denition and justication
are covered in the next section.
2.3 Greens functions
Discrete Schrodinger equation:
H |n = E |n (2.9)
We divide the Hamiltonian and wavefunction of the system into contact
(H
1,2,
|
1,2
) and device (H
d
, |
d
) subspaces:
We dene the Greens function
1
:
(E H)G(E) = I (2.10)
2.3.1 Why do we want to calculate the Greens function?
The Greens function gives the response of a system to a constant perturbation
|v in the Schrodinger equation:
H | = E | + |v (2.11)
The response to this perturbation is:
(E H) | = |v (2.12)
| = G(E) |v (2.13)
Why do we need the response to this type of perturbation? Well, it turns out
that its usually easier (see next section) to calculate the Greens function than
1
Others may (and do) use the opposite sign.
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 16
solve the whole eigenvalue problem
1
and most (all for the one-particle system)
properties of the system can be calculated from the Greens function. e.g., the
wavefunction of the contact |
2
can be calculated if we know the wavefunction
on the device |
d
. From third row of Eq. 2:
H
2
|
2
+
2
|
d
= E |
2
(2.14)
(E H
2
) |
2
=
2
|
d
(2.15)
|
2
= g
2
(E)
2
|
d
(2.16)
where g
2
is the Greens function of the isolated contact 2 ((E H
2
)g
2
= I).
It is important to note that since we have an innite system, we obtain two
types of solutions for the Greens functions
2
, the retarded and the advanced
3
solutions corresponding to outgoing and incoming waves in the contacts.
Notation: We will denote the retarded Greens function with G and the ad-
vanced with G
+
(and maybe G
R
and G
A
occasionally).Here, CAPITAL G denotes
the full Greens function and its sub-matrices G
1
, G
d
, G
1d
etc. Lowercase is used
for the Greens functions of the isolated subsystems, e.g., (E H
2
)g
2
= I.
Note that by using the retarded Greens function of the isolated contact (g
2
)
in Eq. 9 we obtain the solution corresponding to a outgoing wave in the contact.
Using the advanced Greens function (g
+
2
) would give the solution corresponding
to an incoming wave.
2.3.2 Self-Energy
The reason for calculating the Greens function is that it is easier that solving
the Schrodinger equation. Also, the Greens function of the device (G
d
) can be
1
Especially for innite systems.
2
When the energy coincides with energy band of the contacts there are two solutions cor-
responding to outgoing or incoming waves in the contacts.
3
In practice these two solutions are usually obtained by adding an imaginary part to the
energy. By taking the limit to zero of the imaginary part one of the two solutions is obtained.
If the limit 0
+
is taken the retarded solution is found, 0
+
1
G
1d
+ (E H
d
)G
d
2
G
2d
= I (2.18)
(E H
2
)G
2d
2
G
d
= 0 (2.19)
We can solve Eqs. 11 and 13 for G
1d
andG
2d
:
G
1d
= G
1
1
G
d
(2.20)
G
2d
= G
2
2
G
d
(2.21)
substitution into Eq. 12 gives:
1
+ g
1
1
G
d
+ (E H
d
)G
d
+
2
g
2
2
G
d
= I (2.22)
from which (G
d
) is simple to nd:
G
d
= (E H
d
1
2
)
1
(2.23)
where
1
=
+
1
g
1
1
and
2
=
+
2
g
2
2
are the so called self-energies.
Loosely one can say that the eect of the contacts on the device is to add
the self-energies to the device Hamiltonian since when we calculate the Greens
function on the device we just calculate the Greens function for the eective
Hamiltonian H
effective
= H
d
+
1
+
2
.However, we should keep in mind that we
can only do this when we calculate the Greens function. The eigen-values and
-vectors of this eective Hamiltonian are not quantities we can interpret easily.
For normal contacts, the surface Greens functions g
1
and g
2
used to calculate
the self-energies are usually calculated using the periodicity of the contacts, this
method is described in detail in appendix B of [4] and in section 3 of [16].
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 18
2.3.3 The spectral function
Another important use of the Greens function is the spectral function:
A = i(G G
+
) (2.24)
which gives the DOS and all solutions to the Schrodinger equation!
To see this we rst note that for any perturbation |v we get two solutions
(
R
_
and
A
_
) to the perturbed Schrodinger equation:
(E H) |) = |v (2.25)
from the advanced and retarded Greens functions:
R
_
= G|v (2.26)
A
_
= G
+
|v (2.27)
The dierence of these solutions(
R
_
A
_
= (G G
+
) |v = iA|v) is a
solution to the Schrodinger equation:
(E H)(
R
_
A
_
) = (E H)(G G
+
) |v = (I I) |v = 0 (2.28)
which means that | = A|v is a solution to the Schrodinger equation for
any vector |v.
To show that the spectral function actually gives all solutions to the Schrodinger
equation is a little bit more complicated and we need the expansion of the Greens
function in the eigenbasis:
G =
1
E + i H
=
k
|k k|
E + i
k
(2.29)
where the is the small imaginary part (see footnote 3),|v s are all eigenvec-
tors
1
to H with the corresponding eigenvalues
k
. Expanding the spectral function
1
Normalized!
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 19
in the eigenbasis gives:
A = i(
1
E + i H
1
E i H
) (2.30)
A = i
k
|k k| (
1
E + i
k
1
E i
k
) (2.31)
A =
k
|k k|
2
(E
k
)
2
+
2
(2.32)
where is our innitesimal imaginary part of the energy. Letting go to zero
gives:
A = 2
k
(E
k
) |k k| (2.33)
(here (E
k
) is the delta function) which can be seen since
2
(E
k
)
2
+
2
goes to
zero everywhere but at E =
k
,integrating over E (with a test function) gives the
2(E
k
) factor. Eq. 27 shows that the spectral function gives us all solutions
to the Schrodinger equation.
2.4 Response to an incoming wave
In the non-equilibrium case, reservoirs with dierent chemical potentials will in-
ject electrons and occupy the states corresponding to incoming waves in the con-
tacts. Therefore, we want to nd the solutions corresponding to these incoming
waves.
Consider contact 1 isolated from the other contacts and the device. At a
certain energy we have solutions corresponding to an incoming wave that is totally
reected at the end of the contact. We will denote these solutions with |
1,n
where 1 is the contact number and n is a quantum number (we may have several
modes in the contacts). We can nd all these solutions from the spectral function
a
1
of the isolated contact (as described above).
Connecting the contacts to the device we can calculate the wavefunction on
the whole system caused by the incoming wave in contact 1. To do this we note
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 20
that a wavefunction should be of the form
2.5 Charge density matrix
In the non equilibrium case we are often interested in two quantities: the current
and the charge density matrix. Lets start with the charge density (which allows
us to use a self-consistent scheme to describe charging).
The charge density matrix is dened as:
=
k
f(k, ) |
k
k
| (2.34)
where the sum runs over all states with the occupation number f(E
k
, ) (pure
density matrix) (note the similarity with the spectral function A, in equilibrium
you nd the density matrix from A and not as described below). In our case, the
occupation number is determined by the reservoirs lling the incoming waves in
the contacts such that:
f(E
k
,
1
) =
1
1 + e
E
k
1
k
B
T
(2.35)
is the Fermi-Dirac function with the chemical potential
1
and temperature
(T) of the reservoir responsible for injecting the electrons into the contacts.
The wavefunction on the device given by an incoming wave in contact 1 (see
Eq. 32) is:
|
d,k
= G
d
+
1
|
1,k
(2.36)
Adding up all states from contact 1 gives:
d
[contact1] =
_
E=
dE
k
f(E,
1
)(E E
k
) |
d,k
d,k
| (2.37)
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 21
2.6 Boundary conditions
In quantum transport, the treatment of boundary conditions requires signi-
cantly more care than in classical physics due to the nonlocality of quantum
mechanics.[20],[11],[23],[17]. A common problem is to dene Ohmic leads. In the
context of NEGF, we may dene leads as Ohmic if the current is controlled by
the interior of the device rather than by serial resistances or interface states. This
denition has several implications. First, there must be a smooth transition in
the density of states of the leads and the device. Secondly, the same scattering
mechanisms must act within the leads and the device. Thirdly, the carrier dis-
tribution within the leads must be a suitably accelerated Fermi distribution to
reect current conservation. All of these conditions are necessary to avoid quan-
tum mechanical reections and pile up of charge at the interface.[21],[24] This is
illustrated in Figure . Figure shows a full NEGF calculation of the carrier density
in a homogeneously n-doped piece of GaAs for zero bias as a function of position.
Zero bias implies the quantum mechanical current from left to right to equal the
current from right to left. If the leads are treated ballistically, the carriers ac-
cumulate near the device boundaries due to the resistance they meet within the
medium. Only by employing the same scattering mechanisms and self-energies
everywhere, the formalism yields a constant carrier density throughout the sys-
tem. Figure illustrates the calculated carrier density in a biased n++-i-n+ GaAs
structure. If the carrier distribution within the leads is assumed to be in equilib-
rium, one obtains an articial pinch-o behavior; there is a depletion of carriers
near the source side and an accumulation near the drain side of the device.
2.7 The semiclassical limit
Does the NEGF formalism reduce to the semiclassical limit when quantum eects
play no role? [22],[15] For a realistic device structure, this is dicult to prove
rigorously but the following example provides practical evidence that, indeed, the
answer is armative.
Consider a symmetric n-i-n GaAs diode at room temperature and for zero bias.
Since such a device contains neither quantum wells nor barriers, one expects the
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 22
semi-classical Boltzmann equation to adequately describe the carrier density, as-
suming one includes impurity and phonon scattering in the standard fashion and
includes electron-electron scattering at least within the Hartree approximation
via the Poisson equation. We have performed a charge self-consistent NEGF
calculation that takes into account the same type of scattering mechanisms. As
shown in Figure , the NEGF electron density mimics faithfully the semiclassical
results.
Figure 2.2: (Color online) Left: Calculated carrier density for a homogeneously
doped semiconductor at zero bias, attached to leads with an equilibrium distri-
bution. The expected homogeneous density is obtained only if the leads include
the same type of scattering self energies as the device (red curve). Otherwise one
obtains articial charge accumulation (blue curve). Right: GaAs n-i-n structure
at room temperature under bias with asymmetric doping prole as indicated by
the grey lines.Once a current is owing. the charge distribution within the leads
must be a suitably shifted Fermi distribution that reects global current con-
servation (results shown by thc red curve). Otherwise, NEGF calculations yield
articial pinch-o eects (blue lines)
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 23
Figure 2.3: Comparison of fully self-consistent NEGF and semiclassical carrier
dynamics calculations for a standard GaAs resistor (50 nm n-i-n structure) at
zero bias and room temperature
Figure 2.4: (Color onlinc) Carrier dynamics calculation for 50 nm n-i-n structure
at room tcmperature with a 12 nm InGaAs quantum well as intrinsic zone at-
tached to eld-free GaAs leads of the same n-density.The applicd voltage is 150
mV across thc structure.Red curve: calculalion in terms of charge-self-consistent
semiclassical Boltzmann equation.Blue curve: Calculation in terms of strictly bal-
listic NEGF, equivalent to thc solution of Schrodinger equation of open system.
Black curve: Fully self-consistent NEGF calculation
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 24
2.8 Quantum well structures: Why NEGF is es-
sential
A nice illustration of the power of NEGF can be obtained by calculating the local
carrier distribution in a biased n-i-n structure that contains a quantum well. This
is illustrated in Figure for a n-i-n structure with a 12 nm InGaAs quantum well
in the intrinsic zone. The applied voltage across the 50 nm structure is 150
mV and we show 3 results. The red curve shows the semiclassical Boltzmann
results and yields the well-known accumulation of charge near the quantum well
barriers. This method completely ignores the existence of quantum mechanical
bound states and reects the classical Thomas-Fermi density. The blue curve
illustrates another limiting case, namely the solution of the Schrdinger equation in
the absence of any scattering. This reects a strictly coherent, energy conserving,
ballistic transport. In this case, the carrier density within the device is fully
determined by the overlap of the lead wave functions with the device. Since
there are no lead carriers below the GaAs band edge, the quantum well states
in the intrinsic region remain unoccupied. Thus, the electron density within
the quantum well only stems from (continuum state type) lead electrons which
explains the oscillatory density in the intrinsic region. Finally, the black curve
represents the full NEGF calculation. The inelastic scattering processes lead to
a capture of carriers into the quantum well states and lead to a carrier density in
the intrinsic zone that lies in between the semiclassical and the strictly ballistic
quantum mechanical calculations.
This result can be further illustrated by plotting the energy resolved density
for the case of zero bias. Figure depicts this density for a strictly ballistic
calculation that actually represents a NEGF calculation in the limit of all impu-
rity, phonon, and interface scattering self energies set equal to zero. Note that
the Poisson equation is still solved self-consistently with the Schrdinger equation
even in this case. Figure , on the other hand, shows a complete NEGF calculation
that clearly illustrates the carrier capture into the rst and second bound state of
the quantum well. Due to the charging eects caused by the capture, the bottom
of the quantum well raises in energy so that both states actually form resonances
that slightly overlap with the lead states.
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 25
Figure 2.5: Contour graph of calculated energy resolved electron density for 50
nm n-i-n structure at room temperature with a 12 nm lnGaAs quantum well as
intrinsic zone for zero applied bias. The density scale is the analogous to the
one in Fig. 5. but for lower doping. (a) Strictly ballistic NEGF calculation (no
scattering included). (b) Fully self-consistent NEGF calculation
Figure 2.6: Contour graph of calculated energy resolved electron density for 50
nm n-i-n structure at room temperature with a 12 nm InGaAs quantum well as
intrinsic zone for zero applied bias. (a) Fully self consistent NEGF calculation.
(b) NEGF calculation with coupling
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 26
2.9 Limits and simplications in NEGF
As mentioned in the introduction, the NEGF formalism guarantees conserva-
tion of important physical principles only if all Greens functions are evaluated
exactly from a many-body standpoint. It is a very signicant weakness of the
method that even plausible approximations can fail badly. Generally, it is di-
cult to introduce simplications that do not violate basic conservation laws. As
an example, we discuss the so-called decoupling approximation.[10] The NEGF
formalism couples the energy of states with their occupation via 4 coupled integro-
dierential equations. If the carrier density is not too high, it seems plausible to
decouple the equations of and . This approximation can lead to a violation of
Paulis principle, however, since there is no mechanism that prevents the occur-
rence of over-occupied states (i.e. states with occupancy higher than permitted
by the Pauli principle).[10] To exemplify this situation, we consider the same
GaAs-InGaAs-GaAs n-i-n structure as before, but with a slightly higher carrier
concentration in the n-region. Figure shows the energy resolved carrier density
of the n-i-n structure for zero bias, as calculated by the full NEGF approach. By
contrast, Figure shows the result of the decoupling of Greens functions, leading
to unphysically high occupation of the lowest bound state.
2.10 THz quantum cascade lasers: a classics for
NEGF
An important question that we have not addressed so far is whether fully self-
consistent NEGF calculations actually agree with experiment. We have applied
this formalism to GaAs/AlGaAs THz QCLs and included impurity, phonon, inter-
face roughness scattering in the self-consistent Born approximation. In addition,
the electron-electron scattering has been included both in the Hartree approxi-
mation as well as within the so-called GW approximation. For details, we refer
to [18]. Importantly, the calculations contain no tting parameter. Figure 8 de-
picts the calculated current-voltage characteristics of such a QCL for a particular
sheet doping density, together with experimental data.[2] As one see, theory and
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 27
experiment agree very well with one another up to the voltage where lasing starts
and both thermal as well as hot electron eects become relevant that have not
been included in the calculations.
Figure 2.7: Comparison between experimental (Ref. [27]) and calculated (Ref.
([13]) current-voltage characteristics for AIGaAs/GaAs quantum cascade struc-
ture
2.11 Comparison between NEGF and Monte -
Carlo
2.11.1 Monte-Carlo
Monte Carlo simulation performs calculation by building models of possible re-
sults by substituting a range of values-a probability distribution-for any factor
that has inherent uncertainty. It then calculates results over and over, each time
using a dierent set of random values from the probability functions. Depending
upon the number of uncertainties and the ranges specied for them, a Monte
Carlo simulation could involve thousands or tens of thousands of recalculations
before it is complete. Monte Carlo simulation produces distributions of possible
outcome values. By using probability distributions, variables can have dierent
probabilities of dierent outcomes occurring. Probability distributions are a much
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 28
more realistic way of describing uncertainty in variables of a device performance
analysis.
Through the Non-Equilibrium Greens Function (NEGF) formalism, quantum-
scale device simulation can be performed with the inclusion of electron-phonon
scattering. However, the simulation of realistically sized devices under the NEGF
formalism typically requires prohibitive amounts of memory and computation
time. Two of the most demanding computational problems for NEGF simulation
involve mathematical operations with structured matrices called semiseparable
matrices.
Nonequilibrium Greens function method is a very general scheme to include
coherent quantum mechanics and incoherent scattering with phonons and device
imperfections self-consistently. However, it is numerically very demanding and
cannot be used for systematic device parameter scans. For this reason, we also
implement the approximate, numerically e?cient ensemble Monte Carlo method
and assess its applicability on the above mentioned transport problems.
A research paper [2] shows that the approximate treatment of coherent tunnel-
ing and leakage into continuum states limits the applicability of the EMC method
on transport regimes that are dominated by incoherent scattering. When all de-
vice states that contribute to transport are clearly non-degenerate, results of the
current density obtained by the EMC method quantitatively agree with NEGF
results and experiment. Also the simulated spectral gain pro?le is in good agree-
ment for both methods. This is in particular important because the numerical
load of NEGF calculations exceeds the load of the EMC method tremendously
and typically prohibits a systematic improvement of QCL designs.
2.12 Why nonequilibrium Green functions?
Let us briey describe its main features:
The method has as its main ingredient the Green function, which is a func-
tion of two spacetime coordinates. From knowledge of this function one can
calculate time-dependent expectation values such as currents and densities,
electron addition and the total energy of the system.
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 29
In the absence of external elds the nonequilibrium Green function method
reduces to the equilibrium Green function method which has had imporatnt
applications in quantum chemistry.
Nonequilibrium Green functions can be applied to both extended and nite
systems.
The nonequilibrium Green function can handle strong external elds non-
perturbatively. The electron-electron interactions are taken into account by
innite summations.
The approximations within the nonequilibrium Green function method can
be chosen such that macroscopic conservation laws as those of particle num-
ber, momentum and angular momentum are automatically satised
Dissipative processes and memory eects in transport that occur due to
electron-electron interactions and coupling of electronic to nuclear vibra-
tions can be clearly diagrammatically analyzed
2.13 Conclusions
The NEGF formalism provides the framework of choice for consistent carrier dy-
namics calculations of open nanosystems where quantum eects and incoherent
scattering play a comparable role. When implemented with care, it reproduces
the results of the semiclassical Boltzmann equation in the limit where quantum
eects such as resonant tunneling and interference are unimportant. By deni-
tion, it also reproduces the solutions of the Schrdinger or Lippmann-Schwinger
equation when inelastic scattering is turned o. A disadvantage of the method
is the fact that charge and current conservation, and even Paulis principle are
not automatically satised once seemingly reasonable approximations are intro-
duced. Scattering vertices must necessarily be taken into account to innite
order, for example, to strictly obey current conservation and it is dicult to
achieve a numerically satisfactory convergence. Approximations are unavoidable,
though, once one seeks predictions for realistic nano-devices, simply due to the
CHAPTER 2. NON-EQUILIBRIUM GREENS FUNCTIONS 30
excessive numerical eort required to solve the full set of Keldysh equations self-
consistently. In fact, it will take some time before quantitative NEGF solutions
for time-dependent quantum transport calculations become numerically feasible.
In comparison with semiclassical calculations, much more eort is required to
properly take into account the physics of contacts and the contact-device cou-
pling. This is a consequence of the nonlocal nature of quantum mechanics and
the nature of scattering solutions in open quantum systems.
Chapter 3
QDIPs density of states modeling
3.1 DWELL QDIP Structure
A theoretical model of pyramidal-shaped InAs quantum dots placed in an InGaAs
quantum well, which is buried in a GaAs matrix, is shown in [[7]]. The model of
the DWELL is based on a Bessel function expansion of the wave function. The
model can estimate the ground state of the quantum dot. For the higher states
in the quantum well, the model has to be modied to account for the free motion
of electrons perpendicular to the growth direction.
The DWELL detector grown by Molecular Beam Epitaxy, reported in Ref [1],
consists of a ten-period active region of 6nm In
0.15
Ga
0.85
As, 2.4 ML of InAs, 6
nm In
0.15
Ga
0.85
As, and 50 nm GaAs, as shown in Fig. 3.1. The quantum dots
are placed in the In
0.15
Ga
0.85
As quantum well which is in turn surrounded by the
GaAs region.
The TEM image of the DWELL heterostructure is shown in following Fig 3.2.
[1]
The darkest region is the InAs quantum dot. The quantum dots are situated
in the upper half of the quantum well and have a conical shape whose base
dimension is of 11 nm and height is of 6nm. The QD material InAs is deposited
over the substrate and due to the lattice mismatch between deposited material
and substrate, the strain is built up gradually. After a critical thickness (2.4 ML)
is reached, the two-dimensional growth changes into a three-dimensional one and
31
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 32
Figure 3.1: Cross-section schematic of a 10 layer InAs/InGaAs quantum dot in a
well detector,
Figure 3.2: Cross-section TEM image of a single QD layer of DWELL [14]
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 33
dislocation free QD islands begin to grow.
Figure 3.3: Cross-section TEM image of an InAs/InGaAs DWELL Heterostruc-
tures [14]
Figure 3.4: Theoretical Modeling of DWELL QDIP Structure
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 34
For numerical analysis we have modeled the DWELL device as shown in Fig
3.1. 60% of the band gap dierence between InAs and GaAs is counted as the
conduction band oset .The band osets calculated are 477 meV between InAs
and In
0.15
Ga
0.85
As and 93 meV between In
0.15
Ga
0.85
As and GaAs. The conduc-
tion band edge of the In0.15Ga0.85As is selected as reference energy level. A
linear interpolation between two binary values is used to calculate the eective
masses in the dierent materials. The eective masses used for GaAs, InAs and
In
0.15
Ga
0.85
As are 0.067, 0.027and 0.061 (in terms of electron mass) respectively.
Figure 3.5: Conduction band osets and energy levels of QDWELL [14]
For analysis, the device is thought of consisting of array of identical cylinders,
where each cylinder contains one quantum dot. To calculate Hamiltonian of the
device, a cross section along the cylinder axis is taken and disintegrated into a
large number of equally spaced grids. The nite dierence method is used to
solve the dierential equation governing Greens function. The retarded Greens
function of the system is dened as
[E H
op
r
(E)]G
r
(x, x
, y, y
, E) = (x x
)(y y
) (3.1)
Here E is the total energy of electron,
r
is the self energy and H
op
, the Hamil-
tonian operator of the system, is given by
H
op
= .
2
2m(x, y)
+ V (x, y) (3.2)
Here V(x,y) is the potential energy seen by the electron and m(x,y) is the
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 35
Figure 3.6: Cross section of a cylinder is disintegrated into a number of grids
eective mass. The diagonal elements of spectral function is given by
diag(A(x, y, E)) = 2Im[G
r
(x, y; E)] (3.3)
The density of states, which is the number of states per unit energy per unit
volume, is given by
N(E) =
1
2
Tr(A(x, y; E)) (3.4)
3.2 Numerical Analysis
The quantum dot photodetectors under our analysis have an estimated dot den-
sity of 5 10
10
cm
2
, and the average spacing between two adjacent dots is about
60nm. Due to this relatively large distance, in our simplied model of the quan-
tum dot photo-detector, the neighboring dots are assumed to be vertically and
laterally decoupled, and a quantum dot is modeled so as to be surrounded by
semi-innite contact composed of InGaAs and GaAs layer, and InAs wetting
layer. The contact can be thought of being a continuation of cylinder radius and
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 36
the quantum dot exists at the center of the cylinder.
Modeling quantum dot photo-detectors in such a way gives the benet of ex-
ploiting the property of translational invariance of the contact. The Hamiltonian
matrix for the device, which is tridiagonal and Hermitian, is formed by nite
dierence method and is given by
H =
_
_
_
_
_
_
_
_
_
x1
1
0 . . . 0
1
x2
2
. . .
.
.
.
0
.
.
.
.
.
.
.
.
.
Nx1
0 . . . . . .
Nx1
x(Nx)
_
_
_
_
_
_
_
_
_
Here,
x
(i) =
_
_
U
y1
+ 2tx
(i)
+ 2ty
1
ty
1
0 0
ty
1
U
y2
+ 2tx
(i)
+ 2ty
2
ty
2
0 0
.
.
. ty
2
.
.
.
.
.
.
ty
Ny1
0 ty
Ny1
U
y(Ny)
+ 2tx
(i)
+ 2ty
Ny
_
_
(3.5)
(
i) =
_
_
tx
1,i
0 0
0 tx
2,i
.
.
.
.
.
.
.
.
.
.
.
.
0 tx
Ny,i
_
_
(3.6)
3.3 Calculation of Self Energy:
To illustrate the self-energy calculation which accounts for the device leads, we
consider the eect of coupling the active device Hamiltonian to the drain. The
innite Hamiltonian and its Greens function can be partitioned as follows:
In the above equation, the subscript lead indicates innite block of matrices
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 37
(H matrix and G matrix). The matrix block we only care about is G
device
as we
are not interested in the Greens function within the Lead or Source. G
device
can
be expressed in terms of known quantities as follows:
G
device
= [EI h
device
r
]
1
(3.7)
Where the self energy term is
r
=
_
_
0 0
0 0
_
_
_
EI
Nx+1
0
EI
Nx+2
0
0
_
_
1 _
_
0
0 0
_
(3.8)
For evaluating the matrix product in above equation, we only need the rst
block of the inverse of the infnite matrix associated with the Lead. Moreover, the
diagonal blocks of this innite matrix are repeated due to translational invariance
within the Lead.
Nx
=
Nx+1
=
Nx+2
= (3.9)
Using this property, and partitioning the matrix, a close form of the matrix
for the rst block of the inverse (denoted by g
c
) of the innite matrix, can be
obtained as
I = g
c
[EI
Nx+1
c
] (3.10)
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 38
Once g
c
hasbeensolved, wehavetheselfenergyterm
r
=
_
_
g
c
+
0 0
0 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 g
c
+
_
_
(3.11)
t
x
is the coupling energy between adjacent grid points along x direction, and
is given by
t
x
=
2
2m
x
a
2
(3.12)
g
c
is the retarded greens function of a unit cell of the contact, and is solved
from the recursive relation
g
1
c
= ((E + i)I H
c
c
+
) (3.13)
It is noteworthy that only the last vertical slice of the device couples to the
Lead. Therefore, the self-energy for the Lead has a single nonzero block that
perturbs the last diagonal block of the device Hamiltonian. To solve for g
c
, a
basis transformation has to be performed. The eigenvectors of EI diagonalize
g
c
simultaneously. Therefore we change the basis from 2D real space to a basis
that is composed of the eigenvectors of EI (equivalent to a mode-space
transformation at the boundary). This reduces the equation related to gc to a
set of decoupled quadratic equations that can be solved for the diagonal entries
g
c
, in the transformed representation. It should be noted that each of these
equations results in two roots. The root representing outgoing waves is selected
as we are ultimately interested in obtaining the retarded Greens function for the
device. An inverse basis transformation is then applied to evaluate g
c
in 2D real
space. A similar procedure is invoked to solve for the self-energy part associated
with the Lead of other side. The nal size of the self-energy matrix is (N
x
N
y
)
2
for the real-space solution and (N
x
)
2
for the decoupled mode-space solution.
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 39
3.4 Computational Challenges
The device that we have dealt with is of GaAs/InGaAs/InAs heterostructure.
For numerical analysis in FDM (Finite Dierence Method), we have disintegrated
the device structure into a large number of grids. For instance, a 40nm by 30nm
cross section of the device was divided into near about 30000 grids. A pictorial
representation of the device with grid division is as follows:
Figure 3.7: Device structure with grid division
The actual number of grid is too enormous to show in picture. In simulation,
any two neighboring grids, (either in x direction or in y direction) are separated
by 0.2nm distance. Thus in the entire device, there were a total of 30,000 grids.
Along X axis - number of grid per row is
30nm
0.2nm
= 150 grids
Along Y axis - number of grid per row is
40nm
0.2nm
= 200 grids
Hence total no. of grid in the whole device = 200 x 150 = 30,000 We will
justify the reason of taking such a mammoth number of grids although the device
somewhere later, but for instance, lets take a look at what are the problems asso-
ciated with calculation that involve such a large number of grid points. In nite
dierence method, the matrix size of the Hamiltonian of a device is proportional
to the square of the total number of grid point representing the entire device. For
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 40
example, if we divide the device in 6 grids, the corresponding Hamiltonian will
be of 6 x 6 size. Each diagonal element of the Hamiltonian will correspond to a
unique grid points. An illustration of the same is as follows:
Figure 3.8: Device cross section
Corresponding
H =
_
_
_
_
_
_
_
_
_
_
_
g1 a 0 0
a g2 b 0
b g3 c
c g4 d
0 d g5 e
0 e g6
_
_
_
_
_
_
_
_
_
_
_
The diagonal value of the Hamiltonian corresponds to each of the grid point
taken on the device cross section. However, the o-diagonal elements are ap-
pearing due to coupling between the near-most grid points. Now, for the actual
device, on which we have taken a considerable number of grid point in order to
compute its property as precisely as possible, will yield a Hamiltonian having size
of 30,000 x 30,000. This is an incredibly large matrix which will require a huge
memory space. In fact, in a computer with moderate conguration, a MATLAB
program cant declare a matrix which exceeds 10,000 x 10,000 in size, let alone
any operation we can make with it. The built in memory reserved for a MAT-
LAB program overows while storing the data of such an enormous matrix. The
numerical analysis is thus stymied at the beginning of our thesis. We had to nd
a way out to circumvent the problem of memory run-out, and that constituted
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 41
how we confronted our very rst challenge.
3.5 How to confront the challenges
There are several ways out to solve the problem of memory run-out. We will
discuss them one by one, and also have a glance on the associated limitation that
each of the way poses on our ability to analyze the device appropriately.
3.5.1 Reduce the number of grid
This is probably the very rst (and the most naive, however!) idea that may come
across ones mind in such a case of limited computer memory. But it should be
kept in mind that the separation between adjacent grids would have to be such
that even the thinnest region of the device cross section is to be covered by
sucient number of grid points.
In our device, the thinnest region is the wetting layer of InAs, whose thickness
is of 0.5 nm. Hence, if grid separation is more than 0.2 nm, the wetting layer
will contain insucient number of grid, thereby the eect that the wetting layer
exerts on the device characteristics will not be properly reected in the simulated
model.
3.5.2 Non-uniform distribution of grid point
The non-uniform distribution of grid point throughout the device can be a smart
and intelligent way to make an economical use of available memory space while
incorporating the eect of all the region simultaneously. In fact, Comsol simu-
lator provides us with the feature of taking dense mesh points in some region
whereas less dense or less concentrated mesh points in some other region as per
the requirement of the user. However, in numerical analysis method performed
in MATLAB, we have to provide with formula that rightly ts for non-uniform
distribution of grid point. In contrast, the formula we could have laid our hand
on was derived for uniform distribution of grid point. We, thus, were left with
the choice of formulating a new equation for our own pursuit, or leave this way
of solution as an unviable one.
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 42
3.5.3 Disintegrating a large matrix into several smaller
matrices
Segregating a large matrix into several smaller ones can be another potential
solution to our problem. Our mathematical analysis frequently needed matrix
inversion operation, which is considered one of the most tedious and memory
expansive job for a calculating machine. Since the matrix we dealt with is incred-
ibly large in size, a direct command of matrix inversion (i.e. inv (Matrix)) would
denitely end up with memory overow. An alternative way of inverting a matrix
is as follows: Let, A is a matrix, and B is the inverse of the matrix A. So, A*B=
I, where I is an identity matrix of same size as A or B is. If a command in MAT-
LAB is given as follows: B=A/I; Then MATLAB actually performs a solution of
linear equation system which is much less hectic and time saving process.
For instance, if A=
_
2 7
1 5
_
Then A
1
can be calculated as follows:
A
1
= B =
_
b1 b2
b3 b4
_
From here, we can write:
_
2 7
1 5
_
_
b1 b2
b3 b4
_
=
_
1 0
0 1
_
So
_
2 7
1 5
_
_
b1
b3
_
=
_
1
0
_
That is, 2 b1 +7 b2 = 1 And, b1 + 5 b3 = 0
And also
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 43
_
2 7
1 5
_
_
b2
b4
_
=
_
0
1
_
That is, 2 b2 + 7 b4 = 0 And , b2 + 5 b4 = 1 Thus, instead of nding the
inverse matrix altogether, we can individually nd every element of the inverse
matrix and then arrange all the calculated elements in order.
3.5.4 Dealing with Sparse Matrix other than conventional
matrix
Sparse matrix are those, whose elements are mostly zero. A zero dominated sparse
matrix can easily be manipulated to save memory since its zero elements do not
need a separate memory space to occupy. The Hamiltonian that is concerned
in our device simulation is a sparse matrix. It contains non-zero elements along
the diagonal and o-diagonal position. The remaining elements are zero (unless
any special circumstances appear, like: boundary condition). In fact, the sparse
matrix springs o in almost all device simulation, since we usually consider a
tight binding model of the device (a model where wave function of electron at a
certain location is only inuenced by nearmost grid points, coupling occurs only
between neighboring grid points)
3.5.5 Using all the processors in a computer in a parallel
process
By default, a MATLAB program engage only one core in a computer for com-
putational purpose at a time. But special arrangement and coding can be done
so as to compel all the available processors simultaneously and in a parallel way,
which saves time to a great deal. A detail description of How self-parallelism can
be done in a MATLAB program can be viewed from the following website:
a) Parallel Computing Toolbox Perform parallel computations on multicore
computers, GPUs, and computer clusters
http://www.mathworks.com/products/parallel-computing/
b) Multicore - Parallel processing on multiple cores
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 44
http://www.mathworks.com/matlabcentral/leexchange/13775
3.6 Result- Our Findings
A group of researchers led by Professor Sanjay Krishna reported their Spectral
response measurements performed on the DWELL detectors with a Nicolet 870
FTIR spectrometer and a Keithley 428 current-amplier. [1] Figure 3.9 shows the
spectral response obtained from six detectors in which the width of the bottom
InGaAs well was varied from 10 to 60 angstrom. The top InGaAs cap layer was
unchanged to minimize changes in the dimensions of the dots. As expected, the
operating wavelength of the detector showed a monotonic red shift from 7.2 to 11
m from samples A to F. This is signicant since it provides us with a method of
controlling the operating wavelength of a QD detector. Note that in the sample
with the largest well, there is a broad shoulder on the shorter wavelength side,
which is possibly due to a transition from the ground state in the dot to a higher
lying state in the InGaAs quantum well.
Professor Krishna and his collaborators obtained the responsivity from a 15-
stack quantum DWELL detector. The measured responsivity was divided by a
factor of 4 to account for the scattering in the substrate. Far-infrared (FIR) spec-
tral response measurements were also undertaken in collaboration with Pereras
group [9]. The data were obtained using a Perkin-Elmer system 2000 FTIR with
two sets of beamsplitters and windows and were corrected by background spectra.
The resulting three-color response is shown in gure 3.10. The rst two
MIR peaks, i.e. 10 and 5 m, have previously been observed by the same re-
searchers group on similar detector structures [9]. The researchers group believes
that the peak at 10m(124meV < Ec) is probably a transition from a bound
state in the dot to a bound state in the quantum well, whereas the peak around
5m(250meV > Ec) is possibly a transition from a state in the dot to a quasi-
bound state close to the top of the well as shown in the inset to gure . A FIR
peak centred around 25m was also observed in these detectors and is shown in
gure 3.11. The researchers group believes that this peak could possibly be due
to transitions between two states in the QD since the calculated energy spacing
between the dot levels is about 50-60 meV (20 25m).
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 45
Figure 3.9: Progressive red shift in the peak operating wavelength of the detector
as the width of the bottom InGaAs is increased from 10 to 60 angstrom. The
spectre has been vertically displaced for clarity. [7]
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 46
Figure 3.10: Multicolour response in the mid-wave, long-wave and very long-
wavelength regimes with the associated transitions in the inset.[11]
Figure 3.11: The very long-wave infrared (VLWIR) response was observed till 80
K [23]
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 47
3.7 Comparing Experimental results with sim-
ulated ones
The reported structure in [8] has three color response with peaks at wavelengths of
5, 11, and 25m. The corresponding energy transitions E due to photon absorption
at these wavelengths are approximately 250, 113 and 50 meV.
E =
hc
e
eV (3.14)
E
1243
(m)
meV (3.15)
Our calculated DOS for DWELL structure with bottom quantum well of 6nm
(symmetric DWELL structure) is shown in Fig. 3.1. From the calculated DOS,
we get E0, E1, E2, E3 (corresponding to each abrupt rise in DOS prole) as
-270.3meV,-28meV, 64meV, and 112meV and results in E1 - E0, E2 - E1 and E3
- E2 as 242.3meV ( 250meV ), 92meV ( 124meV ) and 48meV ( 50meV ).
Thus it explains the tri-band operation satisfactorily.
Figure 3.12: DOS prole for bottom well width=6nm
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 48
The experimentally measured spectral response of the asymmetric DWELL
with increasing bottom well width shows a monotonic red shift [Ref: Quantum
dots-in-a-well infrared photodetectors, Sanjay Krishna]. This result can be intu-
itively expected because increasing Quantum Well width means decreasing the
connement for electron and hence the neighboring energy levels comes closer,
thus leading to smaller wavelength of photon absorption.
Figure 3.13: Eect of changing well width on DOS
A rigorous theoretical analysis of DOS for various Well width structure yields
the same expected result. From gure 3.13, we see, the width of Quantum
Well hardly has any prominent eect on the position of ground state energy,
though it signicantly changes other levels, causing a monotonic shift of DOS
towards lower energy level with gradually increasing Well width. An interesting
observation is that, the higher energy levels are getting closer to ground state at
a rate faster than lower energy level, thus reducing the relative distance between
any two energy levels we consider for transition. The corresponding energy levels
are shown in table 1 and their dierences are shown (in meV) in table 2.
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 49
Figure 3.14: Eect of changing well width on DOS
Energy in meV 3nm Well 4nm Well 6nm Well
E
0
-270.3 -270.3 -270.3
E
1
-23 -26 -28
E
2
79 74 64
Energy dierence in meV E
1
E
0
E
2
E
0
E
2
E
1
3 nm Well 247.3 349.3 102
4 nm Well 244.3 344.3 100
6 nm Well 242.3 334.3 92
It is a point of interest to see how dierent sites of the device contribute to
dierent energy states. Figure 3.15 shows the potential prole for the DWELL
model with 6 nm bottom half Quantum Well.
From gure 3.16, we see, at the ground level (E= -278meV), energy state
distribution takes a triangular form reveals the fact that at this bottom-most
energy level, quantum state is provided solely by the quantum dot.
Though at E = -62 meV, provision of energy state by the device is of minimal
value (of order 10
8
) and hence has no signicant number of state in practice,
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 50
Figure 3.15: Potential Prole
Figure 3.16: LDOS at E=-270.3208mev
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 51
we nevertheless add here a state distribution (Figure 3.16) to point out the
curious fact that as we are departing more and more from ground state energy,
energy state are getting spread wider and wider from their previous condense
location at quantum dot site. The previous triangle shape state distribution now
assumes an almost dumble like shape.
Figure 3.17: LDOS at E=-61.8225mev
From gure 3.18, we can see, at the rst continuum energy level (E= -15meV),
the density of state is concentrated at wetting layer of InAs
Figure 3.19) reveals that, at second continuum energy level (E= 77meV), the
energy states resides prominently at InGaAs Quantum Well sites.
CHAPTER 3. QDIPS DENSITY OF STATES MODELING 52
Figure 3.18: LDOS at E=-15mev
Figure 3.19: LDOS at E=77mev
Chapter 4
Dipole Moment and Absorption
Coecient
4.1 Transitional Dipole Moment
The Transition dipole moment or Transition moment, usually denoted
nm
for a
transition between an initial state, m, and a nal state, n, is the electric dipole
moment associated with the transition between the two states. In general the
transition dipole moment is a complex vector quantity that includes the phase
factors associated with the two states. An oscillating electric or magnetic mo-
ment can be induced in an atom or molecular entity by an electromagnetic wave.
Its interaction with the electromagnetic eld is resonant if the frequency of the
latter corresponds to the energy dierence between the initial and nal states
of a transition (E = h). The amplitude of this moment is referred to as the
transition moment. Its direction gives the polarization of the transition, which
determines how the system will interact with an electromagnetic wave of a given
polarization, while the square of the magnitude gives the strength of the inter-
action due to the distribution of charge within the system. The SI unit of the
transition dipole moment is the Coulomb-meter (Cm); a more conveniently sized
unit is the Debye (D).
The interaction energy, U, between a system of charged particles and an elec-
53
4. Dipole Moment and Absorption Coecient 54
tric eld, E, is given by:
U = E (4.1)
The dipole moment is dened for a collection of charges by
=
i
q
i
r
i
(4.2)
Where r
i
is the position vector of charged particle i. The expectation value
of the interaction energy is
U =
_
n
_
.
E
_
n
d (4.3)
If we assume that the magnitude of the electric eld is constant over the length
of the molecule (and that is nite only over the length of the molecule) we can
write
U = .E (4.4)
where
=
_
n
n
d (4.5)
i.e. the strength of interaction between a distribution of charges and an electric
eld depends on the dipole moment of the charge distribution.
In order to obtain the strength of interaction that causes a transition between
two states, the transition dipole moment is used rather than the dipole moment.
For a transition between and initial state,
i
, to a nal state
f
, the transition
dipole moment integral is.
fi
=
_
f
i
d (4.6)
the probability for a transition (as measured by the absorption coecient) is
proportional to
fi
fi
fi
may be positive, negative or imaginary. If
fi
then the interaction energy is
4. Dipole Moment and Absorption Coecient 55
zero and no transition occurs - the transition is said to be electric dipole forbidden.
Conversely, if
fi
islarge, thenthetransitionprobabilityandabsorptioncoefficientarelarge.
The intensity of the transition is therefore proportional to
k
j
d
2
The dipole moment operator for an electron in one dimension is
fi
= e
_
f
x
i
dx = e
f
|x|
i
(4.7)
Notes on transitional (dipole) moment: 1. The absorption probability for lin-
early polarized light is proportional to the cosine square of the angle between the
electric vector of the electromagnetic wave and ; light absorption will be maxi-
mized if they are parallel, and no absorption will occur if they are perpendicular.
2. It is frequently said that a transition is polarized along the direction of its
transition moment and this direction is called the polarization direction of the
transition. 3. In the case of a doubly degenerate nal state f, each of the two
components at the same energy has a transition moment and the two moments
dene a plane. The transition is then said to be polarized in that plane, which
also denes its polarizationdirection(s). This is typically the case for some of the
transitions in highly symmetrical molecules.
4.2 Optical Transition using Fermis Golden Rule
Let us consider
1
a semiconductor illuminated by light. The interaction between
photons and electrons in the semiconductor can be described by the Hamiltonian
[5]
H =
1
2m
0
(p eA)
2
+ V (r) (4.8)
Where m
0
is the free electron mass, e = e| for electrons, A is the vector
potential accounting for the presence of the electromagnetic eld, and V(r) is the
periodic crystal potential. The above Hamiltonian can be understood because
1
One of the authors, Mr. Tonmoy Kumar Bhowmick was inspired by his thesis supervisor
Dr. Farseem Mannan Mohammedy to take the course Semiconductor Device Theory from which
he learnt a great deal about semiconductor physics. He is grateful to his thesis supervisor for
his inpiration and encouragement
4. Dipole Moment and Absorption Coecient 56
taking the partial derivatives of H with respect to the momentum variable p and
the position vector r will lead to the classical equation of motion described by
the Lorentz force equation in the presence of an electric and a magnetic eld.
4.3 The Electron-Photon Interaction Hamilto-
nian
The Hamiltonian can be expanded into
H =
p
2
2m
0
+ V (r)
e
2m
0
(p A + A p) +
e
2
A
2
2m
0
= H
o
+ H
(4.9)
Where H
o
is the unperturbed Hamiltonian and H is considered as a pertur-
bation due to light:
H
0
=
p
2
2m
0
+ V (r) (4.10)
H
=
e
m
0
A p (4.11)
The Coulomb gauge
A = 0 (4.12)
Has been used such that p.A=A.p, noting that p = (/i) . The last term
e
2
A
2
2m
0
is much smaller than the terms linear in A, since absolute value of eA is
much smaller than absolute p for most practical optical eld intensities. This can
be checked using
p = k /a
0
(4.13)
Where a
0
5.5angstrom is the lattice constant. Assuming the vector potential
for the optical electric eld of the form
A = eA
0
cos(k
op
r t) (4.14)
4. Dipole Moment and Absorption Coecient 57
A = e
A
0
2
e
ik
op
r
e
it
+ e
A
0
2
e
ik
op
r
e
+it
(4.15)
Where k
op
is the wave vector, is the optical angular frequency and e is a
unit vector in the direction of the optical electric eld, we have
E(r, t) =
A
t
(4.16)
E(r, t) = eA
0
sin(k
op
r t) (4.17)
H(r, t) =
1
A (4.18)
H(r, t) =
1
k
op
eA
0
sin(k
op
r t) (4.19)
since the scalar potential vanishes for the optical eld. The interaction Hamil-
tonian H(r, t) can be written as
H
(r, t) =
e
m
0
A(r, t) p (4.20)
H
(r, t) = H
(r)e
it
+ H
+
(r)e
it
(4.21)
where H
(r, t) is
H
(r, t) =
eA
0
e
ik
op
r
2m
0
e p (4.22)
4.4 Transition Rate due to Electron-Photon In-
teraction
The transition rate for the absorption of a photon, assuming an electron is initially
at state Ea, is given by Fermis golden rule and has been derived in Section 3.7
4. Dipole Moment and Absorption Coecient 58
using the time-dependent perturbation theory:
W
abs
=
2
_
b
a
_
2
(E
b
E
a
) (4.23)
Where E
b
> E
a
has been assumed. Similarly the transition rate for the
emission of photon if an electron is initially at state E
b
is
W
abs
=
2
_
a
b
_
2
(E
a
E
b
+ ) (4.24)
The total upward transition rate per unit volume (s-1cm-3) in the crystal,
taking into account the probability that state a is occupied and state b is
empty, is
R
ab
=
2
V
k
a
k
b
2
ba
2
(E
a
E
b
)f
a
(1 f
b
) (4.25)
Where we sum over the initial and nal states and assume that the Fermi-
Dirac distribution
f
a
=
1
1 + e
(E
a
E
f
)/k
B
T
(4.26)
is the probability that the state a is occupied. A similar expression holds for
f
b
, where E
a
is replaced by E
b
, and (1 f
b
) is the probability that the state b
is empty. The prefactor 2 takes into account the sum over spins, and the matrix
element H
ba
is given by
H
ba
=
_
b
(r)
a
_
(4.27)
H
ba
=
_
b
(r)H
(r)
a
(r)d
3
r (4.28)
With |H
ba| = |H
k
a
k
b
2
ba
2
(E
a
E
b
+ )f
b
(1 f
a
) (4.29)
Noting the even property of the delta function, (x) = (x), the net upward
4. Dipole Moment and Absorption Coecient 59
transition rate per unit volume can be written as
R = R
ab
R
ba
(4.30)
R =
2
V
k
a
k
b
2
ba
2
(E
a
E
b
)(f
a
f
b
) (4.31)
4.5 Optical Absorption Coecient
The absorption coecient (1/cm) in the crystal is the fraction of photons ab-
sorbed per unit distance:
=
no. of photons absorbed per unit volume per second
no. of photons injected per unit area per second
(4.32)
The injected number of photons per unit area per second is the optical inten-
sity S (W/cm2) divided by the energy of a photon ; therefore,
() =
R
S/
(4.33)
() =
n
r
2
A
2
0
/2c
2
V
k
a
k
b
2
| e p
ba
|
2
(E
a
E
b
)(f
a
f
b
) (4.34)
Using the dipole (long wavelength) approximation that A(r) = Ae
ik.r
A
We nd that the matrix elements can be written in terms of the momentum
matrix element:
ba
= e b |r| a = er
ba
(4.35)
H
ba
= b |er E| a =
ba
E (4.36)
p = m
0
d
dt
r =
m
0
i
(rH
0
H
0
r) (4.37)
4. Dipole Moment and Absorption Coecient 60
H
ba
=
_
b
e
m
0
A(r) p
a
_
(4.38)
H
ba
=
e
i
A b |rH
0
H
0
r| a (4.39)
H
ba
=
e(E
a
E
b
)
i
A b |r| a (4.40)
H
ba
= iA b |er| a (4.41)
H
ba
=
ba
E (4.42)
The absorption coecient becomes
() =
n
r
c
0
2
V
k
a
k
b
| e
ba
|
2
(E
a
E
b
)(f
a
f
b
) (4.43)
Where E
b
E
a
= has been used, and E = +iA for the rst term in
A(r, t) with exp(it) dependence In terms of the dipole moment, we write the
absorption coecient as
() =
n
r
c
0
2
V
k
a
k
b
| e
ba
|
2
(E
a
E
b
)(f
a
f
b
) (4.44)
We can see that the factors containing A
2
o
are canceled since the linear optical
absorption coecient is independent of the optical intensity, which is proportional
to A
2
o
. When the scattering relaxiation is included, the delta function may be
replaced by a Lorentzian function with a linewidth .
(E
a
E
b
)
/2
(E
a
E
b
)
2
+ (/2)
2
(4.45)
Where a factor has been included such that the area under the function is
4. Dipole Moment and Absorption Coecient 61
properly normalized:
_
(E
a
E
b
)d() = 1 (4.46)
4.6 Wavefunction Calculation
To calculate the dipole moment in (4.2), the wavefunctions corresponding to the
initial and nal states of the transitions should be obtained. The wavefunctions
corresponding to the discrete energy states in the QDs are localized inside the
QDs. Therefore, assuming zero boundary conditions on a virtual cylinder sur-
rounding the QD, the bound wavefunctions and the corresponding discrete energy
eigenvalues can be calculated. In the continuum states above the QD barrier, the
assumption of zero boundary conditions is not suitable for obtaining the wave-
functions in this part of the DOS. Instead, the method of the Greens function is
used to calculate the wavefunctions in this part of the spectrum. A self-energy
term is added to the Hamiltonian at the cylinder surface to simulate the proper
boundary conditions representing the coupling with the surroundings outside the
cylinder. The bound state wavefunctions in the quantum dots are calculated by
solving the eigenvalue problem:
[EI H]
b
= 0 (4.47)
which gives all the eigenstates and the corresponding eigenvalues and HD is
the Hamiltonian matrix of the isolated cylinder cross section that contains the
Quantum Dot. This procedure is suitable to calculate the bound states since they
are localized inside the quantum dot and hence we can assume zero boundary
condition at the cylinder radius. To get the wave-function in the continuum part
of the spectrum, the Greens function method has been used to simulate the free
motion of electrons above the quantum dot and to add an additional self-energy
term at the cylindrical surface to simulate the open boundary condition. The
retarded Greens function gives the response at any point due to a unit excitation
at any other point. So the Hamiltonian of the isolated cylinder is modied by
adding a self-energy term to simulate the interaction with the surroundings in
4. Dipole Moment and Absorption Coecient 62
addition to the excitation term in the R.H.S. The wavefunctions [Ref: S. Datta,
Electronic Transport in Mesoscopic Systems (Cambridge University, Press, 1995)]
satisfy the equation
_
EI H
D
r
_
c
= S (4.48)
where S is the excitation term due to propagating waves incident from the
lead at the cylinders surface. The retarded Greens function of the system in the
cylindrical representation and in the matrix form is given by
_
EI H
D
r
_
G
D
= (x x
)(y y
) (4.49)
From the above equation, the wavefunction can be related to the Greens
function by the relation:
c
=
_
EI H
D
r
_
1
S = G
D
S (4.50)
If we write the self energy
r
=
m
f
m
|
m
(4.51)
then the excitation S is given by
S =
m
(f
n
f
n
) |
m
(4.52)
which gives the contribution from all propagating modes at same total energy
E. In normal incidence operation, the electric eld is polarized in the in-plane
directions, that is, x or y directions. Therefore for an arbitrary polarized electric
eld in the xy plane, the dipole moment will be given by
bc
= e
c
|r|
b
(4.53)
bc
= e
_
i + y
b
_
(4.54)
which indicates the selection rules for allowed optical transitions. The elec-
4. Dipole Moment and Absorption Coecient 63
trons can be photo-excited from the bound states of quantum dot to the contin-
uum states lying in quantum well or barrier.
4.7 Our Findings and Results
The reported heterostructure in [19], cited as sample B (AlGaAs/InGaAs/InAs),
when compared with sample A (AlGaAs/GaAs/InAlGaAs/InAs), has exhibited
a signicant improvement in photocurrent response for both S-plane (in plane)
and P plane (45 deg to Growth plane) polarized light, which, the authors like to
attribute to the improved Height to Base ratio (H/B) of Quantum Dot in Sample
B than in sample A. In a bid to nd out whether this improvement is predomi-
nantly caused by Height to Base ratio improvement or by some other reason (since
the two samples are of dierent composition), we have chosen a heterostructure
reported in [8]. In this specimen, The self-assembled growth of quantum dots in
the Stranski-Krastanov (SK) growth mode grows Quantum Dots having height
of 6.5nm and base of 11nm on 2.4 ML wetting layer. In theoretical simulation
of absorption coecient, we have changed the base dimension of quantum dot,
keeping its height xed at 6.5nm.
The experimentally measured polarized spectral response of the DDWELL
sample B with H/B=8nm/12nm=0.67, shows a blue shift of peak response over
DDWELL sample A with H/B= 5nm/17nm= 0.29 in case of both S and P po-
larized light. However, we believe, this shift is prominently caused by composi-
tional variation of the two samples, and therefore, other than wavelength shift,
are primarily interested in verifying the claim of magnitude of spectral response
improvement due to H to B ratio improvement in this paper. This result may
be intuitively expected because increasing Quantum Dot Base dimension means
decreasing the zero dimensional connement for electron and thus decreasing
atom-like behavior of QD structure.
A rigorous theoretical analysis of transitional dipole moment for standard
Quantum dot dimension cited in [8] with 6.5nm Height and 11nm Base yields
the gure 4.2. This gure shows two dominant peak and one pseudo-dominant
peak. However, not all the peak be necessarily existent in experimental result of
photocurrent response, since Fermis Golden Rule will eliminate the possibility
4. Dipole Moment and Absorption Coecient 64
of any transition from even to even or odd to odd ? function.
The calculated transitional dipole moment is later used to estimate absorption
coecient at 77K. We considered inhomogeneous broadening (Lorentzian broad-
ening) is dominant in absorption phenomena. The curve of absorption coecient
in Fig 4.2 exhibits not only a smoothened function but also the elimination of
the pseudo appearance of a peak in dipole moment curve.
Figure 4.1: Dipole moment versus wavelength for S polarized inci-
dent light in a standard dot-in-a-well infrared photodetector. QD has
Height/Base=6.5nm/11nm
A relative comparison among absorption coecient of QDIP for various Height
to Base ratio is graphically shown in Fig 4.3 and Fig ??. Fig 4.3, applicable
for S polarized light, indicates, peak of absorption coecient (and henceforth,
probably spectral response) is vividly highest among all three H/B value taken
into account. However, it does not experience a monotonic fall in maxima as H/B
value is declined.
Fig 4.4, applicable for P polarized incident light gives us a rather straight-
forward notion. It indicates a clear improvement in absorption coecient with
the rise of Height to Base ratio.
4. Dipole Moment and Absorption Coecient 65
Figure 4.2: Absorption coecient versus wavelength for S polarized in-
cident light in a standard dot-in-a-well infrared photodetector. QD has
Height/Base=6.5nm/11nm
Figure 4.3: Comparison of absorption coecient of dierent hypothetical QDIPs
having various Height to Base ratio. Incident Light is S polarized (in plane
incidence)
4. Dipole Moment and Absorption Coecient 66
However, in both case of polarization, the maxima of absorption coecient
invariably shifts towards red end with gradual rise in Height to Base ratio.
Table below indicates the relative magnitude of maxima of absorption co-
ecient, for S and P polarization. Notably, since the base diameter of QD is
always larger than the height [6], the polarization dependent infrared absorption
is anisotropic and s -polarized (in-plane) absorption is always less than the vertical
polarized absorption.
Figure 4.4: Comparison of absorption coecient of dierent hypothetical QDIPs
having various Height to Base ratio. Incident Light is P polarized (45 degree
incidence to growth plane)
4.8 Conclusion
Here, we have pointed out the interesting phenomenon of how the quantum dot
dimensions exert impact in determining DWELLs absorption co-ecient. This
analysis will eventually shed some light on the relative change of magnitude of
the photocurrent response in conjunction with change of QD dimensions. The
InAs/InGaAs/GaAs DWELL structure is taken and its quantum dot base is var-
4. Dipole Moment and Absorption Coecient 67
Figure 4.5: Relative change of absorption peak with dot dimension change. Ab-
sorption peak found for 7nm QD base is taken as unity, when incident light is S
polarized.
Figure 4.6: Relative change of absorption peak with dot dimension change. Ab-
sorption peak found for 7nm QD base is taken as unity, when incident light is P
polarized.
4. Dipole Moment and Absorption Coecient 68
ied while its height is kept xed at standard value (6.5nm). In each case, absorp-
tion co-ecient of the device is calculated resulting from both S and P polarized
incident light. The experimentally found fact of responsivity improvement with
quantum dot height to base ratio improvement also matches with our numerically
simulated result within a certain range of observation. The parabolic variation of
absorption maximum with change in QD height to base ratio for S polarized light
can likely be attributed to the race between two possible factors: QD volume and
QD dimension ratio. However, whether absorption maximum will change in a
similar parabolic way for P polarized light as well beyond our observed range or
will retain its linear shape remains a matter of further investigation.
Chapter 5
Conclusions
A summary of the main contributions of this research about DWELL modeling
and optimization is presented in this chapter. Future research recommendations
are provided for improvement and upgrading of the models developed achiev-
ing more accurate device characterization which would lead to improved device
design.
5.1 CONCLUSION
In our thesis, we have explored some interesting and important behavior of Quan-
tum dot-in-a-well detector by simulation and compared with the available exper-
imental data. DWELL detector has one wonderful feature. Its peak wavelength
response can be tuned in two ways: a) by changing the bias voltage, and b) by
changing the dimension of quantum well in the heterostructure. It implies that
a DWELL detector is tunable both before and after fabrication.
In the very rst phase, we have investigated how the change in thickness
of quantum well brings change in Density of Quantum State. In fact, we have
varied the dimension of bottom half well keeping the upper half well xed in
thickness so as to house the quantum dots promptly. A gradual increase of the
width of bottom half well results in a gradual reduction in distance between
any two neighboring energy levels. In particular, the position of ground energy
level remains unaected by the change of quantum well dimension, whereas the
higher energy levels shifts towards the ground level, and hence lessening the inter
69
CHAPTER 5. CONCLUSIONS 70
sub-band gap. This observation eventually substantiates the fact that, if we
increase the height of quantum well, we may expect to obtain a red shift of peak
wavelength response. However, this observation may experience an exception in
continual red shift of wavelength, since an availability of quantum state does not
necessarily prove the existence of an electron in that state. Occupation of state
by an electron is also dependent on Fermi function that we apparently have not
taken into account in this investigation.
In the second phase, we have pointed out the interesting phenomenon of how
the quantum dot dimensions exert impact in determining DWELLs absorption
co-ecient. This analysis will eventually shed some light on the relative change
of magnitude of the photocurrent response in conjunction with change of QD di-
mensions. The InAs/InGaAs/GaAs DWELL structure is taken and its quantum
dot base is varied while its height is kept xed at standard value (6.5nm). In
each case, absorption co-ecient of the device is calculated resulting from both
S and P polarized incident light. The experimentally found fact of responsivity
improvement with quantum dot height to base ratio improvement also matches
with our numerically simulated result within a certain range of observation. The
parabolic variation of absorption maximum with change in QD height to base
ratio for S polarized light can likely be attributed to the race between two possi-
ble factors: QD volume and QD dimension ratio. However, whether absorption
maximum will change in a similar parabolic way for P polarized light as well
beyond our observed range or will retain its linear shape remains a matter of
further investigation. Noteworthy, this work can have future extension to nd
out some more properties inherently related to DWELL detectors mechanism.
In example, the calculation of transmission co-ecient, together with previously
calculated absorption co-ecient will give us responsivity of the detector. Be-
sides, the dark current behavior with variation in temperature and voltage bias
will be a very crucial point to explore. In addition, we are planning to nd a
feasible way out to incorporate the eect of coupling between two neighboring
quantum dots in our mathematical model in a bid to bring the simulated result
closer to the experimentally observed behavior of a DWELL photo-detector.
CHAPTER 5. CONCLUSIONS 71
5.2 FUTURE WORK
The present research lays the foundation for developing quantum transport mod-
els based on NEGF to describe the main characteristics of DWELLs. The goal
of these models is to understand the physical operations of these nano-devices in
order to have an insight of how the detector design can be improved such that it
gives the optimized performance eciently fullling the required characteristics.
There is a trade o between the accuracy of the model and its complexity.
Therefore, for good results a compromise between the accuracy and complexity
is necessary. More research is required to improve and upgrade the models de-
veloped in the thesis to make them more accurate and to calculate more physical
quantities required in fully characterizing the DWELLs. Some of these recom-
mendations will be described in this section
5.2.1 INTERMIXING AND STRAIN EFFECT
The models developed in the thesis are based on a single band eective mass
approximation. The DWELLs modeled are unipolar devices and the values of
the conduction band osets and electron eective masses are used to solve the
dierential equation of the Greens function. As described in chapter 3, all points
in the lattice grid associated to the QD have single values for the potential energy
and electron eective mass and the same thing for the QW and barrier regions.
This means that no intermixing is assumed between the QD material and the
material corresponding to the surrounding barriers. For example, for InAs QDs
grown on GaAs, the model assumed no intermixing between In and Ga to get
a simplied potential energy and eective mass proles. The real situation is
that intermixing between dierent heterostructure materials does occur which
will modify the values of the band osets and eective masses . Also the strain
eld distribution in the QD region will aect the values of band osets and
eective masses such that they will have a specic distribution in the QD and
the surrounding regions. The numerical model developed in the thesis is general
and can accept any input potential energy and eective masses prole if known.
So nding out the exact compositions of the QD and the surrounding barrier in
CHAPTER 5. CONCLUSIONS 72
addition to the eect of the strain eld distribution on the values of the conduction
band osets and eective masses will allow the model to describe a more realistic
conguration of the QD region and hence giving better results.
5.2.2 NON-UNIFORMITY OF QDS SIZE, AND DIS-
TRIBUTION
One of the assumptions used in the model to simplify the calculations is that it
actually describes a uniform distribution of QDs in the layers of the detector in
addition to using the average values of height and diameter of QDs in the cal-
culations. This gives symmetry that allows, for example, calculating the portion
of the current that passes through one QD and then the total current is the QD
density times this portion of the current. In addition it simplies the calculations
of the DOS, energy eigenvalues and the corresponding wavefunctions. The model
used in the thesis considered the uncertainty in the shape and size of QDs by
using average values for the height and diameter of the QDs by adding a small
imaginary part to the electron energy in the equation. Adding this small imagi-
nary part causes broadening to the energy levels such that they have a Gaussian
shape instead of sharp delta functions obtained when all QDs have the same
shape and size. A better description of the random shape and size of QDs can be
done through a sensitivity analysis technique described in . This technique gives
an estimate about the rate of change of a specic physical quantity as a func-
tion of one of the dimensional parameters, for example. In this way there is no
need of multiple simulations to scan all possible values of heights and diameters
of QDs and instead, the average values can be used one time in the simulation
and then the sensitivity analysis will give the corresponding simulation values at
other values of QD sizes. So including this technique may save time of repeating
the simulations in case other QDs need to be tested.
CHAPTER 5. CONCLUSIONS 73
5.2.3 INTERACTION WITH THE LATTICE
The NEGF model gives a straightforward way of including interactions in the
system whether with the surrounding contacts or internally in the QDs and the
surrounding barriers. The types of interactions to be included and the degree
of approximations to be used determine both the accuracy and the complexity
of the calculations. The interactions with the lattice are modeled in its simplest
form. The self-energy due to lattice interactions is assumed to be constant inde-
pendent on energy and position and corresponding to localized elastic scattering
of a constant scattering rate. This simple form of self-energy does not require a
selfconsistent solutions between the Greens functions and the self-energy as in the
case of considering the self-consistent rst order Born approximation for model-
ing electron phonon interactions. A realistic description of the interaction with
the lattice requires including self-energies corresponding to interactions between
electrons and longitudinal and acoustic phonons, but a self-consistent solution
is needed in this case. Assuming constant and uniform scattering rates simpli-
es the calculations but it might not be valid for high applied biases and high
temperature. The electron-electron interaction is considered qualitatively in the
model through a self-consistent solution between the electron density and the
potential energy as shown in chapter 5. A better description of electron-electron
interaction can be done through the Hartree-Fock approximation to calculate the
self-energy due to the charges of the QD and surrounding QDs.
5.2.4 TRANSIENT RESPONSE
Furthermore, there is one important characteristic of DWELL that has not been
discussed in this work which might be useful for some applications. The transient
response of the DWELL should be modeled for uorescent applications where
the time decay of the received pulse is important. The models in the thesis
focus on DC applications such as thermal imaging and focal plane arrays where
the signal is in steady state. For time varying signal, the transient response
for the responsivity is needed. Some techniques to model the transient response
in QWIPs have been shown in using Monte Carlo methods. For an accurate
CHAPTER 5. CONCLUSIONS 74
description of the transient response in a quantum system like a DWELL, the
real time Greens functions can be used to calculate the responsivity as a function
of time . Further research is needed to solve the problem without adding a lot
of complexity due to the increase of the dimensions of the problem by adding
an extra time axis. One simple way would be to obtain a circuit model for the
DWELL in order to estimate a time constant RC for the detector. The time
constant could be obtained by calculating the resistance of the barrier material
between the QD layers and the equivalent capacitance between the contacts of
the detector taking into consideration the QD capacitance in the layers.
References
[1] P. Rotella G. von Winckel A. Stintz A. Amtout, S. Raghavan
and S. Krishna. Theoretical modeling and experimental characterization
of InAs/InGaAs quantum dots in a well detector,. J. App. Phys., 96, 3782
(5 pages),, 2004. 31, 44
[2] Paolo Luglib Christian Jirauschek Alpar Matyas, Tillmann Ku-
bisc. Comparison between semiclassical and full quantum transport analysis
of THz quantum cascade lasers. Physica E: Low-dimensional Systems and
Nanostructures, 2010. 26, 28
[3] S. Datta. Electronic transport in mesoscopic systems. Cambridge Univer-
sity Press, Cambridge, UK,, 1997. 13
[4] M. Paulsson F. Zahid and S. Datta. Advanced Semiconductors and
Organic Nano-Techniques, chapter Electrical Conduction through Molecules.
Academic press,Editor H. Markoc., 2002. 13, 17
[5] H. Haken. Light, Vol. 1, Waves , Photons, Atoms, p. 204. North-Holland,
Amsterdam,, 1986. 55
[6] K. Schmidt T. Ngo J. L. Feng A. Lorke J. P. Kotthaus
J. M. Garca, G. Medeiros-Ribeiro and P. M. Petroff. Appl. Phys.
Lett. 71. Appl. Phys. Lett. 71, 1997. 66
[7] Sanjay Krishna. Quantum dots-in-a-well infrared photodetectors. Infrared
Physics Technology. 31, 45
75
REFERENCES 76
[8] Sanjay Krishna. Quantum dots-in-a-well infrared photodetectors. J. Phys.
D: Appl. Phys. 38 2142-2150, 2005. 47, 63
[9] von Winckel G Stintz A Ariyawansa G Matsik S G Krishna S,
Raghavan S and PereraAGU. Appl. Phys. Lett. 83 2745. Appl. Phys.
Lett. 83 2745, 2003. 44
[10] Bowen. R.C. Jovanovic. D. Lake. R., Klimcck. G. J. Appl. Phys.
81. 7845, 1997. 26
[11] Fisdlchi. M.V. Lau S.E., KUIlla. A. J. Appl. Pllys. 95. 5545, 2004.
21, 46
[12] Keldysh. L.V. Sov. Phys. JETP 20. to t8, 1965. 12
[13] F. Zahid M. Paulsson and S. Datta. Nanoscience. Nanoscience,
engineering and technology handbook, chapter resistance of a molecule. 2002.
13
[14] D. A. Thompson MAK Naser, MJ Deen. Theoretical modeling of quan-
tum dot infrared photodetectors. ETD Collection for McMaster University,
2007. ix, 32, 33, 34
[15] Kubis. T. Lugli. P. Jirauschek. C. Matyas, A. Physica E. Addison-
Wesley, 2009. 21
[16] A. W. Ghosh P. S. Damle and S. Datta. Molecular nanoelectronics,
chapter Theory of nanoscale device modeling. Editor M. Reed., 2002. 12, 13,
17
[17] W. Potz. J. Appl. Phys. 86. 2458, 1989. 21
[18] Pereira M.F. Schmielau. T. Phys. Status Solidi. (b) 246. 329, 2009. 26
[19] Jiayi Shao. Enhanced normal incidence photocurrent in quantum dot in-
frared photodetectors. Journal of Vacuum Science Technology B: Microelec-
tronics and Nanometer Structures, 2011. 63
REFERENCES 77
[20] Venugopa. R. Slizhenko. A., Anantram. M.P .. Govindan. T.R ..
Biegc1. B. J. Appl. Phys. 91. 2343, 2002. 21
[21] Datta S. Lundstrom. M. Venugopal. R., Paulsson. Goosgucn. S.
J. Appl. Phys. 93. 5613, 2003. 21
[22] Jallho. A.P. Wacker. A. Phys. Rev. Lett. 80. 369, 1998. 21
[23] Frcnslcy. W.R. Rev. Mod. Phys. 62. 745, 1990. 21, 46
[24] Datta S. Yanik. A.A., Klimeck. G. Phys. Rev. B. 76. 0452 13, 2007.
21