You are on page 1of 52

Technische Universität München

Fakultät für Elektrotechnik und


Informationstechnik

Non-saturable Metasurfaces for Second Harmonic Generation

Master Thesis
Simon Stich
April 2022

Lehrstuhl für Halbleitertechnologie E26


Prof. Dr. Mikhail Belkin

Walter Schottky Institut


Zentrum für Nanotechnologie und Nanomaterialien
Technische Universität München
Contents

1 Introduction 1

2 Theory 3
2.1 Nonlinear Optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Quantum Mechanical Electric Susceptibility . . . . . . . . . . . . . . 3
2.2.1 Density Matrix Formalism . . . . . . . . . . . . . . . . . . . . 3
2.2.2 Linear Susceptibility . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.3 Second-Order Susceptibility . . . . . . . . . . . . . . . . . . . 12
2.3 Plasmonic Metasurfaces with Intersubband Transitions . . . . . . . . 15
2.3.1 MQW Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.2 Two-State SHG . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.3 Saturation Effects . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.4 Nonlinear Optical Response . . . . . . . . . . . . . . . . . . . 24
2.3.5 Finite Size Excitation . . . . . . . . . . . . . . . . . . . . . . 27

3 Methods 29
3.1 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.1 Optical Simulation . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.2 Thermal Simulation . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 SHG-Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Intersubband Absorption Measurement . . . . . . . . . . . . . . . . . 29

4 Metasurface Fabrication 33
4.1 Wafer Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.1.1 Molecular Beam Epitaxy . . . . . . . . . . . . . . . . . . . . . 33
4.1.2 Waferbonding . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1.3 Substrate Removal . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 Surface Structuring . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2.1 Hardmask Deposition . . . . . . . . . . . . . . . . . . . . . . 35
4.2.2 Electron-Beam Lithography . . . . . . . . . . . . . . . . . . . 35
4.2.3 Hardmask Development . . . . . . . . . . . . . . . . . . . . . 36
4.2.4 Cl-Ar Etching . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Page iv Contents

4.2.5 Top Gold Evaporation . . . . . . . . . . . . . . . . . . . . . . 37

5 Results 39
5.1 Intersubband Absorption Measurement . . . . . . . . . . . . . . . . . 39
5.2 T-Shaped Nanoresonators . . . . . . . . . . . . . . . . . . . . . . . . 39
5.3 V-Shaped Nanoresonators . . . . . . . . . . . . . . . . . . . . . . . . 45

6 Conclusions and Outlook 47


1 Introduction
2 Theory

2.1 Nonlinear Optics

Firstly, we want to focus on the underlying principles of nonlinear optics, which are
important for our metasurfaces.

In nonlinear optics the material polarization depends non-linearly on the applied


electric field:
P = ϵ0 (χ(1) E + χ(2) E2 + χ(3) E3 + ...). (2.1)

In this work we will focus on the nonlinear optical interaction called second-harmonic
generation (SHG). In order to understand this phenomenon, let us consider a laser
beam with an electric field according to:

e = Ee−iωt + E ∗ eiωt = Ee−iωt + c.c.


E(t) (2.2)

This beam hits a crystal which has a nonzero χ(2) element. According to (2.1), a
nonlinear polarization is created:

e (2) (t) = ϵ0 χ(2) EE ∗ + ϵ0 χ(2) (E 2 e−i2ωt + c.c.).


P (2.3)

2.2 Quantum Mechanical Electric Susceptibility

2.2.1 Density Matrix Formalism

∂ψs (r, t)
iℏ = Ĥψs (r, t), (2.4)
∂t
where Ĥ is the Hamiltonian of the system and can be written as

Ĥ = Ĥ0 + V̂ (t). (2.5)


Page 4 2 Theory

Ĥ0 is the Hamiltonian for a free atom and V̂ (t) represents the interaction energy.
The energy eigenstates of the free-atom Hamoltonian Ĥ0 form a complete set of
basis functions, such that the wavefunction of state s can be represented as a sum
of these:

ψs (r, t) = Cns (t)un (r). (2.6)
n

Here, un (r) are the energy eigenstates of the time-independent Schrödinger equation:

Ĥ0 un (r) = En un (r). (2.7)

The energy eigenstates are also assumed to be orthonormal, i.e.



u∗m (r)un (r)d3 r = δmn . (2.8)

Introducing eq. (2.6) into the time-dependent Schrödinger equation yields

∑ dC s (t) ∑
n
iℏ un (r) = Cns (t)Ĥun (r). (2.9)
n
dt n

We can simplify this expression by multiplying u∗m from the left. Through the
orthogonality condition from eq. (2.8) only the part of the sum where n = m
remains on the left side, such that we obtain:

d s ∑
iℏ Cm (t) = Hmn Cns (t). (2.10)
dt n

Hmn represents here the matrix elements of the Hamiltonian operator and is defined
as ∫
Hmn = u∗m (r)Ĥun (r)d3 r. (2.11)

The expectation value of any observable quantity A is obtained by



⟨A⟩ = ψs∗ Âψs d3 r, (2.12)

where  is an Hermitian operator associated to the quantity A. The Dirac notation


provides an alternative expression for this relation:

⟨A⟩ = ⟨ψs | Â |ψs ⟩ = ⟨s| Â |s⟩ . (2.13)


2.2 Quantum Mechanical Electric Susceptibility Page 5

|ψs ⟩ and |s⟩ denote state s. By introducing eq.(2.6) into eq. (2.12) we get

⟨A⟩ = s∗ s
Cm Cn Amn , (2.14)
mn

where Amn denotes the matrix element of the operator  and is defined as

Amn = u∗m Âun d3 r. (2.15)

The density matrix formalism can be employed to express a quantum mechanical


system in a statistical manner. This is especially useful, when we are uncertain of
the precise state of the system. We introduce p(s), which gives the probability that
the system is in state s. The elements of the density matrix elements are given by

s∗ s
ρnm = p(s)Cm Cn , (2.16)
s

which can alternatively be written as

∗C .
ρmn = Cm (2.17)
n

Here, the overbar denotes an average over all possible states of the system, i.e.
an ensemble average. The indices n and m range over all of the eigenstates of
the system. The diagonal elements of the density matrix, i.e. ρnn represent the
probability that the system is in the energy eigenstate n. The other elements ρnm
yield the coherence between the energy eigenstates n and m. ρnm will be nonzero if
the system is in a coherent superposition of energy eigenstate n and m.

We can now use this formalism to derive an expression for the expectation value of
any observable quantity, when the precise state of the system is unknown. We get
the expectation value for observable A by averaging over all possible states of the
system:
∑ ∑
⟨A⟩ = p(s) s∗ s
Cm Cn Amn . (2.18)
s nm

With eq. (2.16) we can rewrite this to



⟨A⟩ = ρnm Amn , (2.19)
mn

which can be simplified further by writing the sum over two indices as two individual
Page 6 2 Theory

sums as following:
∑ ∑∑ ∑
ρnm Amn = ρnm Amn = (ρ̂Â)nn . (2.20)
mn n m n

When we take a closer look at the expression ρnm Amn , it becomes clear, that the
sum over m must vanish:
∫ ∫
ρnm Amn = un ρ̂ um um Âun d r = u∗n ρ̂Âun d3 r = (ρ̂Â)nn .
∗ ∗ 3
(2.21)
| {z }
=1

Here, (ρ̂Â)nn corresponds to the n, n matrix element of the product of the two
operators ρ̂ and Â. By introducing the trace operation, which is defined as Tr(M̂ ) =

n M̂nn for any operator M̂ , we can express the expectation value of the observable
quantity A as:
⟨A⟩ = Tr(ρ̂Â). (2.22)

By finding an expression for the time evolution of the density matrix, we can now
determine the time evolution of the expectation value. In the next step, we want to
find out how the density matrix evolves in time. We start by take the time derivative
of the density matrix elements defined in eq. (2.16) and get:

∑ dp(s) ∑ ( s s∗
)
s∗ s s∗ dCn dCm s
ρ̇mn = Cm Cn + p(s) Cm + C . (2.23)
s
dt s
dt dt n

For now, we assume that p(s) is constant in time, therefore the first part of eq.
(2.23) vanishes. The second part of the equation can be rewritten by recalling
Schrödinger’s equation for the probability amplitudes from eq. (2.10). With this we
obtain:

s∗ dCns −i s∗ ∑
Cm = C Hnν Cνs (2.24a)
dt ℏ m ν
dC s∗ i ∑ i ∑

Cns m = Cns Hmν Cνs∗ = Cns Hνm Cνs∗ (2.24b)
dt ℏ ν
ℏ ν

Combining these two expressions with eq. (2.23) yields


∑ i ∑ s s∗
ρ̇mn = p(s) (Cn Cν Hνm − Cm
s∗ s
Cν Hnν ), (2.25)
s
ℏ ν

which can be simplified with the help of the definition for the density matrix elements
from eq. (2.16):
i∑
ρ̇mn = (ρnν Hνm − Hnν ρνm ). (2.26)
ℏ ν
2.2 Quantum Mechanical Electric Susceptibility Page 7

We can formally carry out the summation over ν and get:

i −i [ ]
ρ̇mn = (ρ̂Ĥ − Ĥ ρ̂)nm = Ĥ, ρ̂ . (2.27)
ℏ ℏ nm

Here, we introduced the commutator, which is defined for any two operators  and
B̂ as [Â, B̂] = ÂB̂ − B̂ Â. Eq. (2.27) can be further expanded with phenomenological
damping terms in order to account for certain interactions that cannot be included
into the Hamiltonian Ĥ, e.g. collisions between atoms. These interaction can cause
a change in the state of the system, which creates a nonzero dp(s)/dt. Since we
previously assumed a vanishing dp(s)/dt, we expand eq. (2.27) by another term
according to:
−i [ ]
ρ̇mn = Ĥ, ρ̂ − γnm (ρnm − ρ(eq)
nm ). (2.28)
ℏ nm
(eq)
The damping term on the right side causes ρnm to relax to its equilibrium value ρnm
with the relaxation rate γnm . Furthermore, γnm = γmn and we assume that

ρ(eq)
nm = 0 for n ̸= m. (2.29)

Unfortunately, eq. (2.28) cannot be solved exactly for physical systems of interest.
However, a perturbative approach can help us to find an approximate solution. This
requires a Hamiltonian, which can be written as a sum of the Hamiltonian for a free
atom Ĥ0 and the energy of the interaction between the atom and an external electric
field:
Ĥ = Ĥ0 + V̂ (t), (2.30)

where V̂ (t) is defined as:


e
V̂ (t) = −µ̂ · E(t). (2.31)

In addition, µ̂ = −er̂ represents the electric-dipole moment operator. In eq. (2.7) we


already defined, that the states n are eigenfunctions un of the unperturbed Hamil-
tonian Ĥ0 , therefore the matrix representation of Ĥ0 is diagonal:

H0,nm = En δnm . (2.32)

By introducing the expression for the Hamiltonian from eq. (2.30) into eq. (2.28)
the commutator splits into two separate parts and we obtain:

−i [ ] i[ ]
ρ̇mn = Ĥ0 , ρ̂ − V̂ , ρ̂ − γnm (ρnm − ρ(eq)
nm ) (2.33)
ℏ nm ℏ nm

With eq. (2.32) the first commutator for the unperturbed Hamiltonian Ĥ0 can be
Page 8 2 Theory

rewritten as:
[ ] ∑
Ĥ0 , ρ̂ = (Ĥ0 ρ̂ − ρ̂Ĥ0 )nm = (H0,nν ρνm − ρnν H0,νm )
nm
ν

= (En δnν ρνm − ρnν δνm Eν ) (2.34)
ν

= En ρnm − Em ρnm = (En − Em )ρnm .

Applying this expression to eq. (2.33) yields:

i[ ]
ρ̇mn = −ωnm ρnm − V̂ , ρ̂ − γnm (ρnm − ρ(eq)
nm ), (2.35)
ℏ nm

where we have defined the angular transition frequency ωnm = (En − Em )/ℏ for
convenience. Similar to eq. (2.34), we can expand the commutator of V̂ and ρ and
obtain:

i∑
ρ̇mn = −iωnm ρnm − (Vnν ρνm − ρnν Vνm ) − γnm (ρnm − ρ(eq)
nm ), (2.36)
ℏ ν

Finding an analytical solution to this equation is not possible for most physical
problems. For this reason we employ a perturbation expansion by replacing Vij with
λVij . λ is dimensionless parameter and ranges from zero to one. It indicates the
strength of the perturbation and is set to unity in order to represent the actual
physical situation. We desire the solution of eq. (2.36) in the form of a power series
in λ according to:
nm + λρnm + λ ρnm + · · · .
ρnm = ρ(0) (1) 2 (2)
(2.37)

Eq. (2.37) solves eq. (2.36) for any value of λ only if the coefficients of each power
of λ satisfy eq. (2.36) separately. We therefore obtain one expression for each power
of λ according to:

mn = −iωnm ρnm − γnm (ρnm − ρnm ),


ρ̇(0) (0) (0) (eq)
(2.38a)
i[ ]
ρ̇mn = −(iωnm + γnm )ρnm −
(1) (1)
V̂ , ρ̂(0)
, (2.38b)
ℏ nm
i [ ]
mn = −(iωnm + γnm )ρnm −
ρ̇(2) (2)
V̂ , ρ̂(1) . (2.38c)
ℏ nm

Here, we only present the first three powers of λ, because all higher powers follow
the same scheme as eq. (2.38b) and eq. (2.38c). The general form is given as:

i[ ]
mn = −(iωnm + γnm )ρnm −
ρ̇(N ) (N )
V̂ , ρ̂(N −1) for N ≥ 1. (2.39)
ℏ nm
2.2 Quantum Mechanical Electric Susceptibility Page 9

The steady-state solution of eq.(2.38a) is:

ρ(0) (eq)
nm = ρnm , (2.40)

(1)
which we can use to solve eq. (2.38b) by time integration. We rewrite ρnm as:

−(iωnm +γnm )t
ρ(1) (1)
nm (t) = Snm (t)e , (2.41)

and its time derivative using the product rule as:

−(iωnm +γnm )t (1) −(iωnm +γnm )t


nm = −(iωnm + γnm )Snm (t)e
ρ̇(1) (1)
+ Ṡnm e . (2.42)

Combining eq. (2.41) and eq. (2.42) with eq. (2.38b) yields the following expression:

i[ ]
(1)
Ṡnm =− V̂ , ρ̂(0) e(iωnm +γnm )t , (2.43)
ℏ nm

which we integrate over time and obtain:


∫ t
i[ ] ′
(1)
Snm = − V̂ (t′ ), ρ̂(0) e(iωnm +γnm )t dt′ (2.44)
−∞ ℏ nm

After substituting this back into eq. (2.41) we get the final expression for the first-
order correction:
∫ t
i[ ] ′
(1)
ρnm (t) = − V̂ (t′ ), ρ̂(0) e(iωnm +γnm )(t −t) dt′ . (2.45)
−∞ ℏ nm

In fact, all higher-order corrections can be calculated with a similar, more general
expression:
∫ t
i[ ] ′
ρ(N )
nm (t) = − V̂ (t′ ), ρ̂(N −1) e(iωnm +γnm )(t −t) dt′ for N ≥ 1. (2.46)
−∞ ℏ nm

2.2.2 Linear Susceptibility

We will now use this result, to derive the linear susceptibility of an atomic system.
As a first step, we rewrite eq. (2.45) in the following form:

−(iωnm +γnm )t
t
i[ ] ′
ρ(1)
nm (t) =e − V̂ (t′ ), ρ̂(0) e(iωnm +γnm )t dt′ . (2.47)
−∞ ℏ nm
Page 10 2 Theory

As a next step, we explicitly express the commutator from eq. (2.45):


[ ] ∑[ ]
V̂ (t), ρ̂(0) νm − ρnν V (t)νm
V (t)nν ρ(0) (0)
=
nm
ν
∑[ ] (2.48)
e
νm − ρnν µνm · E(t),
µnν ρ(0) (0)
=
ν

where we have used the definition for the interaction potential from eq. (2.31). Since
(0) (eq)
ρnm = ρnm (see eq. (2.40)) and therefore the condition from eq. (2.29) also applies
(0)
for ρnm , we can simplify even further by summing over ν and get:
[ ]
= −(ρ(0) e
V̂ (t), ρ̂(0) mm − ρnn )µnm · E(t).
(0)
(2.49)
nm

We substitute this expression into eq. (2.47) and obtain:


∫ t
i −(iωnm +γnm )t e ′ )e(iωnm +γnm )t′ dt′ .
ρ(1) mm − ρnn )µnm e
= (ρ(0) (0)
nm (t) E(t (2.50)
ℏ −∞

e as a sum of its individual frequency


We can express the external electric field E
components according to:

e =
E(t) E(ωp )e−iωt , (2.51)
p

which in combination with eq. (2.50) yields:

∑ ∫ t
i ′
ρ(1)
nm (t) = (ρ(0) − ρ (0)
)µ · E(ωp ) × e−(iωnm +γnm )t e[i(ωnm −ωp )+γnm ]t dt′ .
ℏ mm nn nm
−∞
p | {z }
e−iωp t
= i(ω
nm −ωp )+γnm

(2.52)
The right part of this expression can be evaluated explicitly and is given by the
(1)
underbraces below. ρnm is therefore given as:

1 (0) ∑ µ · E(ωp )e−iωp t


ρ(1)
nm = (ρmm − ρ(0) ) nm
. (2.53)
ℏ nn
p
(ωnm − ωp ) − iγ nm

In eq. (2.12) we have seen, that the expectation value of any observable can be
written as the trace of the density operator and the operator corresponding to that
particular observable. The expectation value for the induced dipole moment is
therefore defined as:

⟨e
µ(t)⟩ = Tr(ρ̂(1) µ̂) = ρ(1)
nm µmn . (2.54)
nm
2.2 Quantum Mechanical Electric Susceptibility Page 11

Note that we have omitted the bar over the quantities for the sake of simplicity.
Substituting the result from eq. (2.53) into eq. (2.54) yields:

∑1 ∑µ · E(ωp )]e−iωp t
mn [µnm
⟨e
µ(t)⟩ = mm − ρnn )
(ρ(0) (0)
. (2.55)
nm
ℏ p
(ωnm − ωp ) − iγnm

Similar to the electric field, we can write ⟨e


µ(t)⟩ as a sum of its frequency components
as:

⟨e
µ(t)⟩ = ⟨µ(ωp )⟩ e−iωp t . (2.56)
p

The linear susceptibility tensor χ(1) (ω) is defined by the equation:

P(ωp ) = N ⟨µ(ωp )⟩ = ϵ0 χ(1) (ωp )E(ωp ), (2.57)

where N is the atomic number density. When we compare this expression to eq.
(2.56) and finally to eq. (2.55) it becomes clear that the linear susceptibility is
defined as:
N ∑ (0) µmn µnm
χ(1) (ωp ) = (ρmm − ρ(0)
nn ) (2.58)
ϵ0 ℏ nm (ωnm − ωp ) − iγnm

We can also find an expression for the Cartesian components of the polarization
P(ωp ) with:
∑ (1)
Pi (ωp ) = N ⟨µi (ωp )⟩ = ϵ0 χij (ωp )Ej (ωp ), (2.59)
j

(1)
where the elements of the linear susceptibility tensor χij are:

(1) N ∑ (0) µimn µjnm


χij (ωp ) = (ρmm − ρ(0) ) . (2.60)
ϵ0 ℏ nm nn
(ωnm − ωp ) − iγnm

We can now separate the sum into two individual parts such that:

(1) N ∑ (0) µimn µjnm N ∑ (0) µimn µjnm


χij (ωp ) = ρmm − ρnn . (2.61)
ϵ0 ℏ nm (ωnm − ωp ) − iγnm ϵ0 ℏ nm (ωnm − ωp ) − iγnm

By interchanging the indices n and m in the second sum, we can recombine the two
parts into:
[ ]
(1) N ∑ (0) µimn µjnm µinm µjmn
χij (ωp ) = ρ + , (2.62)
ϵ0 ℏ nm mm (ωnm − ωp ) − iγnm (ωnm + ωp ) − iγnm

where we have also made use of the fact that ωmn = −ωnm and γmn = γnm . We
usually assume that all electrons reside in the ground state of the system, which is
denoted as state g in this case. As a consequence follows for the diagonal entries of
Page 12 2 Theory

the density matrix:

ρ(0)
gg = 1, ρ(0)
mm = 0 for m ̸= g. (2.63)

2.2.3 Second-Order Susceptibility

In order to calculate the second-order susceptibility, we proceed in a similar manner


as for the linear susceptibility. First, we use the expression for the second-order
correction to ρ̂ from eq. (2.46) (i.e. N = 2) and rewrite it to:

−(iωnm +γnm )t
t
i[ ] ′
ρ(2)
nm (t) =e − V̂ (t′ ), ρ̂(1) e(iωnm +γnm )t dt′ . (2.64)
−∞ ℏ nm

The commutator can once again be expressed in an explicit form (compare to eq.
(2.48)) as: [ ] ∑[ ]
=− e
V̂ (t), ρ̂(1) νm − ρnν µνm · E(t).
µnν ρ(1) (1)
(2.65)
nm
ν

Now that we obtained an expression for the first-order correction to ρ̂ in eq. (2.53)
(1) (1)
ρνm and ρnν obey the form:

1 (0) ∑ µνm · E(ωp )


ρ(1)
νm = (ρmm − ρ(0)
νν ) e−iωp t , (2.66a)
ℏ p
(ωνm − ωp ) − iγνm
1 (0) ∑ µnν · E(ωp )
ρ(1)
nν = (ρνν − ρ(0) ) e−iωp t . (2.66b)
ℏ nn
p
(ω nν − ω p ) − iγnν

e is eq. (2.66) into its frequency components:


We also decompose the electric field E(t)

e =
E(t) E(ωq )e−iωq t . (2.67)
q

By substituting the two eqs. (2.66) and eq. (2.67) into the the commutator expres-
sion from eq. (2.65) we obtain:
[ ] 1 ∑ (0)
V̂ (t), ρ̂ (1)
=− (ρmm − ρ(0)
νν )
nm ℏ ν
∑ [µ · E(ωq )][µ · E(ωp )] −i(ωp +ωq )t
× nν νm
e
pq
(ωνm − ωp ) − iγνm
(2.68)
1 ∑ (0)
+ (ρνν − ρ(0)
nn )
ℏ ν
∑ [µ · E(ωq )][µ · E(ωp )]
× nν νm
e−i(ωp +ωq )t .
pq
(ωnν − ωp ) − iγnν
2.2 Quantum Mechanical Electric Susceptibility Page 13

We combine this result with eq. (2.64) and perform the time integration. This
yields:
∑∑
ρ(2)
nm (t) = e−i(ωp +ωq )t
ν pq
{
(0) (0)
ρmm − ρνν [µnν · E(ωq )][µνm · E(ωp )]
×
ℏ2 [(ωnm − ωp − ωq ) − iγnm ][(ωνm − ωp ) − iγνm ]
} (2.69)
(0) (0)
ρνν − ρnn [µnν · E(ωq )][µνm · E(ωp )]

ℏ2 [(ωnm − ωp − ωq ) − iγnm ][(ωnν − ωp ) − iγnν ]
∑∑
≡ Knmν e−i(ωp +ωq )t .
ν pq

Here, we relabeled the expression in curly braces as Knmν for convenience. Once
again, we are interested in finding the expectation value for the atomic dipole mo-
ment, which is defined according to eq. (2.19) as:

⟨e
µ(t)⟩ = ρnm µmn . (2.70)
nm

The complex amplitudes ⟨µ(ωr )⟩ of the individual frequency components of ⟨e


µ⟩ are
defined by the following:

⟨e
µ(t)⟩ = ⟨µ(ωr )⟩ e−iωr t . (2.71)
r

The second-order correction of ρ̂ from eq. (2.69) contains frequency components at


ωp + ωq , thus the complex amplitude of this particular frequency of ⟨e
µ⟩ is given as:
∑∑
⟨µ(ωp + ωq )⟩ = Knmν µmn , (2.72)
nmν (pq)

from which directly follows the nonlinear polarization at this frequency:


∑∑
P(2) (ωp + ωq ) = N Knmν µmn . (2.73)
nmν (pq)

(2)
The nonlinear susceptibility tensor χijk is related to the Cartesian components of
the nonlinear polarization in the following way:

(2)
∑∑ (2)
Pi (ωp + ωq ) = ϵo χijk (ωp + ωq , ωq , ωp )Ej (ωq )Ek (ωp ). (2.74)
jk (pq)
Page 14 2 Theory

When we compare eq. (2.69), eq. (2.73) and eq. (2.74) it becomes clear that the
nonlinear susceptibility tensor is:

(2)[tent] N
χijk (ωp + ωq , ωq , ωp ) =
ϵ0 ℏ2
∑{ µimn µjnν µkνm
× (ρ(0) − ρ (0)
) (2.75)
mnν
mm νν
[(ωnm − ωp − ωq ) − iγnm ][(ωνm − ωp ) − iγνm ]
}
µimn µjνm µknν
− (ρνν − ρnn )
(0) (0)
.
[(ωnm − ωp − ωq ) − iγnm ][(ωnν − ωp ) − iγnν ]

We marked this result with a superscript, indicating that this is a tentative expres-
sion for the nonlinear susceptibility tensor. One fundamental property of the nonlin-
ear susceptibility is the intrinsic permutation symmetry. This requirement is neces-
sary for a physically meaningful nonlinear susceptibility. From eq. (2.74) we know
(2) (2)
that χijk (ωp + ωq , ωq , ωp )Ej (ωq )Ek (ωp ) is one contribution to the polarization Pi .
The variables j,k,p and q however, only serve as dummy variables. By interchanging
(2)
j with k and p with q we get χikj (ωp + ωq , ωp , ωq )Ek (ωp )Ej (ωq ), which should return
(2)
the same polarization Pi . The change in variables merely flipped the order of the
product of the two electric fields at the end of the expression, which has no effect on
(2) (2)
the result. For this reason we require that χijk (ωp +ωq , ωq , ωp ) = χikj (ωp +ωq , ωp , ωq ).
The current expression for the nonlinear susceptibility in eq. (2.75) does not possess
the required intrinsic permutation symmetry. We therefore expand the result of eq.
(2.75) by adding terms with interchanged ωp and ωq as well as interchanged j and
k.

(2) N
χijk (ωp + ωq , ωq , ωp ) =
2ϵ0 ℏ2
∑ { [
µimn µjnν µkνm
× (ρmm − ρνν )
(0) (0)

mnν
[(ωnm − ωp − ωq ) − iγnm ][(ωνm − ωp ) − iγνm ]
]
µimn µknν µjνm
+ (2.76)
[(ωnm − ωp − ωq ) − iγnm ][(ωνm − ωq ) − iγνm ]
[
µimn µjνm µknν
− (ρνν − ρnn )
(0) (0)
[(ωnm − ωp − ωq ) − iγnm ][(ωnν − ωp ) − iγnν ]
]}
µimn µkνm µjnν
+ .
[(ωnm − ωp − ωq ) − iγnm ][(ωnν − ωq ) − iγnν ]

We can rewrite this expression by changing the dummy indices ν, n and m in the
last two terms to m, ν and n, respectively, and thus the population differences of the
two terms are identical and can be factored out. The resulting general expression
2.3 Plasmonic Metasurfaces with Intersubband Transitions Page 15

for the second-order nonlinear susceptibility thereby becomes:

(2) N ∑ (0)
χijk (ωp + ωq , ωq , ωp ) = (ρ − ρ(0)
νν )
2ϵ0 ℏ2 mnν mm
{
µimn µjnν µkνm
×
[(ωnm − ωp − ωq ) − iγnm ][(ωνm − ωp ) − iγνm ]
µimn µknν µjνm
+ (2.77)
[(ωnm − ωp − ωq ) − iγnm ][(ωνm − ωq ) − iγνm ]
µinν µjmn µkνm

[(ωνn − ωp − ωq ) − iγνn ][(ωνm − ωp ) − iγνm ]
}
µinν µkmn µjνm
− .
[(ωνn − ωp − ωq ) − iγνn ][(ωνm − ωq ) − iγνm ]

In this thesis we focus on the SHG process, i.e. the pumps ωp and ωq are the same.
Furthermore, the quantum engineered nonlinear susceptibility only affects electric
fields along the MQW’s growth direction (in our case the z-direction). For this
(2)
reason, the second-order nonlinear susceptibility of interest becomes χzzz (2ω, ω, ω)
and is given by:

N ∑ (0)
χ(2)
zzz (2ω, ω, ω) = (ρmm − ρ(0)
νν )
ϵ0 ℏ mnν
2
{
µzmn µznν µzνm
× (2.78)
[(ωnm − 2ω) − iγnm ][(ωνm − ω) − iγνm ]
}
µznν µzmn µzνm
− .
[(ωνn − 2ω) − iγνn ][(ωνm − ω) − iγνm ]

In the following we want to calculate the second-order susceptibility for some exem-
plary quantum well configuration.

2.3 Plasmonic Metasurfaces with Intersubband Transitions

In general, Metamaterials present materials with engineered properties that are not
found in nature. Here, we want to focus on optical metamaterials, more precisely
metasurfaces, which reflect and thereby change the properties of the incoming light,
e.g. its polarization or frequency.

For our work, frequency doubling or second harmonic generation is the nonlinear
optical effect of interest. In our metasurface design we incorporate the extremely
large nonlinear optical responses from n-doped multi-quantum-well semiconductor
heterostructures together with the strong light-enhancement of plasmonic metasur-
Page 16 2 Theory

faces.

Typically, plasmonic metasurfaces consist of some type of nonlinear material (di-


electric) covered by metallic nanoresonators. The idea behind this approach is to
drastically enhance the light-matter interaction by creating strong electric fields (sur-
face plasmons) at the metal-dielectric boundary. Optically thin materials cannot
generate large nonlinear signals when compared to bulk materials. They therefore
require much stronger fields in order to produce similar nonlinear responses. These
strong fields can be achieved with metallic nanoresonators, as they enable strong
local field enhancement at the metal interface. Also, the generated fields follow the
nanoresonators profile allowing for precise control over polarization and resonances
at specific frequencies [Metasurfaces_KRASNOK].

In order to further enhance the nonlinear response, we employ n-doped multi-


quantum-well (MQW) semiconductor heterostructures as a dielectric material. They
present a class of artificial semiconductor materials and allow us to engineer non-
linear susceptibilities χ(2) and also χ(3) , which outperform the ones found in con-
ventional materials by many orders of magnitude [Metasurfaces_KRASNOK].
MQWs are essentially stacked layers of different III/V semiconductor materials that
form a specific potential landscape along the growth direction for the charge carri-
ers (electrons in our case) inside the material. Due to the varying bandgaps of the
individual layers, we create quantum wells which quantize the electrons motion in
the growth direction. By carefully choosing the widths of the individual quantum
wells and their doping we can achieve precise control over the materials frequency
response and more importantly can create extremely large nonlinear responses. Fig-
ure 2.1 shows an exemplary metasurface design for SHG, with so called Γ-shaped
gold nanoresonators patterned on top MQWs.

2.3.1 MQW Design

To this date, the extreme nonlinearities inside the MQWs were achieved in so-called
fully-resonant or semi-resonant configurations. In both cases the quantum system
possesses three states and the transition energy from the third state to the ground
state corresponds to the emitted SH energy, i.e. 2ℏω. In order to get appreciable
interaction with light at ℏω the 1-2-transition exactly corresponds to that required
energy in a fully-resonant configuration. Theoretically, this maximizes the obtained
(2)
intrinsic nonlinearity χzzz , however also results in a poor saturation performance
at elevated input light intensities. The strong interaction of the 1-2-transition with
the impinging light causes the excitation of charge carriers from the ground state
2.3 Plasmonic Metasurfaces with Intersubband Transitions Page 17

Figure 2.1: Plasmonic metasurface design for SH generation with gold nanores-
onators on top of MQW [nature_plasm].

|3⟩

ℏω |2⟩SR 2ℏω
|2⟩FR
FR
E12 = ℏω SR
E12 E13 = 2ℏω

|1⟩

Figure 2.2: Quantum states for the fully-resonant (FR) and semi-resonant (SR) con-
figuration.

to state 1 which thereby cannot contribute to the SHG process anymore. In semi-
resonant configurations, slight detuning of the 1-2-transition from the pump energy
can remedy this by decreasing the amount of excited electrons in state 1. On the
other hand, the overall second-order nonlinearity is reduced when compared to the
fully-resonant configuration.

Having the 1-2-transition close to the pump energy creates another disadvantage:
The intersubband transition generates considerable optical losses that lead to a
decreased field enhancement, which in turn heavily influences the overall antenna
performance. The imaginary part of the complex refractive index which arises from
the intersubband transitions can be calculated with the help of eq. (2.62) when
the quantum states of the system are known. In fig. 2.3 we have plotted the real
(nz ) and imaginary (kz ) part of the complex refractive index for an exemplary fully-
resonant SHG structure. The optical losses, indicated by kz , show two distinct peaks
at 124 meV and 248 meV, which correspond to the transition energies E12 and E13 ,
Page 18 2 Theory

Wavelength λ (µm)
10 9 8 7 6 5 4
0.7 4
nz

248 meV
0.6 kz 3.8

0.5 3.6

0.4 3.4

nz
kz

3.2
0.3
3
124 meV

0.2
2.8
0.1
2.6
0
100 150 200 250 300 350 400
Photon Energy ℏω (meV)
.
Figure 2.3: Real and imaginary parts of the complex refractive index in the growth
direction z of a MQW stack in a fully-resonant configuration. The struc-
ture is optimized for pump energy of 124 meV (≈ 10 µm)

respectively. Optical losses at the pump frequency lead to decreased field intensities
inside the MQW volume and thereby impair the SHG process. This becomes clear
when we take a look at the expression for the overlap integral, which enables the
enhancement of the intrinsic nonlinearity of the MQW material. For our application
the overlap integral can be written as:

∫ ( ω
)2 2ω
overlap 1 Ez[x] Ez[y]
I[yxx] = ω 2ω
dV. (2.79)
V VUC Einc[x] Einc[y]

The two fractions in the integral correspond to the field-enhancement inside the
MQW volume for x-polarized light at ω and y-polarized light at 2ω. Since the
enhancement factor of the pump contributes quadratically to the overlap integral
expression, low losses at the pump frequency are of utmost importance in order to
maximize the metasurface performance. Again, slight detuning of the 1-2-transition
from the pump energy can reduce the optical losses and thereby help to increase the
overlap integral.
2.3 Plasmonic Metasurfaces with Intersubband Transitions Page 19

Wavelength λ (µm)
109 8 7 6 5 4
0.3 3.5
Ec nz
800 kz
State 1
3.4
State 2
Energy (meV)

600 State0.2
3
E23 = 50 meV 3.3
z23 = 2.21 nm

nz
kz
400
E12 = 260 meV z13 = 0.8 nm 3.2

130 meV

260 meV
z12 = 1.04 nm 0.1
200 3.1

0 0 3
2 4 6 8 10 12100 150 200 250 300 350 400
Position z (nm) Photon Energy ℏω (meV)

(a) Proposed quantum well configu- (b) Real and imaginary part of the
ration for a two-state system. complex refractive index for the
two-state SHG structure.

2.3.2 Two-State SHG

We will now introduce a novel MQW design that enables us to alleviate poor sat-
uration performance at high input intensities and also possesses almost no losses
at the pump energy. This potentially allows us to generate extremely high field
intensities inside the MQW, beneficially affecting the overlap integral. The MQW
configurations introduced so far employ a three-level system in order to achieve huge
values for the nonlinear susceptibilities as described in the previous section. Our
approach was to completely remove the second state and have two upper states that
are separated only by tens of meV. Due to the small energy difference of state 2 and
3, we can consider them as one artificial state. Our proposed quantum well design
is shown in fig. 2.4a.

Eliminating the 1-2-transition at the pump energy bears several important advan-
tages: Optical losses are significantly reduced, because this design simply lacks an
intersubband transition for the input light to strongly interact with. Figure 2.4b
shows the real and imaginary part of an exemplary quantum well design for a two-
state SHG system optimized for a pump energy of approximately 130 meV. The
optical losses, i.e. the imaginary part kz , peak at 260 meV, corresponding to the
1-2-transition energy E12 depicted in fig. 2.4a. We also note that kz is very small
at the pump energy of 130 meV.
Page 20 2 Theory

Another advantage is that the pump light cannot to populate the second state since
its energy is well below the 1-2-transition. It is therefore possible to obtain extremely
large saturation intensities, i.e. saturation effects do not appear even at high input
intensities. If we assume even population for the upper two states the saturation
factor S SHG can be written as:

1 4IS12 + Izω
S SHG (Izω ) = , (2.80)
4 IS12 + Izω

where IS12 denotes the saturation intensity for the 1-2-transition. For extremely
large Izω , this expression approaches 0.25. At these high intensities our assumption
about the populations of the upper two states does not hold anymore and eq. (2.80)
becomes inaccurate. However, we assume that we stay in the accurate regime for
the intensities considered for our applications. The saturation intensity for the 1-2-
transition IS12 can be calculated by:

IS12 = (2.81)

In order to get proper thermal stability the Fermi level must always remain in
appropriate distance below the first excited state. This ensures that most of the
electrons reside in the ground state at operational temperature and thereby do not
cause any kind of saturation. The Fermi level is predominantly determined by the
density of electrons and follows in a two-dimensional system (i.e. quantum wells)
the following relationship:
πℏ2
EF = n2D . (2.82)
m
Here, n2D denotes the sheet density of electrons and scales linearly with the doping.
This limits the maximum doping before upper states are being populated at thermal
equilibrium. On the other hand, the nonlinear susceptibility is directly proportional
to the average bulk electron density and thus benefits from high doping. Our goal
therefore is to introduce as much doping as possible while still keeping the Fermi
level sufficiently far below the first excited state. Naturally, this also implies that a
more elevated second state allows for larger doping. Since the two-state SHG has the
second state at 2ℏω instead of ℏω as the fully-resonant design, we are therefore able
to introduce approximately twice as much doping and thus get a twofold increase in
(2)
the intrinsic second-order nonlinear susceptibility χzzz .

In the following we want to explore the contributions of the individual states to


the overall second-order nonlinearity. The expression to evaluate the second-order
2.3 Plasmonic Metasurfaces with Intersubband Transitions Page 21

nonlinearity is:

N ∑ (0)
χ(2)
zzz (2ω, ω, ω) = (ρ − ρ(0)
νν )
ϵ0 ℏ2 mnν mm
{
µzmν µznν µzmn
× (2.83)
[(ωmn − 2ω) − iγmn ][(ωmν − ω) − iγmν ]
}
µzmν µznν µzmn
− .
[(ωnν − 2ω) − iγnν ][(ωmν − ω) − iγmν ]

For the three state system for our application generally holds m, n, ν ∈ {1, 2, 3}.
We calculate the second-order nonlinear susceptibility under the assumption that
(0)
the ground state, i.e. state 1, is the only populated state. It follows that ρii = 0
for i ̸= 1 and therefore specific parts of the sum do not contribute to the overall
second-order susceptibility. In fig. 2.5 we plotted all triplets (m, n, ν) with a non-
zero contribution. The resonances at the individual transition energies are clearly
visible and it becomes apparent that the (1, 1, 2) triplet constitutes the main part
(2)
of the overall χzzz at the designated pump energy of 130 meV. For this combination
of m, n and ν, the inner part of the sum in eq. (2.83) becomes:
( )
(0) (µz12 )2 µz11 (µz12 )2 µz11
ρ11 − ,
[−2ω − iγ11 ][(ω12 − ω) − iγ12 ] [(ω12 − 2ω) − iγ12 ][(ω12 − ω) − iγ12 ]
(2.84)
where we get resonances at 0, E12 and most importantly at E12 /2, i.e. the pump
energy. Triplets with other combinations of m, n and ν only contribute little at the
energies of interest.

Since the quantization only affects the electrons along the growth direction and
their motion remains unconstrained in the plane of the layers, these strong nonlin-
earities can only be excited with electric fields perpendicular to the layers. This
is particularly hard to achieve because light impinging the material with normal
incidence does not possess electric fields along this direction and direct excitation
by light orthogonal to the layers is virtually impossible. The nanoresonators on top
of the MQW help to convert the orthogonally incident light into z-polarized electric
field inside the MQW volume. The generated electric fields along the z-direction
for different polarizations of incident light in case of the previously introduced Γ-
shaped nanoresonators are shown in fig. 2.6. Here, the units are given in terms of
enhancement, i.e. the ratio of the electric field strength in the z-direction and of
the incident radiation. The incident electric field is derived from the time averaged
Poynting vector, which also denotes the input light intensity and is given by:

1 2
Iinc = Sinc = ϵo c0 Einc . (2.85)
2
Page 22 2 Theory

·105
112
4 113
122
123
2 132
133
231
sum
0

−2

−4

0 50 100 150 200 250 300 350 400


Photon Energy ℏω (meV)

Figure 2.5: Imaginary part of the individual parts of the sum contributing to the
(2)
second-order nonlinear susceptibility χzzz .

This can be rearranged for the electric field to:



Einc = 2Z0 Iinc , (2.86)

where Z0 refers to the free space impedance of approximately 377 Ω. We can observe
greatly enhanced electric fields close to the metal-MQW-interface at the far ends of
the crossed bars. These strong fields inside of the MQWs together with the giant
nonlinear response of the materials promote the generation of high levels of nonlinear
signals.

Even higher field enhancements were achieved in ref. [ADOM] by removing the
MQW material, which is not covered by the resonator. This essentially creates a
metal-semiconductor nanocavity and due to the permittivity contrast between the
MQWs and the air gaps concentrates the electric field better in the MQW volume.
These highly concentrated electric fields in combination with a renewed quantum
well system lead to record high nonlinear signals. Also, a new antenna shape, which
is shown in fig. 2.7a, was developed to independently control the fundamental
and second-harmonic resonances. The long arm of the T-shape along the x-axis
creates a resonance at the pump wavelength and the resonance frequency can be
controlled with its length. The short arm along the y-axis generates a resonance at
2.3 Plasmonic Metasurfaces with Intersubband Transitions Page 23

Figure 2.6: Field enhancement for different polarization of incident radiation due
to surface plasmons inside the MQW layers. The electric fields in the
z-direction are plotted in terms of their enhancement factor i.e. Ez /Einc .

(a) Schematic of the T-shaped metal- (b) Field enhancement for the pump-
semiconductor nanoresonator for and second-harmonic frequency.
SHG.

Figure 2.7: T-shaped antennas used to achieve record-high SHG signal [ADOM].

the second harmonic frequency accordingly. Unlike the Γ-shaped resonators, which
allowed incident electric fields both along the x- and y-axis, this geometry requires
the impinging electric field to be polarized along the x-axis. In this case, the MQW
structure and the antennas were optimized for a pump wavelength of λ = 10 µm
(SHG wavelength λSHG = 5 µm accordingly). Fig. 2.7b depicts the distribution of
the electric field enhancement in the MQW volume for the fundamental frequency
(top) and the second-harmonic frequency (bottom), which is significantly larger than
for the Γ-shaped antennas in fig. 2.6.

2.3.3 Saturation Effects

The nonlinear susceptibility tensors for the MQWs stem from the transitions between
(2)
the different energy levels. For the calculation of χzzz we usually assume that all
electrons reside in the ground state. However, since the incident light excites the
electrons, some of them are being promoted into upper states as well. The amount
Page 24 2 Theory

of electrons leaving the ground state is proportional to the light intensity along the
z-direction at the fundamental frequency ω inside the MQW volume Izω1 . The large
field enhancement factors from the antennas cause huge field intensities inside the
MQWs. Note that the electric field is related to the intensity as:

Izω1 = 2nωMQW ϵ0 c0 |Ezω |2 , (2.87)

where nMQW denotes the refractive index of the MQW and c0 the speed of light in
vacuum [saturation]. As a consequence, a significant amount of electrons populate
excited states and our initial assumption about the state population becomes invalid.
This phenomenon is called saturation and limits the overall nonlinear process. In
order to properly take saturation effects into account during our calculations we
introduce the intensity-dependent nonlinear susceptibility:

(2)
χzzz,sat (Izω ) = S SHG (Izω )χ(2)
zzz , (2.88)

where we have just expanded the nonlinear susceptibility by an additional intensity-


dependent factor, the so called saturation factor S SHG (Izω1 ). This factor is unity at
low intensities, indicating negligible saturation effects. We also note that the non-
linear susceptibility became position-dependent, because the intensity varies across
the MQW volume.

2.3.4 Nonlinear Optical Response

In SHG processes the time- and position dependent incident electric field can be
written as:
Einc (r, t) = Eωinc (r)eiωt + c.c., (2.89)

where ”c.c.” denotes the complex conjugate and r the position. Eωinc1 (r) is composed
of the electric fields in x- and y-directions according to:

Eωinc (r) = Einc[x]


ω ω
(r)ex + Einc[y] (r)ey . (2.90)

ex and ey are the unit vectors along the x- and y-directions, respectively. This
incident field generates an electric field of the same frequency inside the quantum
well:
EMQW (r, t) = Eω (r)eiωt + c.c., (2.91)
2.3 Plasmonic Metasurfaces with Intersubband Transitions Page 25

which in turn induces a nonlinear polarization at twice the frequency:


∑ (2)
P2ω
i (r, t) = ϵ0 χijk (r)Ejω (r)Ekω (r)ei2ωt . (2.92)
jk

(2)
As discussed previously, all elements of the nonlinear susceptibility tensor χijk van-
(2)
ish, except χzzz , which relates the fields orthogonal to the MQW layers. For this
reason, the last equation can be simplified to:

Pz2ω (r) = ϵ0 χ(2) ω ω


zzz (r)Ez[a] (r)Ez[b] (r), (2.93)

ω
where a and b can either be x or y. Also, (r)Ez[a] (r) denotes the electric field com-
ponent in the z direction at point r inside the MQW generated by an incident field
of frequency ω polarized in the direction a. In electromagnetism the displacement
current density is defined as the time derivative of the electric displacement field
D, which can be expressed by the polarization and the electric field in the following
way:
D = ϵo E + P. (2.94)

Differentiation with respect to time yields the displacement current density:

∂E ∂P
JD = ϵ0 + . (2.95)
∂t ∂t

Here, the last term corresponds to the so called polarization current density, which
we will denote with a simple J for simplicity. As a result, we can now derive an
expression for the polarization current density for our polarization in the z-direction
from eq. (2.93):
Jz2ω (r) = i2ωϵ0 χ(2) ω ω
zzz (r)Ez[a] (r)Ez[b] (r). (2.96)

With the help of Lorentz reciprocity theorem, we can link this polarization current
density to the electric field radiated towards free-space at frequency 2ω. The Lorentz
reciprocity theorem states:
∫ ∫
E2ω 2ω
MQW (r)JMQW (r)dV = E2ω 2ω
FS (r)J(2D) (r)ds, (2.97)
VUC S

where the left side corresponds to an integration over the unit-cell volume (”VUC ”)
and the right side denotes a surface integral. The surface has the same area as the
unit-cell and is located in free-space in the far field. We are interested in finding
the electrical field radiated into free-space at frequency 2ω, thus we want to solve
eq. (2.97) for E2ω 2ω
FS . The surface current density J(2D) can be expressed in terms of
Page 26 2 Theory

an incident electric field with the help of the free-space impedance η:

E2ω
inc
J2ω
(2D) = . (2.98)
η

As a result, we can express the electric field generated by the unit cell as:
∫ ( 2ω
)
2ω 1 Ez[m] (r)
E[mab] = i2k χ(2) ω ω
zzz (r)Ez[a] (r)Ez[b] (r) 2ω
dV. (2.99)
S VUC Einc[m]


Here, we converted the expression ωϵ0 η to the free-space wavevector k and E[mab] cor-
responds to the m-polarized electric field with frequency 2ω generated by the fields
ω ω
Einc[a] and Einc[b] , which are both impinging the metasurface at normal incidence.

On the other hand, we can write the overall nonlinear polarization response of the
metasurface with an effective transverse polarizability density:

(2)
P2ω ¯ ω ω ω
eff = ϵ0 χ̄eff (I )Einc Einc , (2.100)

(2)
where χ̄¯eff (Iω ) is the effective nonlinear susceptibility tensor. In order to account
for saturation effects, it also depends on the intensity. Similar to eq. (2.96), we can
associate this to an effective transverse polarization current according to:

J2ω
P2ω
eff = eff . (2.101)
i2ω

By relating this current density to the radiated electric field through:

2E2ω
J2ω
eff = , (2.102)
ηd

where d is the thickness of the structure, we arrive at the following expression for
the radiated electric field:

(2)
E2ω = i2kdχ̄¯eff (I ω )Eωinc Eωinc . (2.103)

We can rewrite this for the case of individual input and output polarizations to
obtain a complementary expression to eq. (2.99):

2ω (2)
E[mab] = i2kdχeff[mab] (I ω )Einc[a]
ω ω
Einc[b] . (2.104)

Now, we combine eq. (2.99) with eq. (2.104) to get our final expression for the
2.3 Plasmonic Metasurfaces with Intersubband Transitions Page 27

individual elements of the effective transverse nonlinear susceptibility tensor:


∫ ω ω 2ω
(2) 1 Ez[a] (r) Ez[b] (r) Ez[m] (r)
χeff[mab] (I ω ) = χ(2)
zzz (r) ω ω 2ω
dV. (2.105)
V VUC Einc[a] Einc[b] Einc[m]

When we neglect saturation, the MQW’s intrinsic nonlinear susceptibility becomes


position-invariant and we can rewrite this expression as:

(2) ∫ ω ω 2ω
∫ ω ω 2ω
(2) χzzz Ez[a] (r) Ez[b] (r) Ez[m] (r) VUC
ξ[a] ξ[b] ξ[m] dV
χeff[mab] = ω ω 2ω
dV ≡ χ(2)
zzz , (2.106)
V VUC Einc[a] Einc[b] Einc[m] | V
{z }
overlap
I[mab]

where the factor on the right hand side denoted by the underbrace is the so called
overlap integral. The overlap integral essentially enhances the intrinsic nonlinear
susceptibility and thus is of paramount importance for a well performing metasur-
face. This enhancement arises from the coupling of incident light into the MQW
and conversion of the electric fields into the z-direction and is expressed We can
understand the importance of this effective transverse nonlinear susceptibility when
we take a look at the expression for the conversion efficiency for a SHG process
[saturation]:
I 2ω 1 (2)
2
ηSHG = ω = ηI ω k 2 d2 χeff[mab] (I ω ) . (2.107)
I 2
The conversion efficiency is quadratically proportional to the effective nonlinear
susceptibility, thus improving the overlap integral provides a massive boost for the
performance of the metasurface and poses a major challenge for the metasurface
design and fabrication.

2.3.5 Finite Size Excitation

During our experiments we will use lasers in order to excite the nonlinear processes.
For that reason we assume a gaussian beam profile described by:

I ω (r) = I0ω e−2r


2 /w 2
. (2.108)

Here, w denotes the beam radius and r the distance from the beam center. The
beam radius w can be estimated using the knife edge technique. The overall power
of the beam is related to the intensity profile according to:
∫ ∫
P0ω = I ω (r)ds, (2.109)
Page 28 2 Theory

I0 I ω (r)
I 2ω (r)

Intensity I(r) (a.u.)


0.8

0.6

0.4
w

0.2 2 w

−40 −20 0 20 40
Radius r (µm)

Figure 2.8: Intensity profiles for a 20 µm wide pump beam (red) and the resulting
SH beam (blue).

from which we obtain the following relationship after carrying out the integration:

2P0ω
I0ω = . (2.110)
πw2

Since the SH intensity has a square dependence on the input intensity at ω, also the
SH intensity profile is squared accordingly:

I 2ω (r) = I02ω e−4r


2 /w 2
. (2.111)

By substituting w with wSH = w/ 2 we get the equivalent expression to eq. (2.108)
for the SH case. This indicates that the resulting SH intensity profile possesses a

beam width reduced by a factor of 1/ 2, which is also depicted in fig. 2.8. Following
a similar integration as in eq. (2.109) yields the relationship between the peak SH
intensity and the SH power:
4P 2ω
I02ω = 0 2 . (2.112)
πw
As a result, the conversion efficiency in terms of intensity is twice as large as the
power conversion efficiency. This directly follows from the division of eq. (2.112)
and eq. (2.110):
I02ω P02ω
= 2 . (2.113)
I0ω P0ω

For the sake of compactness, we omit the 0-subscript for the peak intensity values
in all equations and calculations.
3 Methods

3.1 Simulation

In this section we want to discuss the employed simulations to predict and confirm
our experimental results. We will deal with optical as well as thermal considerations.
All of our simulations are carried out with COMSOL Multiphysics.

3.1.1 Optical Simulation

We use optical simulations to calculate important characteristics of different meta-


surface configurations. We mainly employ these simulations to find optimal antenna
shapes, i.e. to maximize the emitted SH power. Figure 3.1 illustrates a unitcell of
the metasurface containing one antenna. In order to reproduce the behaviour of
a complete metasurface rather than just one antenna, we apply periodic boundary
conditions to each sides marked by dashed lines. The optical parameters of the ma-
terials are taken partly from literature as well as from our own calculations. We can
deduce the electric susceptibility in the out-of-plane direction for the MQW from
the intersubband-absorption measurement.

3.1.2 Thermal Simulation

3.2 SHG-Measurement

3.3 Intersubband Absorption Measurement

The quantum configuration of the MQWs grown in the MBE is measured in a


multipass geometry using an infrared Fourier transform interferometer. For this, a
small rectangular piece of the wafer is cleaved off and two opposite sides are polished
Page 30 3 Methods

Iinc
Gold
A MQW

Figure 3.1: Caption

Figure 3.2: Caption

to have 45° facets as indicated in fig. 3.3. One edge is illuminated at normal incidence
and the resulting transmission is measured with a mercury cadmium telluride (MCT)
detector. The intersubband absorption coefficient αW can be expressed as:
( )
1 ITM
αW =− ln (10) log10 . (3.1)
Lint ITE

Here, ITM and ITE correspond to the measured intensities for transverse-magnetic
and transverse-electric polarized incoming light, respectively. Furthermore, Lint de-
notes the interaction length according to:

np LQW N
Lint = , (3.2)
cos (θ)

where np is the number of passes through the quantum well layer, LQW is the height
of the individual quantum well sections, N denotes the total number of these sections
and θ corresponds to the angle to which the wafer piece is polished to (45° in our
case). The number of passes np can be estimated with the samples length L and
thickness d:
L
np = tan (θ). (3.3)
2d
3.3 Intersubband Absorption Measurement Page 31

Gold
MQW 45°

d
InP

Figure 3.3: Multipass geometry for intersubband absorption measurement. The two
facets are polished at 45°. The contributions to the interaction length
Lint are indicated by the thick red lines inside the MQW region.

As a consequence, the interaction length Lint is the accumulated length of the thick
red line inside of the MQW-region in fig. 3.3.
Once we have obtained the intersubband absorption coefficient, we can deduce the
actual doping level Ne of our structure. This can differ from the nominal doping
and is therefore important to determine. The absorption coefficient of a plane wave
propagating through a material is generally defined as:

α = 2n′′ ω/c, (3.4)

where n′′ denotes the imaginary part of the complex refractive index n. The linear
dielectric constant and the refractive index obey the following relationship:

n(ω) = ϵ(1) (ω). (3.5)

Combining the last two equations yields an expression for the intersubband absorp-
tion coefficient in terms of the dielectric constant for electric fields polarized along
the out-of-plane direction:

2ω √
αW = Im{ ϵ⊥ } sin 2 (θ), (3.6)
c

where we have also added the correction factor sin 2 (θ) accounting for the incidence
angle with respect to the layers. Furthermore, we can calculate the out-of-plane
dielectric constant with the help of the linear susceptibility found in eq. (??)
4 Metasurface Fabrication

In this section we will focus on the fabrication process of metasurfaces from start to
finish. All major processing steps are illustrated in fig. 4.1. The process starts with
the molecular beam epitaxy (MBE), where the quantum wells are grown on top of
an etch stop (4.1a). In a second step, this wafer is bonded to a host InP-wafer with
gold (4.1b). The InP-subtrate and the etch stop from the MBE wafer are removed
during substrate removal, such that the quantum wells are now directly on top of a
gold layer (4.1c). SiN is added as a hardmask, which is followed by spin-coating a
photoresist (4.1d). After exposing and developing the resist (4.1e), the underlying
SiN hardmask is developed with dry etching (4.1f). At this point, the resist can be
removed and the quantum wells are etched with the newly developed Cl-Ar recipe
(4.1g). After removing the hardmask (4.1h), the top surface is covered in gold, using
evaporation (4.1i).

4.1 Wafer Preparation

The plasmonic metasurfaces considered in this work require a complex geometry and
layer structure. This layer structure consists of a substrate material, a layer of gold,
a series of multiple quantum wells and another gold layer on top. These advanced
requirements demand some preliminary fabrication steps before the actual geome-
try of the metasurface antennas can be processed. The quantum wells are grown
during molecular beam epitaxy along with some other important layers required for
substrate removal. During wafer bonding, a layer of gold is implemented between
the substrate and the quantum wells. After the substrate has been removed, the
antenna patterns can be developed.

4.1.1 Molecular Beam Epitaxy

The metasurfaces fabricated during this work are capable of achieving an excep-
tionally high second-order nonlinear susceptibility χ(2) with the help of quantum
Page 34 4 Metasurface Fabrication

InP substrate
Gold
Quantum wells Quantum wells Quantum wells
Etchstop Etchstop Gold
InP substrate InP substrate InP substrate

(a) Wafer after epitaxy (b) Bonded wafers (c) After substrate removal

Photoresist
SiN hardmask SiN hardmask
Quantum wells Quantum wells Quantum wells
Gold Gold Gold
InP substrate InP substrate InP substrate

(d) Hardmask and resist (e) Developed resist (f) Developed hardmask

Gold Gold Gold


InP substrate InP substrate InP substrate

(g) After Cl-Ar etch (h) Removed hardmask (i) After gold evaporation

Figure 4.1: Illustration of the metasurface fabrication process from start to finish.

confinement of electrons in quantum wells. The quantum wells consist of several


layers of semiconductors with different bandgap energies, e.g. a series of AlInAs and
GaInAs layers are grown on top of an etch stop using molecular beam epitaxy. The
etch stop is used in a later step for removing the semi-insulating InP substrate.

4.1.2 Waferbonding

During waferbonding, two wafers with gold on their surfaces are brought into contact
and are thermocompressively welded together. In our case, the wafer from the MBE
is bonded onto a host wafer consisting of InP. As a result, we obtain a gold layer
sandwiched between two semiconductor wafers.

4.1.3 Substrate Removal

In this step we want to remove the InP substrate and the etchstep, revealing the
quantum well layers on top of the gold layer. First, the InP substrate is etched
using highly concentrated hydrochloric acid (HCl:H2 O [4:1]). Once a clear surface is
visible, we switch to a less reactive solution (HCl:H2 O [2:1]) to finish the substrate
removal at a slower rate. The etch stop layer is removed with phosphoric acid
4.2 Surface Structuring Page 35

(H3 PO4 :H2 O2 :H2 O [1:1:8]) followed by a light HCl etch (HCl:H2 O [1:1]) for a thin
layer of InP. After finishing the removal of the substrate, the end result consists of
a InP-substrate with the quantum well layers on top and a layer of gold sandwiched
in between and can now be processed further by structuring the antenna shapes.

4.2 Surface Structuring

The patterning of the antennas on the surface is done with electron-beam lithography
on a hardmask material. A hardmask is needed because a typical organic photoresist
would not withstand the high table temperatures and the Ar-sputtering of the RIE
system. After spin-coating the hardmask with photoresist specifically designed for
electron-beam lithography, the pattern is exposed by the electron beam and in the
end developed.

4.2.1 Hardmask Deposition

As a first step, Si3 N4 is deposited as a hardmask material on the wafer’s surface


using a PECVD machine. In a plasma consisting of Silane (SiC4 ) and Nitrogen,
Si3 N4 is formed and can bond to the wafers surface. Our recipe choice is already well
established and grows Si3 N4 at a rate of approximately 10 nm/s. In order to achieve
a depth of 300 − 400 nm during etching, 100 nm of Si3 N4 and therefore 10 minutes of
PECVD is needed. Technically, much less Si3 N4 is required assuming a selectivity
of 10 for the Cl-Ar etch, but we also want to account for process fluctuations using
a conservative safety margin.

4.2.2 Electron-Beam Lithography

Before spinning on the resist, we remove any residual humidity on the wafers surface
by baking it for at least 20 minutes at 100 ◦C, followed by spinning on HDMS for pro-
moted adhesion. As a next step we coat the wafer in photoresist (SX-ARN 7520.17),
which is then baked on the hotplate for 120 s at 85°. The electron-beam lithography
is carried out at 30 kV, which is the highest possible EHT voltage at our machine.
Generally, finer details can be achieved with higher EHT voltages. The correct ex-
posure dosage must be determined beforehand using a dosetest. After exposure, we
develop the photoresist in the developer AR 300-47 for approximately 100 s. Inspect-
ing the sample with the microscope helps to determine the exact delevoping time.
Page 36 4 Metasurface Fabrication

Figure 4.2: Si3 N4 hardmask after development.

Before moving on to the etching of the hardmask, the photoresist must be tempered
for 120 s at 85° using the hotplate. During this step, the photoresist develops the
required strength to endure plasma etching.

4.2.3 Hardmask Development

Once the resist is developed, the Si3 N4 can be etched, revealing the semiconductor
surface. This step can be easily monitored using in-situ control. By measuring
the intensity of a red lasers reflection, pointed vertically on the wafer’s surface
during etching, the etch rate can be monitored. Due to constructive and destructive
interference, the reflected intensity oscillates because the distance between the laser
and the reflection surface increases as the wafer is etched away. Once no Si3 N4
remains, the intensity oscillation stops, indicating the end of process. Overetching
for 15 − 30s can improve the pattern’s detail.
After developing the hardmask, the photoresist must be removed, either by dipping
the wafer in acetone or with the help of an oxygen plasma in the barrel etcher. As
a last step, the wafer should be inspected with a microscope in order to check for
any resist residues.

4.2.4 Cl-Ar Etching

This is the most critical step of the fabrication process, so we need to take special
care. In order to avoid any mistakes due to changes in process parameters, it is
4.2 Surface Structuring Page 37

(a) Si3 N4 -covered nanoresonators after (b) Nanoresonators after removal of the
etching the MQW. Si3 N4 -hardmask.

important to always run a test etch on a dummy sample before proceeding to the
main sample. This also helps to ensure that the RIE recipe remains functional when
processing the main sample. During the course of this work, it sometimes happened
that the developed RIE recipe stopped working, even when no parameters have
changed. This could be due to the fact, that many etching processes with different
materials are carried out at our RIE machines and therefore the material deposition
on the inner chamber walls constantly changes. A dummy sample also helps to
determine the current etch rate, as it can also vary over time. Once assured of a
functional etching recipe, we can continue to process the main sample. The in-situ
laser intensity signal indicates when all of the quantum wells are etched through,
because the underlying metal layer stops the etch. Overetching for approx. 10 s
increases the etch quality slightly. After etching, the results should be evaluated
using the SEM. At this point it is still possible to re-etch the sample for a second
time if needed. As a last step, the Si3 N4 hardmask is removed in the RIE using the
same recipe as for the hardmask development.

4.2.5 Top Gold Evaporation

It is possible to potentially ruin the metasurface with this step, therefore special
care is required when adding the topmost gold-layer to the structure. The gold
deposition is done by evaporation of solid gold using an electron beam for heating.
It is of utmost importance, that no gold deposits on the sidewalls of the antennas, as
this would render the metasurface completely inoperable. It is for this reason that
the gold vapor must hit the wafer surface as vertically as possible, which requires
high vacuums in the evaporation chamber. One possible alternative approach to
overcome this high-risk processing step is to add the gold cover right after substrate
removal. This gold layer is then etched through during the Cl-Ar etching step,
Page 38 4 Metasurface Fabrication

Figure 4.4: Finished nanoresonators after gold evaporation.

yielding a similar final metasurface.


5 Results

We will now evaluate and discuss the experimental results obtained from numerous
experiments for different metasurface configurations.

5.1 Intersubband Absorption Measurement

We begin our results discussion by evaluating the grown MQW structure. By mea-
suring the intersubband absorption spectrum described in the previous sections, we
can draw conclusions about the energy eigenstates and the intersubband transitions
of the layers obtained from the MBE. We can also estimate the actual doping level,
the dipole moments and the transition linewidths. The best fit is achieved with
E12 = 255 meV and E13 = 310 meV. The dipole moments z12 and z13 are evaluated
as 0.95 nm and 0.5 nm, respectively. The transition linewidths ℏγ12 and ℏγ13 are both
40 meV. Since we cannot deduce the linewidths for the remaining transitions, we
assume them as 15 meV. The introduced doping is estimated as N = 8 × 1018 cm−3 ,
which is already much larger when compared to the doping level found for the semi-
resonant SHG from [ADOM]. With these high levels of doping, we are able to
(2)
achieve huge second-order nonlinear susceptibilities χzzz as shown in fig. 5.1. Here,
(2)
we calculated the resulting value of χzzz with the help of eq. (2.83) and the fitted
parameters. As intended, the maximum value is reached at around 130 meV.

5.2 T-Shaped Nanoresonators

We begin the discussion of the results by investigating the measurements obtained


from the T-shaped nanoresonators. In order to achieve efficient in-coupling of the
incoming light at ω and out-coupling of the SH-radiation at 2ω various antenna
arrays with different x- and y-scalings were patterned during EBL and processed at
once. We were able to obtain well performing nanoresonators during two separate
fabrication processes. We will now discuss the measurement results of the two best-
Page 40 5 Results

·104 ·105
Measurement 3
1.2 Fit
2.5

χzzz (pm/V)
2
α (cm−1 )

0.8 130 meV

260 meV
1.5

130 meV

(2)
0.4 1

0.5

0 0
100 150 200 250 300 350 400 100 150 200 250 300
Photon Energy ℏω (meV) Photon Energy ℏω (meV)

Figure 5.1: Left: Intersubband absorption measurement (red) with theoretical fit
(orange). Right: Absolute value of the second-order nonlinear suscepti-
(2)
bility χzzz resulting from the fitted parameters.

performing antenna shapes. SEM images of these two antenna shapes are depicted
in fig. 5.2. For the sake of simplicity we will from now on refer to the antenna on
the left as ”Sample 1” and to the antenna on the right as ”Sample 2”.

The absorption spectra of the two antenna arrays are shown in fig. 5.4. We have also
plotted the wavelength dependence of the intrinsic second-order nonlinear suscepti-
(2)
bility |χzzz |, which peaks at 129 meV (λmax ≈ 9.6 µm). It is obvious that maximum
performance is obtained when the nanoresonators are able to efficiently couple in
light at λmax . Peaks in the absorption spectrum of x-polarized light point to proper

(a) Sample 1. (b) Sample 2.

Figure 5.2: SEM images of the two antenna shapes with the best performance.
5.2 T-Shaped Nanoresonators Page 41

sx Lx
Sample 1 Sample 2
dy Lx 2160 2130
Ly
Ly 1140 905
dx 340 765
dy 370 350
dx sy
sx 423 1000
sy 390 300

Figure 5.3: Schematic of the T-shape Table 5.1: Dimensions of the


unit cell with all antenna di- two samples mea-
mensions and the horizontal sured with SEM.
and vertical spacings. All dimensions are
given in nm.

incoupling because they indicate a resonance of the antenna. From fig. 5.4 we can
observe that sample 1 and 2 (shown in red and blue, respectively) absorb x-polarized
(2)
incoming light almost equally well in the peak region of |χzzz |. Strong absorption
of y-polarized light at the SH-wavelength ensures efficient outcoupling, because the
T-shaped antennas are designed to re-emit y-polarized SH-radiation towards free
space. If we assume a pump wavelength of 9-10 µm (indicated by the light red inter-
val in fig. 5.4) the corresponding SH wavelengths lay between 4.5 µm and 5 µm. The
absorption spectra for y-polarized light show broad peaks extending into the SH-
region (light green interval) for both samples, as depicted by the orange and cyan
curves in fig. 5.4. It is remarkable that, despite the different geometry, the spectra
for both samples follow a similar trend for the most part. Major differences are only
apparent in the absorption spectra for x-polarized light at higher energies, which
are not relevant for us. It is also worth noting that the y-absorption peak height in
the SH region for sample 1 is lower than for sample 2, which could potentially hint
to a more efficient outcoupling of SH-radiation for sample 2.

In fact, fig. 5.5 shows that the performance of sample 2 is superior. It depicts the
measured SH-power/intensity against the square of pump power/intensity in order
to illustrate the quadratic relationship deduced from eq. (2.107):

1 (2)
2
I 2ω = η(I ω )2 k 2 d2 χeff . (5.1)
2

In order to better illustrate saturation, we have also fitted each measurement with
a linear function. The SH power for sample 1 deviates slightly from the linear trend
and starts to slope down at high pump powers. The downward sloping could hint
towards the onset of saturation phenomena. Saturation could either be due to ex-
Page 42 5 Results

Wavelength λ (µm)
10 9 8 7 6 5 4
1 1.5 · 105

0.8

χzzz (2ω, ω, ω) (pm/V)


1 · 105
0.6
1-R

0.4
Sample 1 x
Sample 1 y 5 · 104

(2)
Sample 2 x
0.2
Sample 2 y
(2)
|χzzz |
0 0 · 100
100 150 200 250 300 350 400
Photon Energy ℏω (meV)

Figure 5.4: Absorption spectra of Sample 1 and 2. A reasonable region for the pump
energy is indicated by the interval in light red, together with the resulting
SH interval in light green.

tremely high field intensities in some regions of the MQW or caused by thermal
heating from the incident radiation. From the large absorption value at the pump
wavelength in fig. 5.4 we can deduce that a major part of the energy of the incident
light pulse gets absorbed and thereby transferred into the antenna volume. Here, we
can neglect the generated SH radiation as the SH conversion efficiency is minuscule
(≤ 1 % in our case). As a result, we assume that all of the absorbed light energy
is converted into thermal energy and causes the antennas to heat up. Thermal
management in MQW based structures often poses a challenge because the thermal
conductivity in the layer growth direction is significantly reduced when compared
to bulk materials. Heat build-up can affect the antenna performance in a similar
way as high field intensities. Through thermal excitation, electrons in the ground
state are able to populate elevated states and thereby cannot contribute to the SHG
process anymore. In order to estimate the possible consequences for SH performance
and confirm our assumption we have carried out a transient thermal analysis. In
fig. 5.6 we show the temperature distribution inside the MQW region at the end of
a 40 ns light pulse with 70 kW/cm2 intensity, approximately corresponding to the
maximum input power under experiment conditions. We can observe a large tem-
perature gradient in the z-direction inside the MQW region, which is presumably
caused by the poor thermal conductivity perpendicular to the layers. According to
5.2 T-Shaped Nanoresonators Page 43

Pump Intensity squared (kW2 /cm4 )


0 2000 4000 6000
30
Sample 1
Sample 2 10
25

SH Intensity I02ω (W/cm2 )


SH Power P02ω (µW)

2 8
20
/W
mW
15 15 6
0.

10 2 4
/W
mW
0.1 2
5

0 0
0 0.05 0.1 0.15 0.2
Pump Power squared (W2 )

Figure 5.5: SH power measurement for sample 1 and 2.

the simulation, a maximum temperature of almost 150 ◦C is reached near the top
gold-MQW interface. Excess heat at the bottom interface can easily be conducted
through the gold layer into the InP-substrate. The large heat build-up near the top
interface is critical for the saturation behaviour of the metasurface, because the sur-
face plasmons already create huge field intensities in that region. The superposition
of these two effects could potentially lead to the onset of saturation, even for our
MQW structures, which are optimized for extremely large saturation intensities.

So far, we have just discussed the measurement results for sample 1, let us now
analyze the SH measurements for sample 2. From fig. 5.5 it becomes apparent that
sample 2 not only performs better overall, but also achieves to follow the linear trend,
even for the highest applied pump powers. As pointed out previously already, the
absorption value for sample 2 in the SH region is larger than for sample 1, presumably
pointing towards a more efficient outcoupling of the generated SH-radiation, which
in turn would affect the overlap integral and thereby explain the increased slope of
the linear fit for sample 2 in fig. 5.5. The difference in antenna geometry is most
likely the reason for the enhanced SH absorption. More precisely, the increased dx
and shorter Ly in comparison to sample 1 seem to benefit the overall performance
of sample 2. Aside from the larger spacing in x-direction, this presents the major
geometry change for sample 2.
Page 44 5 Results

Figure 5.6: Temperature distribution inside the MQW after 40 ns exposure at


70 kW/cm2 .

From these measurements and with the help of eq. (5.1) we can also deduce the
(2)
effective nonlinear susceptibility χeff and the resulting overlap integral. From eq.
(5.1) we get an expression for the slope of the linear fit a, which can be rearranged
(2)
to solve for the absolute value of χeff :

(2) 2a
χeff = . (5.2)
ηk 2 d2

Using the slopes from fig. 5.5, we obtain an effective nonlinear susceptibility of
2.37 × 105 pm/V and 2.89 × 105 pm/V for sample 1 and sample 2, respectively.
From the absorption measurements we obtained an estimation for the peak absolute
(2)
value of the intrinsic nonlinear susceptibility χzzz of approximately 1.35 × 105 pm/V.
The overlap integral, i.e. the enhancement of the intrinsic nonlinearity, therefore
evaluates to 1.75 for sample 1 and 2.14 for sample 2.

Furthermore, a transient thermal analysis for the particular shape of the antenna of
sample 2 yielded a maximum temperature inside the MQW volume of approximately
130 ◦C at the end of a 40 ns light pulse. This 20 ◦C difference compared to sample 1
could beneficially affect the saturation performance.
5.3 V-Shaped Nanoresonators Page 45

Pump Intensity (kW/cm2 )


0 10 20 30 40 50 60 70 80
0.008 0.016
Sample 1

Intensity Conversion Efficiency(%)


Power Conversion Efficiency (%)

Sample 2 0.014

0.006 0.012

0.01

0.004 0.008

0.006

0.002 0.004

0.002

0 0
0 100 200 300 400
Pump Power (mW)

Figure 5.7: Caption

5.3 V-Shaped Nanoresonators

Now, we want to discuss the results for V-shaped nanoresonators. With optical
simulations we were able to achieve huge field enhancements at the tips of the two
arms of the V-shape.
Page 46 5 Results

Pump Intensity squared (kW2 /cm4 ) Pump Intensity (kW/cm2 )


0 2000 4000 6000 0 10 20 30 40 50 60 70 80
30 0.008
Sample 1 Power Conversion Efficiency (%) Sample 1
Sample 2 10 Sample 2
25

SH Intensity I02ω (W/cm2 )


0.006
SH Power P02ω (µW)

2 8
20
/W
mW
15 15 6
0. 0.004

10 2 4
/W
mW 0.002
0.1 2
5

0 0 0
0 0.05 0.1 0.15 0 0.2 100 200 300 400
2
Pump Power squared (W ) Pump Power (mW)
6 Conclusions and Outlook

You might also like