materials

Article
Atomistic Simulations of Plasma-Enhanced Atomic
Layer Deposition
Martin Becker and Marek Sierka *
Otto Schott Institute of Materials Research, Friedrich Schiller University Jena, 07743 Jena, Germany
* Correspondence: marek.sierka@uni-jena.de




Received: 25 July 2019; Accepted: 12 August 2019; Published: 15 August 2019


Abstract: Plasma-enhanced atomic layer deposition (PEALD) is a widely used, powerful layer-by-layer
coating technology. Here, we present an atomistic simulation scheme for PEALD processes, combining
the Monte Carlo deposition algorithm and structure relaxation using molecular dynamics. In contrast
to previous implementations, our approach employs a real, atomistic model of the precursor. This
allows us to account for steric hindrance and overlap restrictions at the surface corresponding to the
real precursor deposition step. In addition, our scheme takes various process parameters into account,
employing predefined probabilities for precursor products at each Monte Carlo deposition step. The
new simulation protocol was applied to investigate PEALD synthesis of SiO2 thin films using the
bis-diethylaminosilane precursor. It revealed that increasing the probability for precursor binding to
one surface oxygen atom favors amorphous layer growth, a large number of –OH impurities, and the
formation of voids. In contrast, a higher probability for precursor binding to two surface oxygen
atoms leads to dense SiO2 film growth and a reduction of –OH impurities. Increasing the probability
for the formation of doubly bonded precursor sites is therefore the key factor for the formation of
dense SiO2 PEALD thin films with reduced amounts of voids and –OH impurities.

Keywords: plasma-enhanced atomic layer deposition; Monte Carlo simulation; molecular dynamics
simulations; density functional theory; ReaxFF reactive force field

1. Introduction
Atomic layer deposition (ALD) is a coating technology based on self-terminating reactions of
gaseous precursor molecules with surface functional groups of the substrate. As shown in Figure 1,
a precursor vapor (usually an organometallic compound) is introduced in the chemical reactor. It reacts
with the functional groups at the surface of the substrate, until the surface is saturated (first half-cycle).
Excess precursor and reaction byproducts are purged by inert gases. Next, a second precursor is pulsed
in the reactor, such as an oxidizing agent like H2 O, ozone, O2 , or O2 plasma, which reacts with the
remaining ligands of the precursor material (second half-cycle). The reactor is purged again. The
ALD process is continued by repeating the above pulsing and purging steps, which constitute the
ALD cycle, until the desired film thickness is reached. These reactions depend only on the abundancy
and variety of the surface functional groups for a specific precursor, and not on the shape of the
substrate. Hence, the ALD technology leads to extremely conformal coatings on highly curved [1–5] or
nanostructured substrates. Additionally, the composition of the thin films can be controlled with atomic
precision by alternating the organometallic compound for binary or more complex mixtures [1,5].
Plasma-enhanced ALD (PEALD), which utilizes oxygen plasma, is applied in order to increase the
reactivity of the oxidizing agent [6]. This enables, for example, room temperature deposition of SiO2
thin films, which is possible only at 200 ◦ C in typical thermal processes using O3 [7]. However, recently
developed highly reactive precursors allow for thermal ALD processes at temperatures approaching
room temperature [8,9]. The O radical species present in the plasma allow for additional tailoring of

Materials 2019, 12, 2605; doi:10.3390/ma12162605 www.mdpi.com/journal/materials

Materials 2019, 12, 2605 2 of 11

material properties [10,11]. For example, in the case of TiO2 coatings, PEALD films are denser and
have a higher refractive index [12].
Atomistic simulations of PEALD processes are of particular interest for the fundamental
understanding of the growth processes, mechanisms of material densification and crystallization, void
formation, intermixing, and mechanical properties of the coatings at an atomic level [13]. However, such
simulations are very challenging due to the usually significant number of competing chemical reactions
and the thickness-dependent structure relaxation of the growing surface [13–15]. For mechanistic
questions, such as reaction paths, reaction energies and barriers electronic structure approaches are the
methods of choice [13]. In particular, the advent of modern density functional theory (DFT) based on
accurate gradient-corrected and hybrid functionals allowed for the identification and understanding of
elementary chemical reactions of a number of ALD processes [13]. For example, the understanding
of SiO2 ALD mechanisms has proceeded through a number of computational studies employing
DFT [16]. The reactivities have been investigated for a number of aminosilane precursors [17–23],
which display low activation energies for reactions with surface hydroxyl groups. Despite the progress
in the applications and development of electronic structure methods, their computational cost is still too
high to reliably simulate the overall ALD growth [13,15]. This task can only be achieved by employing
more approximate methods capable of extending the spatiotemporal scale accessible for simulations.
The two such approaches applied for studies of ALD growth dynamics are the atomistic kinetic Monte
Carlo (KMC) model and molecular dynamics (MD) [24].
KMC is a stochastic method intended to simulate the time evolution of processes that occur with
priori known transition rates among states. The main advantage of the KMC method is its ability
to model a broad range of time scales [24]. However, it relies on an accurate knowledge of rate
constants for all elementary reaction steps. Such information is difficult to obtain by employing DFT
due to the deficiencies of existing exchange–correlation functionals and requires more accurate and
computationally much more demanding electronic structure methods. Another disadvantage of the
KMC approach is its inability to simulate the morphology of the growing ALD film, i.e., crystalline
vs. amorphous growth. Atomistic KMC simulations using DFT-derived reaction mechanisms and
activation energies have been used, e.g., to investigate ALD growth of HfO2 [14,25].
In MD simulations, the time evolution of interacting atoms is described directly by solutions of
the corresponding equations of motion [24]. Therefore, the MD method, in combination with carefully
parameterized potential functions or reactive force fields, is able to describe relaxation phenomena
and defects in ALD-generated films. However, the temporal scale of MD simulations (ps to ns) is
very small compared to the duration of ALD pulses. This problem has been addressed by Hu et al. in
MD simulations of Al2 O3 ALD by separating the large time scale of surface reactions (ns to s) from
the small time scale of structural relaxation (ps) [15]. Their deposition algorithm assumes that ALD
reactions occur only on the active –OH groups on the growing surface and that the products of the
metal precursor pulses can be fully hydroxylated, i.e., –Al(CH3 )2 is fully converted to –Al(OH)2 . The
direct (large time scale) simulations of surface reactions are replaced by a deposition algorithm. This
starts with a hydroxylated surface and randomly picks one of the available surface –OH groups for
the deposition of the ALD product (–Al(OH)2 ). Approximate steric and overlap restrictions with
neighboring atoms are checked and, if satisfied, an H atom in the selected –OH group is replaced by
–Al(OH)2 . After each deposition step, the structure is relaxed using MD simulation. Unlike KMC,
this MD-based approach enables us to study the thickness-dependent evolution of the microscopic
structures of ALD layers. In addition, the influences of operating parameters, such as precursor
type, temperature, external fields, initial surface structure (crystalline vs. amorphous), number and
distribution of –OH, etc. on the ALD process can be investigated at the atomic level.
Due to a high computational cost, MD simulations of large systems are rarely performed using
electronic structure methods. Instead, more approximate approaches, such as interatomic potential
functions or force fields, are used. However, the applicability of many interatomic potentials is
restricted to one element oxidation state and a small number of polymorphs [26]. A relatively new and
Materials 2019, 12, 2605 3 of 11

promising trend in the development of simulation methods for nanomaterials is the use of so-called
Materials 2019, 12, 2605
reactive force fields, such as the ReaxFF approach of van Duin and co-workers [27]. This 3has of 11
been
applied to a wide range of materials including amorphous and crystalline SiO2 [28,29] and Al2 O3 [30],
and promising trend in the development of simulation methods for nanomaterials is the use of
yielding very reactive
so-called good agreement
force fields,with
such experimental
as the ReaxFF approach data. Although
of van Duinparameterization
and co-workers [27]. of ReaxFF
This is
relatively
has been applied to a wide range of materials including amorphous and crystalline SiO2 [28,29] and of
complex, it is very powerful since it allows for an accurate description of the chemistry
nanostructures
Al2O3 [30], with
yielding a computational
very good agreement cost farwith below that of quantum
experimental mechanical
data. Although methods. of
parameterization In our
recentReaxFF
study, is werelatively
demonstratedcomplex, that the
it is structure
very powerful of since
amorphous
it allowsSiO could
for2 an be accurately
accurate description described
of the at
chemistry
the ReaxFF levelofby nanostructures with a computational
employing relatively small semi-amorphous cost far below that of
periodic quantum
models [31].mechanical
A simulation
methods. In our recent study, we demonstrated that the structure
cell containing only 64 SiO2 units yielded a structure that properly reproduced experimental of amorphous SiO2 could be
structural
accurately described at the ReaxFF level by employing relatively small
parameters, mechanical properties, and IR spectra of bulk silica, using subsequent refinement at the semi-amorphous periodic
models [31]. A simulation cell containing only 64 SiO2 units yielded a structure that properly
DFT level.
reproduced experimental structural parameters, mechanical properties, and IR spectra of bulk silica,
In this article, we present an atomistic simulation scheme for a PEALD process that combines the
using subsequent refinement at the DFT level.
Monte Carlo (MC)
In this deposition
article, we presentalgorithm and structure
an atomistic simulationrelaxation
scheme forusing
a PEALDMD.process
Our scheme is based on
that combines
the onetheproposed
Monte Carlo by (MC)
Hu etdeposition
al. [15]. However,
algorithmin and contrast
structure to relaxation
their implementation
using MD. Our that directly
scheme deposits
is based
hydroxylated products of the second ALD half-cycle, we use a real, atomistic
on the one proposed by Hu et al. [15]. However, in contrast to their implementation that directly model of the precursor
deposited in the
deposits first half-cycle.
hydroxylated productsIn addition,
of the secondfor theALDMC deposition
half-cycle, step,
we use our atomistic
a real, approachmodel introduces
of the site
precursor
occupation deposited in
probabilities forthe first half-cycle.
precursor reaction Inproducts.
addition, for Thethe MC deposition
performance step,simulation
of our our approach scheme
introduces site
is demonstrated foroccupation probabilities
PEALD synthesis for precursor
of SiO thin reaction
films, using products.
a The performance of(BDEAS)
bis-diethylaminosilane our
2
simulation scheme is demonstrated for PEALD synthesis of SiO2 thin films, using a
precursor with the structural formula H2 Si(NEt 2 )2 , Et = C2 H5 . We show that the resulting structure of
bis-diethylaminosilane (BDEAS) precursor with the structural formula H2Si(NEt2)2, Et = C2H5. We
the SiO2 coating strongly depends on the occupation probabilities for precursor products. Densification
show that the resulting structure of the SiO2 coating strongly depends on the occupation
of the SiO2 film and reduction of –OH impurities are observed for increasing occupation probability of
probabilities for precursor products. Densification of the SiO2 film and reduction of –OH impurities
doubly arecoordinated
observed for surface
increasing sites.
occupation probability of doubly coordinated surface sites.

Figure 1. 1.Schematic
Figure representation
Schematic representation of aofsurface
a surface
usingusing plasma-enhanced
plasma-enhanced atomic
atomic layer layer (PEALD).
deposition deposition
(PEALD).
2. Method and Implementation
2. Method and Implementation
2.1. PEALD Simulation Procedure
2.1. PEALD Simulation Procedure
Figure 2 shows the implemented protocol for simulation of the first PEALD deposition
Figure 2 shows
half-cycle. the implemented
In contrast to the originalprotocol
scheme of forHu
simulation
et al. [15], of the first
it used PEALD
the MC methoddeposition
to simulatehalf-cycle.
the
In contrast to the original
full precursor scheme
deposition of Hu
process. et al.
In the [15],
first stepitofused
eachthe MC method
precursor to simulate
deposition thethe
half-cycle, fullsurface
precursor
–OH groups
deposition were
process. Inenumerated. Then,
the first step during
of each each MCdeposition
precursor step, a free surface –OHthe
half-cycle, group was randomly
surface –OH groups
selected. The surface was scanned for neighboring free –OH groups
were enumerated. Then, during each MC step, a free surface –OH group was randomly and, depending on their number
selected.
and precursor
The surface type, a MC
was scanned formove was performed
neighboring that creates
free –OH groups one of possible
and, precursor
depending products.
on their numberThis and
precursor type, a MC move was performed that creates one of possible precursor products. This part
Materials 2019, 12, 2605 4 of 11

is also different than in the original scheme of Hu et al. [15], where direct deposition of the product
of the second half-cycle (i.e., –Al(OH)2 ) was performed. Steric hindrance and overlap restrictions
at the surface corresponding to the real precursor deposition step were accounted for only in an
approximate way, by defining an exclusion zone around each successfully deposited –Al(OH)2 group.
Materials 2019, 12, 2605 4 of 11
In contrast, our approach employs a real, atomistic model of the precursor. After each attachment step,
the structure wasdifferent
part is also relaxed, than
eitherinbythelocal structure
original scheme optimization
of Hu et al. or bywhere
[15], short simulated annealing
direct deposition of theusing
product
molecular of the second
dynamics (MD). The half-cycle
energy(i.e., –Al(OH)
of the 2) was performed.
chemisorption Steric hindrance
was calculated and usedand overlap the
to calculate
restrictions
acceptance at the
criterion surface corresponding
employing to the real precursorrule
the usual Metropolis–Hastings deposition step
[32]. The were accounted
procedure for
was repeated
only in an approximate way, by defining an exclusion zone around
until all enumerated surface –OH groups were either reacted or excluded from MC moves due each successfully deposited – to
Al(OH)2 group. In contrast, our approach employs a real, atomistic model of the precursor. After
steric hindrance and overlap with chemisorbed precursor molecules. Alternatively, a predefined
each attachment step, the structure was relaxed, either by local structure optimization or by short
concentration of surface –OH groups can be left unreacted. The final precursor monolayer structure
simulated annealing using molecular dynamics (MD). The energy of the chemisorption was
was relaxed and equilibrated using MD. The second ALD half-cycle was modeled by simply replacing
calculated and used to calculate the acceptance criterion employing the usual Metropolis–Hastings
all organic surface groups with
rule [32]. The procedure was–OH groups,
repeated untilfollowed by relaxation
all enumerated surface –OHand equilibration
groups were eitherusingreacted
MD. This
corresponds to fullfrom
or excluded conversion
MC moves of thedueproducts
to stericofhindrance
the precursorand pulse
overlap to with
surface hydroxyl species.
chemisorbed precursorSuch
an approach
molecules.is particularly
Alternatively, well suited forconcentration
a predefined simulating ALD growth
of surface –OHof systems
groups can with
be primary reactions
left unreacted.
that have relatively low activation energies [14]. It has been shown that, unlike thermal ALD,ALD
The final precursor monolayer structure was relaxed and equilibrated using MD. The second PEALD
can behalf-cycle
efficiently was modeled even
performed by simply replacing
at room all organic
temperature surface
[6,7], groups
justifying thewith –OH groups,
assumption of lowfollowed
activation
by relaxation
energies. It has alsoand equilibration
been shown byusing MD. Thisthat
calculations corresponds to full conversion
oxygen plasma of the products
can quantitatively oxidize of organic
the
precursor pulse to surface hydroxyl species. Such an approach is particularly well suited for
surface precursor species [17].
simulating ALD growth of systems with primary reactions that have relatively low activation
The selection of created chemisorbed precursor products depends on predefined occupation
energies [14]. It has been shown that, unlike thermal ALD, PEALD can be efficiently performed even
probabilities
at roomwhich remain[6,7],
temperature constant during
justifying thethe simulatedofthin
assumption lowfilm growth.energies.
activation For example,
It has considering
also been a
precursor yielding two
shown by calculations possible chemisorbed precursor products P1 or P2 requires the corresponding
 that oxygen plasma can quantitatively oxidize organic surface precursor
occupation
speciesprobabilities
[17]. pP1 , pP2 ∈ [0, 1] , with
The selection of created chemisorbed precursor products depends on predefined occupation
probabilities which remain constant during pP1 + pthe
P2 =simulated
1 thin film growth. For example, (1)
considering a precursor yielding two possible chemisorbed precursor products P1 or P2 requires the
This approach occupation
corresponding can be easily generalized
probabilities {𝑝𝐏𝟏to
, 𝑝a
𝐏𝟐set of N products
∈ [0,1]}, with {P1, P2, . . . , PN} with occupation
probabilities pP1 , pP2 , . . . , pPN ∈ [0, 1] fulfilling

𝑝𝐏𝟏 +the
𝑝𝐏𝟐condition
=1 (1)
This approach can be easily generalized to a set of 𝑁 products {P1, P2, …, PN} with occupation
probabilities {𝑝p , 𝑝+
𝐏𝟏P1
pP2, 𝑝+
𝐏𝟐 , …
. . . + pPNfulfilling
𝐏𝑵 ∈ [0,1]}
=1 the condition (2)
𝑝𝐏𝟏 + 𝑝𝐏𝟐 + ⋯ + 𝑝𝐏𝑵 = 1 (2)

Figure 2. Monte
Figure Carlo
2. Monte (MC)
Carlo simulation
(MC) simulationprotocol
protocol for growthofofPEALD
for growth PEALD thin
thin films.
films.

2.2. Implementation
Materials 2019, 12, 2605 5 of 11

2.2. Implementation
The implementation of our PEALD simulation protocol used the Python programming language.
Materials 2019, 12, 2605 5 of 11
It was designed in a modular way employing the Atomic Simulation Environment (ASE) [33], which
was usedThe for implementation
storing and manipulating
of our PEALD structure models
simulation as well
protocol as the
used for Python
the reading and writing
programming
of alllanguage.
necessary input
It was and output
designed files. way
in a modular Structure relaxation
employing and Simulation
the Atomic simulatedEnvironment
annealing steps
(ASE)were
[33], whichemploying
accomplished was used for storing
the andUtility
General manipulating
Latticestructure
Programmodels
(GULP) as well
[34].asThe
for Python
the reading and ASE
library
writing of all necessary input and output files. Structure relaxation and simulated annealing
provided a built-in GULP interface which enabled single-energy calculations, structure optimizations, steps
and MDweresimulations.
accomplished employing the General Utility Lattice Program (GULP) [34]. The Python library
ASE provided a built-in GULP interface which enabled single-energy calculations, structure
optimizations,
3. Test and MD simulations.
Application—PEALD of SiO2
As a test,
3. Test the PEALD simulation
Application—PEALD of SiOprotocol
2 was applied to investigate the synthesis of SiO2 thin
films using a bis-diethylaminosilane (BDEAS) precursor with the structural formula H Si(NEt ) ,
As a test, the PEALD simulation protocol was applied to investigate the synthesis of SiO22 thin 2 2
Et = C 2 H5using
films . BDEAS belongs to the class of
a bis-diethylaminosilane aminosilanes
(BDEAS) widely
precursor used
with the in ALD,formula
structural and it has been successfully
H2Si(NEt 2)2, Et =
applied
C2Hfor improving
5. BDEAS belongssurface characteristics
to the class in PEALD
of aminosilanes of SiO
widely used [10,35].
in 2ALD, It has
and it hasbeen
beensuccessfully
shown in [17]
that Si precursors
applied with amino
for improving ligands
surface can dramatically
characteristics in PEALD oflower
SiO2 the activation
[10,35]. energies
It has been shownfor reactions
in [17] that of
aminosilane with surface hydroxyl groups, making BDEAS-based PEALD a very good case for of
Si precursors with amino ligands can dramatically lower the activation energies for reactions testing
aminosilaneprotocol.
our simulation with surface hydroxyl groups, making BDEAS-based PEALD a very good case for
testing our simulation protocol.
3.1. Computational Details
3.1. Computational Details
3.1.1. Model Systems
3.1.1. Model Systems
Hydroxylated SiO2 substrate and deposited precursor products are modeled using two types
Hydroxylated SiO2 substrate and deposited precursor products are modeled using two types of
of cluster models of two neighboring –OH groups and a two-dimensional (2D) periodic surface.
cluster models of two neighboring –OH groups and a two-dimensional (2D) periodic surface. Figure
Figure 3a,b shows both cluster models. The smaller, flexible cluster model has the composition Si O H ,
3a,b shows both cluster models. The smaller, flexible cluster model has the composition Si2O7H26, 7 6
whilewhile
the bigger andand
the bigger more rigid
more cage
rigid cagecluster
clustermodel
modelhashas the
the composition
composition SiSi
8O
OH148.HThe
814 8 . The 2D periodic
2D periodic
substrate model shown in Figure 3c is a hydroxylated (0001) surface containing
substrate model shown in Figure 3c is a hydroxylated -quartz (0001) surface containing 216 atoms,
α-quartz 216 atoms,
48H48 and lattice parameters 𝑎 = 14.778 Å , 𝑏 = 17.077 Å =
= = γ
with the unit cell composition Si O H and lattice parameters a 14.778 Å, b 17.077 Å, ◦.
with the unit cell composition
48 Si 48O120
120 , 𝛾90.081
=
90.081°.

Figure
Figure 3. (a,b)
3. (a,b) TwoTwo cluster
cluster models
models of two
of two neighboring
neighboring –OH
–OH groupsonona ahydroxylated
groups hydroxylatedSiO
SiO22 surface.
surface. The
The shaded lower part indicates the “surface” part of the clusters. (c) 2D periodic model
shaded lower part indicates the “surface” part of the clusters. (c) 2D periodic model of a hydroxylated of a
hydroxylated -quartz
α-quartz (0001) surface. (0001) surface.

3.1.2.3.1.2. Methods
Methods
All calculations
All DFT DFT calculations employed
employed our implementation
our implementation of Kohn–Sham
of Kohn–Sham DFTDFT for molecular
for molecular andand
periodic
periodic systems [36–40] within the TURBOMOLE program package [41,42]. The Perdew–Burke–
systems [36–40] within the TURBOMOLE program package [41,42]. The Perdew–Burke–Ernzerhof
Ernzerhof (PBE) exchange–correlation functional [43], triple zeta valence plus polarization
(PBE) exchange–correlation functional [43], triple zeta valence plus polarization (def2-TZVP) [44] basis
(def2-TZVP) [44] basis sets, and Grimme dispersion correction (DFT-D3) [45,46] were used. MD and
sets, and
MC Grimme dispersion
simulations employedcorrection (DFT-D3)
the ReaxFF [45,46]
reactive force were
field [27]used.
using MD and MC simulations
GULP-provided combined employed
force
the ReaxFF reactive force field
field parameters [27,47,48]. [27] using GULP-provided combined force field parameters [27,47,48].

3.2. Surface Reactions

3.2. Surface andand
Reactions Plasma Pulse
Plasma PulseModel
Model
Figure 4 shows
Figure thethe
4 shows possible elementary
possible elementarysurface reactionsfor
surface reactions forPEALD
PEALD synthesis
synthesis of 2SiO
of SiO 2 thin
thin filmsfilms
usingusing a BDEAS
a BDEAS precursor.
precursor. Thehalf-cycle
The first first half-cycle startsthewith
starts with the deposition
deposition of BDEASof onBDEAS on the
the hydroxylated
SiO2 hydroxylated SiOreaction
surface and its 2 surfacewith
and its reaction
–OH withThere
groups. –OH groups.
are twoThere are two
possible possible(1reactions
reactions and 2a in (1 Figure
and 4)
2a in Figure 4) that yield the organic precursor bonded to one or two surface oxygen atoms, denoted
Materials 2019, 12, 2605 6 of 11

Materials
that 2019,
yield the12,organic
2605 precursor bonded to one or two surface oxygen atoms, denoted as P1 and 6 ofP2,
11

respectively. Reaction 2b converts P1 to P2. After the plasma pulse, P1 and P2 lead to a surface Si atom
as P1 and P2, respectively. Reaction 2b converts P1 to P2. After the plasma pulse, P1 and P2 lead to a
with three or two –OH groups, denoted as S1 and S2, respectively. Alternatively, S1 can be converted
surface Si atom with three or two –OH groups, denoted as S1 and S2, respectively. Alternatively, S1
to site S2 by reaction 3, which releases H2 O. We have concentrated on reactions 1, 2a, 2b, and 3, since
can be converted to site S2 by reaction 3, which releases H2O. We have concentrated on reactions 1,
calculations indicate that oxidation of the organic precursor by oxygen radical species does not alter
2a, 2b, and 3, since calculations indicate that oxidation of the organic precursor by oxygen radical
bonds between the precursor Si atom and the surface [17].
species does not alter bonds between the precursor Si atom and the surface [17].

Figure 4. Elementary surface reactions during PEALD deposition of SiO2 film using a
Figure 4. Elementary surface reactions during PEALD deposition of SiO2 film using a
bis-diethylaminosilane (BDEAS) precursor. P1 and P2 are products of the precursor deposition.
bis-diethylaminosilane (BDEAS) precursor. P1 and P2 are products of the precursor deposition. S1
S1 and S2 are products of the plasma pulse.
and S2 are products of the plasma pulse.
3.3. Reaction Energies
3.3. Reaction Energies
As the first step, we investigated the energetics of the elementary reactions using DFT and the
Asmodels
cluster the firstofstep,two we investigated
neighboring OH the energetics
groups of the elementary
on a hydroxylated reactions
SiO2 surface, usinginDFT
shown and3a,b.
Figure the
cluster
The aimmodels of twostudies
of the cluster neighboring
was toOH obtaingroups on ainto
insights hydroxylated
the intrinsicSiO 2 surface,
energetics ofshown in Figure
the reactions 3a,b.
without
Thespatial
the aim ofand the steric
clusterconstraints
studies was to obtain
present insights
at silica into All
surfaces. the three
intrinsic energetics
reactions (1, 2a,ofand
the 2b)
reactions
of the
without the
precursor spatial
with the and steric
cluster constraints
model in Figure present
3a areatexothermic,
silica surfaces.
with Allreaction
three reactions
energies(1, ∆E2a, and
r of 2b)
−58.3,
of the precursor
−64.3, −6.0 kJ/mol, with the cluster Reaction
respectively. model in 3, Figure 3a are S1
converting exothermic, with reaction
to S2, is slightly energieswith
endothermic, ∆𝐸r anof
∆Er of 37.5 kJ/mol. For the bigger cluster model in Figure 3b, the calculated reaction energies ∆Er of
−58.3, −64.3, −6.0 kJ/mol, respectively. Reaction 3, converting S1 to S2, is slightly endothermic, with
an ∆𝐸r of1,37.5
reactions 2a, kJ/mol.
2b, andFor the–67.0,
3 are bigger –30.5, +36.5,
cluster model andin+63.0
Figure 3b, therespectively.
kJ/mol, calculated reaction step, ∆𝐸
energies
In the next wer
of reactions 1,
investigated the2a,energetics
2b, and 3of are
the–67.0,
same–30.5, +36.5,
reactions onand +63.0 kJ/mol,α-quartz
a hydroxylated respectively.
(0001)Insurface
the next step, 3c).
(Figure we
the energetics of the same reactions on a hydroxylated -quartz
The calculated reaction energies ∆Er were −63.0, −18.4, +44.6, and +60.9 kJ/mol for reactions 1, 2a,
investigated (0001) surface (Figure
3c). and
2b, The 3,calculated
respectively. reaction value for∆𝐸
Theenergies r were −63.0,
exothermic −18.4, 1,
reaction +44.6, and +60.9
involving onlykJ/mol for reactions
one surface OH group1, 2a,
2b, and
and 3, respectively.
yielding The value
site P1, is very similar forforexothermic reactionshown
all three models 1, involving
in Figure only3.one surfaceenergies
Reaction OH group for
and yielding site P1, is very similar for all three models shown in Figure
reaction 2a increase with increasing rigidity of the structure, while remaining exothermic. In contrast, 3. Reaction energies for
reaction 2b
reaction 2a is
increase
slightlywith increasing
exothermic rigidity
for the small, offlexible
the structure,
cluster while
model,remaining
but becomes exothermic.
endothermic In contrast,
for the
reaction 2b is slightly exothermic for the small, flexible cluster model,
structurally rigid models of the bigger cluster and surface. Reaction 3 is endothermic for all models,but becomes endothermic for
the structurally rigid models of the bigger cluster and surface. Reaction
but reaction energies increase with increasing rigidity of the structure. Thus, reactions 1 and 2a are 3 is endothermic for all
models,energetically
always but reaction energies
favorable. increase with increasing
In contrast, reactions 2b rigidity
and 3,oftransforming
the structure.P1 Thus,
to P2reactions
and S1 1toand S2,
2a are always energetically favorable. In contrast, reactions 2b and 3, transforming
respectively, are energetically unfavorable in a rigid atomic environment. This can be attributed to P1 to P2 and S1 to
S2, respectively,
constraints of theare energetically
surface on the one unfavorable
hand, sinceinbringing
a rigid atomic environment.
two surface oxygen atomsThis can be enough
close attributed to
to constraints
create sites P2 of andthe S2surface on thewith
is connected one hand,
energysince bringing
penalty. On thetwo surface
other hand,oxygen atoms closethe
this demonstrates enough
high
to createof
stability sites
theP2 Si–Oandbonds,
S2 is connected
especially with in theenergy
case ofpenalty.
S1. On the other hand, this demonstrates the
high stability of the Si–O bonds, especially in the case of S1.

3.4. PEALD Simulations

The cluster and periodic calculations demonstrate that the actual amount of precursor
deposition products P1 and P2 as well as the plasma products S1 and S2 formed at each PEALD
cycle will be influenced by the actual surface structure and the availability of neighboring –OH
Materials 2019, 12, 2605 7 of 11

3.4. PEALD Simulations

Materials 2019, 12, 2605 7 of 11
The cluster and periodic calculations demonstrate that the actual amount of precursor deposition
products P1 and P2 as well as the plasma products S1 and S2 formed at each PEALD cycle will be
groups. In addition, temperature, pressure, and concentration of precursor species, and plasma
influenced by the actual surface structure and the availability of neighboring –OH groups. In addition,
energy are expected to play an important role. Accurate simulations including all these parameters
temperature, pressure, and concentration of precursor species, and plasma energy are expected to
would be a very complex task. As described earlier, our scheme takes the process parameters into
play an important role. Accurate simulations including all these parameters would be a very complex
account in a simplified way, employing predefined occupation probabilities at each deposition MC
task. As described earlier, our scheme takes the process parameters into account in a simplified way,
step. The probabilities for the creation of P1 (precursor bonded to one surface O atom) and P2
employing predefined occupation probabilities at each deposition MC step. The probabilities for the
(precursor bonded to two surface O atoms) are 𝑝𝐏𝟏 and 𝑝𝐏𝟐 = 1 − 𝑝𝐏𝟏 , respectively. The value
creation of P1 (precursor bonded to one surface O atom) and P2 (precursor bonded to two surface
𝑝𝐏𝟐 = 1 means that each Monte Carlo step attempts to create P2, and P1 is created only if P2 cannot
O atoms) are pP1 and pP2 = 1 − pP1 , respectively. The value pP2 = 1 means that each Monte Carlo
be formed due to missing neighboring surface –OH groups or steric constraints. In contrast, 𝑝𝐏𝟐 = 0
step attempts to create P2, and P1 is created only if P2 cannot be formed due to missing neighboring
means that no P2 is directly created during MC moves. As shown in Figure 5, these probabilities
surface –OH groups or steric constraints. In contrast, pP2 = 0 means that no P2 is directly created
have significant impacts on the structures of PEALD layers. The simulated structures of SiO2
during MC moves. As shown in Figure 5, these probabilities have significant impacts on the structures
deposited for probabilities 𝑝𝐏𝟐 = 0 (𝑝𝐏𝟏 = 1), 𝑝𝐏𝟐 = 0.5 (𝑝𝐏𝟏 = 0.5), and 𝑝𝐏𝟐 = 1 (𝑝𝐏𝟏 = 0) clearly
of PEALD layers. The simulated structures of SiO2 deposited for probabilities pP2 = 0 (pP1 = 1),
demonstrate the densification of the growing SiO2 layer with increasing 𝑝𝐏𝟐 . For 𝑝𝐏𝟐 = 0, each
pP2 = 0.5 (pP1 = 0.5), and pP2 = 1 (pP1 = 0) clearly demonstrate the densification of the growing SiO2
precursor molecule is allowed to react with only one surface –OH group, yielding P1. The plasma
layer with increasing pP2 . For pP2 = 0, each precursor molecule is allowed to react with only one surface
pulse converts P1 to S1, with three hydroxyl groups connected to the deposited silicon atom. This
–OH group, yielding P1. The plasma pulse converts P1 to S1, with three hydroxyl groups connected to
increases the number of hydroxyl groups compared to the initial state, thus favoring the creation of
the deposited silicon atom. This increases the number of hydroxyl groups compared to the initial state,
voids, –OH impurities, and amorphous growth of deposited films. In contrast, for 𝑝𝐏𝟐 = 1, after the
thus favoring the creation of voids, –OH impurities, and amorphous growth of deposited films. In
first ALD half-cycle, the maximum possible number of P2 precursor sites is created. The plasma
contrast, for pP2 = 1, after the first ALD half-cycle, the maximum possible number of P2 precursor sites
pulse converts P2 to S2, with two hydroxyl groups connected to the deposited silicon atom. In this
is created. The plasma pulse converts P2 to S2, with two hydroxyl groups connected to the deposited
case, the number of hydroxyl groups remains unchanged after the deposition step compared to the
silicon atom. In this case, the number of hydroxyl groups remains unchanged after the deposition
initial surface state, and, in combination with compact Si–O–Si bridges, it results in a dense film
step compared to the initial surface state, and, in combination with compact Si–O–Si bridges, it results
growth. Figure 6 shows the average growth per cycle (GPC) and mass densities of deposited SiO2
in a dense film growth. Figure 6 shows the average growth per cycle (GPC) and mass densities of
thin films obtained by 30 simulations of seven PEALD cycles for each of 𝑝𝐏𝟐 = 0.0, 0.25, 0.5, 1.0.
deposited SiO2 thin films obtained by 30 simulations of seven PEALD cycles for each of pP2 = 0.0, 0.25,
Virtually identical results were obtained for MD simulations at temperatures ranging from 300 to
0.5, 1.0. Virtually identical results were obtained for MD simulations at temperatures ranging from 300
1000 K at a time scale of up to 10 ps. GPC and mass densities were basically converged after seven
to 1000 K at a time scale of up to 10 ps. GPC and mass densities were basically converged after seven
simulated PEALD cycles. Decreasing GPC and increasing density were observed for increasing 𝑝𝐏𝟐 ,
simulated PEALD cycles. Decreasing GPC and increasing density were observed for increasing pP2 ,
confirming the observed densification of thin films shown in Figure 5. The very low values for GPC
confirming the observed densification of thin films shown in Figure 5. The very low values for GPC
and density in the case of 𝑝𝐏𝟐 = 0 are mainly due to the strong steric effects when only creation of
and density in the case of pP2 = 0 are mainly due to the strong steric effects when only creation of singly
singly coordinated P1 and S1 is allowed and creation of P2 and S2 is completely excluded. This
coordinated P1 and S1 is allowed and creation of P2 and S2 is completely excluded. This confirms
confirms again the key role of doubly coordinated surface sites for the densification of SiO2 PEALD
again the key role of doubly coordinated surface sites for the densification of SiO2 PEALD thin films.
thin films.

Figure 5. Simulated structures of PEALD-deposited SiO2 with different occupation probabilities pP1
Figure 5. Simulated structures of PEALD-deposited SiO2 with different occupation probabilities 𝑝𝐏𝟏
and pP2 .
and 𝑝𝐏𝟐 .
Materials 2019, 12, 2605 8 of 11
Materials 2019, 12, 2605 8 of 11

Figure 6. Simulated (a) growth per cycle and (b) mass density of deposited SiO2 films as a function of
Figure 6. Simulated (a) growth per cycle and (b) mass density of deposited SiO2 films as a function of
the probability pP2 .
the probability 𝑝𝐏𝟐 .
The actual reaction conditions were accounted for only in an approximate way in our occupation
The actual reaction conditions were accounted for only in an approximate way in our
probability model. For instance, possible precursor desorption due to the substrate heating was not
occupation probability model. For instance, possible precursor desorption due to the substrate
directly simulated. Possible influences from precursor temperature and concentration were not directly
heating was not directly simulated. Possible influences from precursor temperature and
simulated, since our MC scheme deposits one precursor molecule after the other, disregarding the
concentration were not directly simulated, since our MC scheme deposits one precursor molecule
actual reaction path. However, we expect that our model can describe plasma energy effects, since the
after the other, disregarding the actual reaction path. However, we expect that our model can
presence of plasma ions in the second half-cycle may affect reactions 2b and 3 (Figure 4), and therefore
describe plasma energy effects, since the presence of plasma ions in the second half-cycle may affect
the occupation probabilities of doubly coordinated products P2 and S2.
reactions 2b and 3 (Figure 4), and therefore the occupation probabilities of doubly coordinated
Our simulation scheme has been tested for BDEAS as a bis-aminosilane precursor containing two
products P2 and S2.
amino ligands. For mono-aminosilane precursors, we expect simulated structures resembling the case
Our simulation scheme has been tested for BDEAS as a bis-aminosilane precursor containing
of pP1 = 1 in Figure 5, but with denser films and less steric hindrance effects, since in this case there are
two amino ligands. For mono-aminosilane precursors, we expect simulated structures resembling
no remaining precursor amino ligands after the first deposition half-cycle. Aminosilane precursors
the case of 𝑝𝐏𝟏 = 1 in Figure 5, but with denser films and less steric hindrance effects, since in this
with three or four amino ligands can lead, in principle, to even higher surface densification, since they
case there are no remaining precursor amino ligands after the first deposition half-cycle.
can increase the number of compact Si–O–Si bridges. However, the higher number of amino ligands
Aminosilane precursors with three or four amino ligands can lead, in principle, to even higher
favors steric hindrance effects, and triply and quadruply occupied surface sites strongly reduce the
surface densification, since they can increase the number of compact Si–O–Si bridges. However, the
availability of surface –OH groups, thus inhibiting the overall film growth. In summary, the size and
higher number of amino ligands favors steric hindrance effects, and triply and quadruply occupied
number of precursor amino ligands are expected to be determining factors for the steric effects and
surface sites strongly reduce the availability of surface –OH groups, thus inhibiting the overall film
availability of –OH groups in our MC simulation scheme.
growth. In summary, the size and number of precursor amino ligands are expected to be
4.determining
Conclusionsfactors for the steric effects and availability of –OH groups in our MC simulation
scheme.
In this study, we presented an atomistic simulation scheme for PEALD that combines the Monte
Carlo deposition algorithm and structure relaxation using molecular dynamics. In contrast to previous
4. Conclusions
implementations, our approach employed a real, atomistic model of the precursor. This allowed
In this study,
us to account wehindrance
for steric presentedand an atomistic simulationatscheme
overlap restrictions for PEALD
the surface that combines
corresponding the
to the real
Monte Carlo
precursor deposition
deposition step.algorithm
In addition,andourstructure
schemerelaxation using
took various molecular
process dynamics.
parameters into In contrast
account in ato
previous implementations, our approach employed a real, atomistic model
simplified way, employing predefined occupation probabilities for precursor products at each Monte of the precursor. This
allowed
Carlo us to account
deposition step. Thefornew
steric hindrance
simulation and overlap
protocol restrictions
was applied at the surface
to investigate the PEALDcorresponding
synthesis ofto
the real precursor deposition step. In addition, our scheme took various process
SiO2 thin films using a bis-diethylaminosilane precursor. Initial analysis of the precursor products parameters into
account in a simplified way, employing predefined occupation probabilities
with hydroxylated SiO2 surfaces employing DFT calculations demonstrated a strong dependence of for precursor products
at each energies
reaction Monte Carloon thedeposition step. The
surface structure, new simulation
in particular protocol was
for the precursor applied
bonded to twotosurface
investigate the
oxygen
PEALD synthesis of SiO 2 thin films using a bis-diethylaminosilane precursor. Initial analysis of the
atoms. PEALD simulations employing our new method revealed that precursor binding to one surface
precursor
oxygen atom products with hydroxylated
favors amorphous SiO2asurfaces
layer growth, employing
large number of –OHDFT calculations
impurities, and the demonstrated
formation ofa
strongIndependence
voids. of reaction
contrast, precursor bindingenergies
to two on the surface
surface structure,
oxygen atoms leadsinto particular
dense SiO2for filmthe precursor
growth and
bonded to two surface oxygen atoms. PEALD simulations employing our new
reduction of –OH impurities. Increasing the probability for the formation of doubly bonded precursor method revealed that
precursor
sites binding
is therefore to one
the key surface
factor oxygen
for the atomoffavors
formation dense amorphous
SiO2 PEALDlayer growth,
thin films with a reduced
large number of –
amounts
OH impurities, and
of voids and –OH impurities.the formation of voids. In contrast, precursor binding to two surface oxygen
atoms leads to dense SiO2 film growth and reduction of –OH impurities. Increasing the probability
for the formation of doubly bonded precursor sites is therefore the key factor for the formation of
dense SiO2 PEALD thin films with reduced amounts of voids and –OH impurities.
Materials 2019, 12, 2605 9 of 11

Author Contributions: Conceptualization, M.B. and M.S.; methodology, M.B. and M.S.; software, M.B. and M.S.;
validation, M.B.; formal analysis, M.B. and M.S.; investigation, M.B. and M.S.; resources, M.S.; data curation,
M.B. and M.S.; writing—original draft preparation, M.B.; writing—review and editing, M.S.; visualization, M.B.;
supervision, M.S.; project administration, M.S.; funding acquisition, M.S.
Funding: This research was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)
in the framework of the Priority Programme 1959 “Manipulation of matter controlled by electric and magnetic
field: Towards novel synthesis and processing routes of inorganic materials” within the project SI 938/8-1.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Ghazaryan, L.; Kley, E.B.; Tünnermann, A.; Szeghalmi, A. Nanoporous SiO2 thin films made by atomic layer
deposition and atomic etching. Nanotechnology 2016, 27, 255603. [CrossRef] [PubMed]
2. Pfeiffer, K.; Schulz, U.; Tünnermann, A.; Szeghalmi, A. Antireflection Coatings for Strongly Curved Glass
Lenses by Atomic Layer Deposition. Coatings 2017, 7, 118. [CrossRef]
3. Pfeiffer, K.; Dewald, W.; Szeghalmi, A. Antireflection coating with consistent near-neutral color on
complex-shaped substrates prepared by ALD. Opt. Lett. 2019, 44, 3270–3273. [CrossRef] [PubMed]
4. Pfeiffer, K.; Ghazaryan, L.; Schulz, U.; Szeghalmi, A. Wide-Angle Broadband Antireflection Coatings Prepared
by Atomic Layer Deposition. ACS Appl. Mater. Interfaces 2019, 11, 21887–21894. [CrossRef] [PubMed]
5. Ghazaryan, L.; Sekman, Y.; Schröder, S.; Mühlig, C.; Stevanovic, I.; Botha, R.; Aghaee, M.; Creatore, M.;
Tünnermann, A.; Szeghalmi, A. On the Properties of Nanoporous SiO2 Films for Single Layer Antireflection
Coating. Adv. Eng. Mater. 2019, 21, 1801229. [CrossRef]
6. Profijt, H.B.; Potts, S.E.; van de Sanden, M.C.M.; Kessels, W.M.M. Plasma-Assisted Atomic Layer Deposition:
Basics, Opportunities, and Challenges. J. Vac. Sci. Technol. A 2011, 29, 050801. [CrossRef]
7. Potts, S.E.; Profijt, H.B.; Roelofs, R.; Kessels, W.M.M. Room-Temperature ALD of Metal Oxide Thin Films by
Energy-Enhanced ALD. Chem. Vap. Depos. 2013, 19, 125–133. [CrossRef]
8. Kim, D.H.; Lee, H.J.; Jeong, H.; Shong, B.; Kim, W.H.; Park, T.J. Thermal atomic layer deposition of
device-quality SiO2 thin films under 100 ◦ C using an aminodisilane precursor. Chem. Mater. 2019. [CrossRef]
9. Lee, Y.S.; Choi, D.W.; Shong, B.; Oh, S.; Park, J.S. Low temperature atomic layer deposition of SiO2 thin films
using di-isopropylaminosilane and ozone. Ceram. Int. 2017, 43, 2095–2099. [CrossRef]
10. Faraz, T.; Knoops, H.C.M.; Verheijen, M.A.; van Helvoirt, C.A.A.; Karwal, S.; Sharma, A.; Beladiya, V.;
Szeghalmi, A.; Hausmann, D.M.; Henri, J.; et al. Tuning Material Properties of Oxides and Nitrides
by Substrate Biasing during Plasma-Enhanced Atomic Layer Deposition on Planar and 3D Substrate
Topographies. ACS Appl. Mater. Interfaces 2018, 10, 13158–13180. [CrossRef]
11. Faraz, T.; Arts, K.; Karwal, S.; Knoops, H.C.M.; Kessels, W.M.M. Energetic ions during plasma-enhanced
atomic layer deposition and their role in tailoring material properties. Plasma Sources Sci. Technol. 2019, 28,
024002. [CrossRef]
12. Ratzsch, S.; Kley, E.B.; Tünnermann, A.; Szeghalmi, A. Influence of the oxygen plasma parameters on the
atomic layer deposition of titanium dioxide. Nanotechnology 2015, 26, 024003. [CrossRef] [PubMed]
13. Elliott, S.D. Atomic-scale simulation of ALD chemistry. Semicond. Sci. Technol. 2012, 27, 074008. [CrossRef]
14. Shirazi, M.; Elliott, S.D. Atomistic Kinetic Monte Carlo Study of Atomic Layer Deposition Derived from
Density Functional Theory. J. Comput. Chem. 2014, 35, 244–259. [CrossRef] [PubMed]
15. Hu, Z.; Shi, J.X.; Heath Turner, C. Molecular dynamics simulation of the Al2 O3 film structure during atomic
layer deposition. Mol. Simulat. 2009, 35, 270–279. [CrossRef]
16. Fang, G.Y.; Xu, L.N.; Ma, J.; Li, A.D. Theoretical Understanding of the Reaction Mechanism of SiO2 Atomic
Layer Deposition. Chem. Mater. 2016, 28, 1247–1255. [CrossRef]
17. Fang, G.Y.; Xu, L.N.; Cao, Y.Q.; Wang, L.G.; Wu, D.; Li, A.D. Self-catalysis by aminosilanes and strong surface
oxidation by O-2 plasma in plasma-enhanced atomic layer deposition of high-quality SiO2 . Chem. Commun.
2015, 51, 1341–1344. [CrossRef] [PubMed]
18. Han, B.; Zhang, Q.; Wu, J.; Han, B.; Karwacki, E.J.; Derecskei, A.; Xiao, M.; Lei, X.; O’Neill, M.L.;
Cheng, H. On the Mechanisms of SiO2 Thin-Film Growth by the Full Atomic Layer Deposition Process
Using Bis(t-butylamino)silane on the Hydroxylated SiO2 (001) Surface. J. Phys. Chem. C 2012, 116, 947–952.
[CrossRef]
Materials 2019, 12, 2605 10 of 11

19. Huang, L.; Han, B.; Han, B.; Derecskei-Kovacs, A.; Xiao, M.; Lei, X.; O’Neill, M.L.; Pearlstein, R.M.;
Chandra, H.; Cheng, H. First-Principles Study of a Full Cycle of Atomic Layer Deposition of SiO2 Thin Films
with Di(sec-butylamino)silane and Ozone. J. Phys. Chem. C 2013, 117, 19454–19463.
20. Baek, S.B.; Kim, D.H.; Kim, Y.C. Adsorption and surface reaction of bis-diethylaminosilane as a Si precursor
on an OH-terminated Si (0 0 1) surface. Appl. Surf. Sci. 2012, 258, 6341–6344. [CrossRef]
21. Jeong, Y.C.; Baek, S.B.; Kim, D.H.; Kim, J.S.; Kim, Y.C. Initial reaction of silicon precursors with a varying
number of dimethylamino ligands on a hydroxyl-terminated silicon (0 0 1) surface. Appl. Surf. Sci. 2013, 280,
207–211. [CrossRef]
22. Li, J.Y.; Wu, J.P.; Zhou, C.G.; Han, B.; Karwacki, E.J.; Xiao, M.C.; Lei, X.J.; Cheng, H.S. On the Dissociative
Chemisorption of Tris(dimethylamino)silane on Hydroxylated SiO2 (001) Surface. J. Phys. Chem. C. 2009, 113,
9731–9736. [CrossRef]
23. Murray, C.A.; Elliott, S.D.; Hausmann, D.; Henri, J.; LaVoie, A. Effect of Reaction Mechanism on Precursor
Exposure Time in Atomic Layer Deposition of Silicon Oxide and Silicon Nitride. ACS Appl. Mater. Interfaces
2014, 6, 10534–10541. [CrossRef] [PubMed]
24. Karakasidis, T.E.; Charitidis, C.A. Multiscale modeling in nanomaterials science. Mater. Sci. Eng. C 2007, 27,
1082–1089. [CrossRef]
25. Dkhissi, A.; Estève, A.; Mastail, C.; Olivier, S.; Mazaleyrat, G.; Jeloaica, L.; Rouhani, M.D. Multiscale Modeling
of the Atomic Layer Deposition of HfO2 Thin Film Grown on Silicon: How to Deal with a Kinetic Monte
Carlo Procedure. J. Chem. Theory Comput. 2008, 4, 1915–1927. [CrossRef]
26. Spagnoli, D.; Gale, J.D. Atomistic theory and simulation of the morphology and structure of ionic nanoparticles.
Nanoscale 2012, 4, 1051–1067. [CrossRef] [PubMed]
27. van Duin, A.C.T.; Dasgupta, S.; Lorant, F.; Goddard, W.A. ReaxFF: A reactive force field for hydrocarbons.
J. Phys. Chem. A 2001, 105, 9396–9409. [CrossRef]
28. Deetz, J.D.; Faller, R. Parallel Optimization of a Reactive Force Field for Polycondensation of Alkoxysilanes.
J. Phys. Chem. B 2014, 118, 10966–10978. [CrossRef] [PubMed]
29. Norman, P.; Schwartzentruber, T.E.; Leverentz, H.; Luo, S.J.; Meana-Paneda, R.; Paukku, Y.; Truhlar, D.G. The
Structure of Silica Surfaces Exposed to Atomic Oxygen. J. Phys. Chem. C. 2013, 117, 9311–9321. [CrossRef]
30. Gubbels-Elzas, A.; Thijsse, B.J. Ionic motion during field-assisted oxidation of aluminium studied by
molecular dynamics simulations. Comput. Mater. Sci. 2014, 90, 196–202. [CrossRef]
31. Hühn, C.; Wondraczek, L.; Sierka, M. Dynamics of ultrathin gold layers on vitreous silica probed by density
functional theory. PCCP 2015, 17, 27488–27495. [CrossRef] [PubMed]
32. Amar, J.G. The Monte Carlo method in science and engineering. Comput. Sci. Eng. 2006, 8, 9–19. [CrossRef]
33. Larsen, A.H.; Mortensen, J.J.; Blomqvist, J.; Castelli, I.E.; Christensen, R.; Dulak, M.; Friis, J.; Groves, M.N.;
Hammer, B.; Hargus, C.; et al. The atomic simulation environment-a Python library for working with atoms.
J. Phys. Condens. Matter 2017, 29, 273002. [CrossRef] [PubMed]
34. Gale, J.D.; Rohl, A.L. The General Utility Lattice Program (GULP). Mol. Simulat. 2003, 29, 291–341. [CrossRef]
35. Kim, H.; Chung, I.; Kim, S.; Shin, S.; Jung, W.; Hwang, R.; Jeong, C.; Hwang, H. Improved film quality of
plasma enhanced atomic layer deposition SiO2 using plasma treatment cycle. J. Vac. Sci. Technol. A 2015, 33,
01A146. [CrossRef]
36. Burow, A.M.; Sierka, M.; Mohamed, F. Resolution of identity approximation for the Coulomb term in
molecular and periodic systems. J. Chem. Phys. 2009, 131, 214101. [CrossRef] [PubMed]
37. Burow, A.M.; Sierka, M. Linear Scaling Hierarchical Integration Scheme for the Exchange-Correlation Term
in Molecular and Periodic Systems. J. Chem. Theory Comput. 2011, 7, 3097–3104. [CrossRef] [PubMed]
38. Lazarski, R.; Burow, A.M.; Sierka, M. Density Functional Theory for Molecular and Periodic Systems Using
Density Fitting and Continuous Fast Multipole Methods. J. Chem. Theory Comput. 2015, 11, 3029–3041.
[CrossRef] [PubMed]
39. Lazarski, R.; Burow, A.M.; Grajciar, L.; Sierka, M. Density Functional Theory for Molecular and Periodic
Systems Using Density Fitting and Continuous Fast Multipole Method: Analytical Gradients. J. Comput.
Chem. 2016, 37, 2518–2526. [CrossRef]
40. Becker, M.; Sierka, M. Density Functional Theory for Molecular and Periodic Systems Using Density Fitting
and Continuous Fast Multipole Method: Stress Tensor. J. Comput. Chem. 2019. [CrossRef]
41. Ahlrichs, R.; Bär, M.; Häser, M.; Horn, H.; Kölmel, C. Electronic-Structure Calculations on Workstation
Computers—The Program System Turbomole. Chem. Phys. Lett. 1989, 162, 165–169. [CrossRef]
Materials 2019, 12, 2605 11 of 11

42. Furche, F.; Ahlrichs, R.; Hättig, C.; Klopper, W.; Sierka, M.; Weigend, F. Turbomole. WIREs Comput. Mol. Sci.
2014, 4, 91–100. [CrossRef]
43. Perdew, J.P.; Burke, K.; Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett.
1996, 77, 3865–3868. [CrossRef] [PubMed]
44. Weigend, F.; Ahlrichs, R. Balanced basis sets of split valence, triple zeta valence and quadruple zeta valence
quality for H to Rn: Design and assessment of accuracy. PCCP 2005, 7, 3297–3305. [CrossRef] [PubMed]
45. Grimme, S.; Antony, J.; Ehrlich, S.; Krieg, H. A consistent and accurate ab initio parametrization of density
functional dispersion correction (DFT-D) for the 94 elements H-Pu. J. Chem. Phys. 2010, 132, 154104.
[CrossRef] [PubMed]
46. Grimme, S.; Ehrlich, S.; Goerigk, L. Effect of the Damping Function in Dispersion Corrected Density
Functional Theory. J. Comput. Chem. 2011, 32, 1456–1465. [CrossRef]
47. Abolfath, R.M.; van Duin, A.C.T.; Brabec, T. Reactive Molecular Dynamics Study on the First Steps of DNA
Damage by Free Hydroxyl Radicals. J. Phys. Chem. A 2011, 115, 11045–11049. [CrossRef]
48. Rahaman, O.; van Duin, A.C.T.; Goddard, W.A.; Doren, D.J. Development of a ReaxFF Reactive Force Field
for Glycine and Application to Solvent Effect and Tautomerization. J. Phys. Chem. B 2011, 115, 249–261.
[CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).