Sie sind auf Seite 1von 42

CHAPTER 81

Atomistic Simulations of Dislocations in FCC


Metallic Nanocrystalline Materials
H. VAN SWYGENHOVEN, P.M. DERLET
Materials Science & Simulation, NUM-ASQ, Paul Scherrer Institute, CH-5232 Villigen PSI, Switzerland

Dedicated to the memory of F.R.N. Nabarro, whos scientific excellence and dedication
has been an inspiration to us

2008 Elsevier B.V. All rights reserved Dislocations in Solids


1572-4859, DOI: 10.1016/S1572-4859(07)00001-0 Edited by J. P. Hirth
Contents
1. Bulk nanocrystalline plasticity 3
2. Introduction to atomistic simulation 4
2.1. Methodology 4
2.2. Timescale restrictions 7
2.3. Lengthscale restrictions 9
2.4. Empirical potentials for FCC metals 10
2.5. Nanocrystalline sample construction 12
2.5.1. Geometrical construction 12
2.5.2. Cluster compaction 14
2.5.3. Quenching from the melt 14
2.6. Atomic visualization 14
2.6.1. Local energy 15
2.6.2. Atomic coordination 16
2.6.3. Common neighbor analysis 17
2.6.4. Positional disorder 18
2.6.5. Local stress and strain 18
2.6.6. Centro-symmetric parameter 19
2.6.7. Free volume 19
3. Atomistic simulations of deformation in bulk 3D nanocrystalline metals 19
3.1. The role of the generalized stacking fault energy curve 20
3.2. Case study one: dislocations in a defect free grain 24
3.3. Case study two: a grain containing twin defects 28
4. Experimental-computational synergy tools 31
4.1. X-ray diffraction 31
4.2. Phonons 34
5. Discussion and concluding remarks 36
Acknowledgements 39
References 40
1. Bulk nanocrystalline plasticity
It is not an easy task to write a chapter on dislocations in FCC nanocrystalline metals for
a book that is a tribute to F.R.N. Nabarro and edited by J.P. Hirth, who both are among
the world experts in dislocation theory. Our current understanding of dislocations and their
role in the mechanical behavior of metals still relies on the ground work performed in
the 5060s by F.R.N. Nabarro, J.P. Hirth, J. Weertman and J.R. Weertman, A.H. Cottrell,
J. Friedel and W.T. Read as is testified by the number of comprehensive books [16], still
frequently used as reference in undergraduate and graduate courses on physical and me-
chanical properties of crystalline solids. What is different to those years is that contempo-
rary times possess new techniques such as atomistic and mesoscopic computer simulation
techniques allowing the investigation of the dynamical details of microstructures and their
interactions with dislocations. For instance, mesoscopic simulation methods such as dislo-
cation dynamics [79], a front tracking method in which dislocations are discretized into
segments allowing for instance the simulation and visualization of the dynamics of Frank
Read dislocation sources and their pile-up in a finite volume [10], are now used to study the
dynamics of hardening mechanisms. Given a reasonable accurate description of the atomic
forces, atomistic simulations such as molecular dynamics (MD), allow the modeling of the
atomistic details of the interactions of dislocation with a microstructure with atomic scale
resolution. MD has not only revealed unprecedented detail of dislocation nucleation, prop-
agation, multiplication and the interaction mechanism with its microstructural environment
often visualizing and detailing the mechanisms that were introduced in the basic works
published in the 5060s but has also emphasized the complexity of dislocation theories.
For example, atomistic simulations are used to study the many possible interaction mech-
anisms between dislocations and interfaces, a field that has proven its importance in the
understanding of mechanical behavior of nanocrystalline metals.
Nanocrystalline (nc) metals are by definition polycrystalline structures with a mean
grain size below 100 nm and therefore their microstructure contains a significant volume
fraction of interfacial regions separated by nearly-perfect crystals [11]. Compared to its
coarse-grain counterpart, the mechanical behavior of a fully-dense nc metal is character-
ized by a significantly enhanced yield stress and a limited tensile elongation [12]. From
the early beginning of experimental investigations in the mechanical behavior of nc met-
als, it has been proposed that the small grain sizes limit the conventional operation of
dislocation sources and estimations of strength were made based on simple FrankRead
sources [11,13]. This simple picture of dislocation nucleation, together with the early as-
sumption that grain boundaries (GBs) in nc metals are highly disordered to the extent that
the material could be considered as a two phase system where the grain boundaries act as a
glue between perfect grains, had lead for a while to the belief that dislocations were of no
importance to nc metals. However, later, several high resolution transmission electron mi-
croscopy (HR-TEM) studies have revealed that grain boundaries in pure nc metals are not
4 H. Van Swygenhoven and P.M. Derlet Ch. 81

amorphous or glue-like [14,15]. Most TEM studies performed on undeformed and de-
formed nc metals confirm the idea that there is no dislocation network in the grain interiors
as is the case in coarse-grained polycrystals. With the increase in computational power, one
can deform three-dimensional (3D) computational nc samples containing enough grains in
order to mimic an nc metal. These simulations have revealed that the deformation behavior
is still to a major extent governed by dislocations: however the dislocations are not nucle-
ated via classical FrankRead sources situated within the grain interior, but are nucleated
at GB ledges, with dislocations traveling through the grain and finally being absorbed in
opposite GBs. The merit of the atomistic simulations of nc GB networks lies in the revela-
tion of the details of such dislocation mechanisms. However it has to be mentioned that the
concept of dislocation nucleation from GBs was already introduced by Hirth in 1972 [16,
17] and experimentally observed a few years later [11,18] well before the experimental era
of nc metals.
In this chapter the dislocation mechanisms suggested by MD during plastic deformation
studies of nc FCC metals are reviewed and the existence of these mechanisms is discussed
in terms of experimental observations. Although MD is a very well known technique and
several reference books exist, the chapter starts with a basic introduction to MD. This ap-
proach is justified by the fact that MD contains some subtle and less subtle caveats that
especially in the research of nc metals has lead to some hasty interpretations resulting in
questionable conclusions concerning the deformation mechanism in nc metals. Therefore,
in order to be able to fully appreciate the MD technique, Section 2 is devoted to the method-
ology, sample construction methods and sample analyzing methods. The main dislocation
mechanism resulting from a vast number of simulations performed in the last couple of
years is summarized in Section 3, including a case-study for a defect-free grain and a grain
that contains twins prior to deformation. Section 4 discusses some tools that have been de-
veloped in order to be able to verify the synergy between MD simulations and experimental
measurable properties. Finally the chapter concludes with a section where simulation re-
sults are evaluated within the framework of experimental results, together with an outlook
mentioning some issues to be resolved. This chapter does not aim to be a comprehensive
review of all data published on the deformation mechanisms in nc metals, but rather to
provide the authors opinion on the current state-of-the-art.

2. Introduction to atomistic simulation


2.1. Methodology

The MD technique involves simulating the classical motion of every atom within a cho-
sen atomic configuration, and has been applied to systems traditionally ranging from dilute
gases, liquids and condensed matter systems such as bulk metals and semi-conductors con-
taining microstructures including GBs and layered geometries. MD has also been applied
to systems characterized by finite sized length scales such as nanowires, clusters and free
standing components of complex geometry. The enduring attraction of the technique, given
a reasonably accurate description of the atomic forces, is that by starting from the atom, one
can, in principle, model the complex stress signature and defect energetics/dynamics of all
2.1 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 5

material structures. As with all methods, MD contains subtle and not so subtle caveats that
restrict its applicability and accuracy to provide experimentally realizable material parame-
ters. This is particularly the case when modeling mechanical properties of materials where
the traditional experimental time and length scales are significantly larger than those typi-
cally seen in MD simulations. In the present section we introduce the basic computational
algorithm of MD and then address its important limitations.
Given the availability of a multi-atom energy function:
V (r1 , . . . , rN ),
where the r1 , . . . , rN constitute the 3N coordinates of the atomic configuration, the ba-
sic numerical technique of MD will involve solving the associated Newtons equation of
motion:
mi r i (t) = F (r1 , . . . , rN ) = i V (r1 , . . . , rN ).
The numerical approach of MD generally involves approximating the double derivative
associated with the atoms acceleration via a simple finite difference representation, giving
for example
[ri (t + 2t) 2ri (t + t) + ri (t)]  
mi = i V r1 (t), . . . , rN (t) .

t 2
Thus if the configuration at times t + t and t are known, then the configuration at the
latter time t + 2t can be easily calculated. The above equation represents the simplest
finite difference operator and in practice is not numerically stable for reasonable values of
t + t in terms of total energy conservation. As a result more stable integrators have been
developed such as the Verlet and Gear predictor/corrector integrators resulting in time-steps
of the order of a femto-second for a condensed matter system [19].
The actual implementation of the numerical solution of 3N equations of motion requires
a starting configuration the initial boundary condition. For the predictor/corrector algo-
rithms, this generally involves defining the initial spatial location of each atom and its
velocity. More generally for a condensed matter system this entails geometrically build-
ing the structure of interest through for example assuming a particular crystallography and
then choosing the shape or morphology of these crystalline regions. Velocities are usually
chosen randomly such that the total kinetic energy defines the required temperature of the
system:
3N 1
kb T = mi r i r i .
2 2
i
With the starting configuration defined, the MD equations of motions are iterated to evolve
the system through physical time until the system reaches thermal equilibrium.
All thermodynamic variables can easily be measured and controlled within the frame-
work of equilibrium MD. For example the temperature, T , for a mono-atomic system with
mass m can easily be calculated using the above equation. Using this formula, one can
control
the temperature of the MD system by rescaling the atomic velocities by the factor
Tactual /Tdesired every certain number of MD steps, eventually leading to an equilibrated
system at the desired temperature. There exist more elaborate approaches for the control
6 H. Van Swygenhoven and P.M. Derlet Ch. 81

of temperature through a fictitious damping term, the magnitude and sign of which, is con-
trolled by the difference between the desired temperature and the actual temperature of the
system [19]. Under equilibrium conditions, all such methods are expected to be equivalent.
For a system where the force on a particular atom can be represented by two body terms,
the instantaneous total stress of the system can be calculated via the virial stress that is
given by
 
1  1 1
=

mr r +
F (rij )rij .
 2 i i 2
i j

For atomic configurations that are on average isotropic, the hydrostatic pressure is gener-
ally calculated and is given by one third of the trace of the stress tensor.
There exist two modes of tensile deformation that have been widely used in the atom-
istic simulation of plastic deformation of periodic systems: that of constant strain rate and
constant stress simulations. The most widely used approach is to apply a global stress to
a simulation cell under full three-dimensional (3D) periodicity via the ParrinelloRahman
technique [20]. Within this framework, absolute atomic coordinates are represented via

ri = B si ,
where B is a square matrix of rank 3, and si are reduced dimensionless atomic coordinates
ranging from 0.5 to 0.5. Thus B has units of length and under orthorhombic geometry
conditions, the diagonal components of B are the periodicity lengths. Higher derivatives of
the atomic coordinates are represented in a similar fashion. The dynamical MD variables
are now B and si , all of which follow differential equations, with the driving force of B
being the difference between the applied global stress and the actual global stress of the
simulation cell calculated via the virial theorem.
The usage of such an approach produces strain versus time data, an example of which
is shown in Fig. 1(a). In this example, an applied uniaxial tensile load of 1.5 GPa has
been applied to a nc Al sample with a mean grain size of 10 nm. The MD was performed
at 300 K [2123]. The measured strain rate between 30 and 60 ps of simulation time is
5 107 /s and from detailed analysis of the atomic scale processes the plastic strain
arises from both slip activity within the grains and GB mediated processes such as GB slid-
ing. Previous work has shown that, for this regime when the simulation time is extended
into the nanoseconds, the strain rate can reduce by an order of magnitude, indicating that
the plastic response of the system as shown in Fig. 1(a) is far from being in a dynami-
cal equilibrium [24]. Fig. 1(a) also shows the resulting strain versus time curve when the
applied load is increased to 1.55 GPa where the strain rate has now increased by an or-
der of magnitude due to increased slip activity within the grains. Increasing the stress to
1.6 GPa further (not shown) increases the strain rate by an order of magnitude resulting in
all grains undergoing slip activity with the final result being the eventual destruction of the
sample.
Constant strain rate simulations have the advantage of producing a stressstrain curve.
There are many ways in which such a simulation can be performed. Presently we consider
a procedure that exploits the framework of the ParrinelloRahman technique under an or-
thorhombic geometry for the simulation cell, where a strain rate is imposed along a given
2.2 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 7

(a) (b)
Fig. 1. (a) Strain versus simulation time curve derived from a constant uniaxial tensile load simulation and
(b) stress versus strain curve derived from a constant strain rate simulation.

direction, say in the z direction, via a choice of B zz . The remaining degrees of freedom,
namely Bxx and Byy , then follow the normal ParrinelloRahman equations of motions to
model the lateral Poisson contraction of the simulation cell. Fig. 1(b) displays typical stress
strain data for imposed strain rates covering three orders of magnitude and a strong strain
rate sensitivity of the maximum stress carried by the simulated material is observed. Indeed
at the strain rate of 1 109 /s a large maximum is seen followed by a reduction in the
stress. This maximum has often been used as defining a yield stress [2527], however as
seen in Fig. 1(b), this common feature appears to be an artifact of the high strain rate and is
reduced by 25% as the strain rate drops two orders of magnitude. Given that much of the
published work involves simulations with strain rates between 1 108 /s and 1 109 /s
this is a lucid example of the dangers of deriving quantitative material parameters such as
a yield stress and relating them to experimentally derived corresponding quantities.

2.2. Timescale restrictions

Any instantaneous atomic configuration can be described by its 3N spatial coordinates,


thus its temporal evolution can be visualized as a trajectory in a 3N -dimensional space,
the so-called phase space of the physical system. What is quite typical for equilibrated
condensed matter systems is that the region of phase space that the system occupies over
a period of time is generally quite localized very little might happen during the course
of a simulation. For example in an MD simulation of a perfect FCC lattice at 300 K, once
equilibrated the atoms essentially oscillate about their mean FCC lattice positions. In terms
of energy, the system exists in an energy minimum that generally can be well described
by a multi-dimensional quadratic function the so-called harmonic energy well. If now,
one atom is removed and the equilibration is repeated, a vacancy defect has been created
with the dimensionality of the phase space reducing to 3N 3. Such a system will also
remain for a significant amount of time in one region of the (3N 3)-dimensional phase
8 H. Van Swygenhoven and P.M. Derlet Ch. 81

Fig. 2. Schematic of a rare event, where initially the system oscillates in region A in phase space. At a particular
time, a thermal fluctuation results in the system moving to region B of phase space. The dark curve represents the
(3N 3)-dimensional trajectory and the thicker light curve idealizes the potential energy surface.

space, until at some point, one of the twelve atoms neighboring the vacancy moves to the
vacancy site, thereby shifting the vacancy to a new location. In terms of the (3N 3)-
dimensional phase space, the system now moves to a new region of phase space where it
remains localized in a new harmonic energy well until the next hopping event occurs. This
atomic scale process corresponds to vacancy diffusion at 300 K and can be measured via a
simple lattice hopping model:
1 2
D= d ,
6
where D is the vacancy diffusion coefficient, d is the nearest neighbor FCC lattice distance
and is the rate of the vacancy hopping. The rate, , is what would actually be measured
in the MD simulation.
From statistical mechanics, the rate of a thermally activated process at temperature T
may be written as
 
E
 = 12 exp ,
kb T
where kb is Boltzmanns constant, E is the migration energy, and is the attempt fre-
quency. The factor of 12 counts for all possible atoms that may hop to the vacancy site.
The exponential factor (the so-called Boltzmann factor) represents the probability of a suc-
cessful attempt at a vacancy hop and is the number of times per unit time that the system
attempts to perform such a hop. E may be considered the effective energy barrier the
system must overcome through thermal fluctuations (that are scaled by energy kb T ) to
achieve the actual hop. Fig. 2 details schematically the actual process of a hop in terms of
an idealized trajectory in the (3N 3)-dimensional phase space. The system spends a long
time in region A, oscillating back and forth in a small region of space at a time scale that
defines , until at some point in time one of the neighboring atoms gains enough energy for
it to migrate to the vacant site, resulting in the system moving to region B in phase space.
Since such a process does not occur so often at 300 K, it usually is referred to as a rare
event.
Just how rare such an event is can be estimated via the attempt rate , which will be
typically of the order of the period of oscillation of thermal atomic motion, or equivalently
2.3 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 9

the typical frequency of a phonon in the system. The latter corresponds to 5 THz, and
E for Al is known to be approximately 0.65 eV [28]. Thus at 300 K,  is typically 700/s,
or alternatively, the average time between hops is 1 ms. Given that the timestep per MD
iteration is 1 fs, this constitutes a vacancy hop every 1012 MD iterations. With wall
clock times for MD simulations being typically between 0, 1 and 1 s per MD step such a
simulation becomes practically impossible.
The above is a simple example of the study of a thermally activated process and since
much of what goes on in the condensed matter state can be associated with such thermally
activated processes it is a potent indicator of the time-scale limitations of the MD tech-
nique. A pertinent example of this in the field of nc plastic deformation simulation was
already seen in Fig. 1(b), in which the stress carried by the computer generated nc ma-
terial decreases as function of the applied strain rate by reducing the strain rate slower
atomic scale processes that can further relieve stress become increasingly active resulting
in a reduction of the global simulation stress.
There have been a number of attempts at developing modifications to the traditional MD
technique, which in an autonomous way, promote the simulated atomic system to leave
its local equilibrium configuration. The most well known approach is the so-called hyper-
dynamics in which a bias-potential is introduced when the system is within an harmonic
well, and subsequently turned off as the system undergoes a transition to a new local equi-
librium configuration [29,30]. An underlying problem has been that to determine whether
or not a system is in local equilibrium requires information derived from the second deriv-
atives of the total potential energy (the Hessian), which is a computationally intensive cal-
culation. Other approaches such as temperature accelerated dynamics [31] and the parallel
replica method [32], which rely on an easy way in which to identify a rare event process,
have been applied successfully to wide range of material problems. In a complex atomic
environment, where many atomic processes occur separately and independently such as in
a GB, it remains unclear if such methods can be used effectively and efficiently. Recently,
the parallel-replica method has been applied with some success to the study of stick-slip
behavior of GBs in simple bi-crystal geometries [33]. An alternative approach to the study
of such rare event atomic scale processes is to employ the nudged elastic band (NEB)
technique [34] to determine the minimum energy pathway between a state and end atomic
configuration. This has been used in the study of atomistic crack propagation [35] and more
recently slip transfer across twin boundaries [36].

2.3. Lengthscale restrictions

With the advent of massively parallel computing it now becomes possible to routinely sim-
ulate tens to hundreds of millions of atoms through the spatial decomposition of the atomic
configuration amongst the available processors. Despite this computing advancement, spa-
tial limitations are always present, since even with a billion atoms, the corresponding size
regime remains below a micrometer.
The most common approach to reduce the number of atoms is to impose periodic bound-
ary conditions on the sample, that is, the 3D simulation box is effectively surrounded by
26 exact copies of itself. Thus an atom close to the boundary of the simulation box will be
10 H. Van Swygenhoven and P.M. Derlet Ch. 81

(a) (b)
Fig. 3. (a) Sketch of two-dimensional periodic boundary conditions, where the central atomic configuration is
repeated around borders to simulate an FCC lattice of infinite size. (b) Cross-section of a computer generated nc
sample under full three-dimensional periodic boundary conditions.

close to atoms that are close to the boundary on the opposite side of the simulation box.
Fig. 3(a) provides a schematic 2D example of such a periodic boundary condition geom-
etry. Such a procedure does impose restrictions on the effective bulk microstructure. For
example, when constructing a grain network care must be taken in the shape and position
of grains for the total configuration to be compatible with the imposed periodicity. Such
issues become particularly evident in the construction of GB networks in which extended
connectivity between misorientations of neighboring grains is required, or specific distrib-
utions of grain sizes and grain shape are needed. Fig. 3(b) displays a 2D cross-section of a
3D nc structure in which the simulation cell contains fifteen grains of random orientation.
Under periodic boundary conditions, the effective bulk nc (super cell) structure exhibits a
patterning that is far from experimental reality.

2.4. Empirical potentials for FCC metals

Although the condensed matter state remains fundamentally a quantum system, the differ-
ence in masses between the atoms and the electrons, the latter of which contribute to the
materials bonding properties, allows for a classical force to be defined between the atoms.
This is, in essence, the adiabatic approximation where the atomic and electronic degrees
of freedom can be decoupled from each other, allowing for the electronic degrees to be
integrated out with respect to atomic motion. The precise form of the classical inter-atomic
potential is thus of quantum origin. For closed shell systems, where there is not a strong
electronic contribution to the bonding, a simple pair potential will suffice, which at short
range will be repulsive and in the long range will be attractive. For metallic systems, the
bonding originates from a combination of a screened ionion type interaction described by
a pair term and an electronic band-energy term. For simple, sp valent metals and d-state
2.4 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 11

transition metals, this band term can be well represented by the width of the band of elec-
tronic states, resulting in a term that is equal to the square root of the sum of bond overlap
integrals between neighboring atoms. Such an approach is referred to as the 2nd moment
tight binding method [37,38]; a more general version being the so-called embedded atom
method which has its theoretical origins in density functional theory [39]. For these mod-
els, the total energy is given by

 1

E= F [i ] + V (rij ) ,
2
i j

where F [ ] is the so-called embedding function (a square-root function for 2nd moment
tight binding method), i = j (rij ) is the local electron density at atom i (constructed
from the electron density, (rij ), arising from each surrounding atom j ). For simple pair
potential systems, the embedding function can either be zero or linear in the local electronic
density. This method displays the essential unsaturated nature of the metallic bond, in
which, if one bond is broken, the remaining bonds are strengthened. Generalizations of this
approach have been applied to ferromagnetic bcc Fe in which the magnetism is modeled
explicitly through a GinzburgLandau mean field approach [40,41].
For systems where there is strong hybridization between the electronic states of neigh-
boring atoms, such as covalent systems and partially filled d-state systems, additional
angular terms are required to describe the sensitivity of the band energy to local varia-
tions in atomic environment. For saturated sp 3 bonding systems, the three body Stillinger
Weber [42] and Tersoff [43] potentials have been used successfully for Si. More recently
the bond-order method has been developed to describe in a systematic way the inter-atomic
potential for a general hybridized system, by employment of the higher moments of the dis-
tribution of electronic states [44,45]. An excellent review of empirical potentials is given
in Ref. [46].
The development of such empirical inter-atomic potentials for a given system generally
involves searching for an optimal set of parameters for the chosen analytical interaction
model, with respect to experimental and ab initio calculated material properties. Such a
database of properties is generally restricted to equilibrium atomic configurations of the
system such as lattice constants, cohesive energy, elastic constants, local and extended
defect including vacancies, interstitials and stacking faults for the FCC, BCC and HCP
phases. Performing simulations for systems not contained within in the data base used
to develop a particular potential, thus involves the implicit assumption that the model is
transferable. If the chosen analytical inter-atomic model captures the essential physics of
the materials bonding, then it is not unreasonable to make such an assumption. However
one must always keep in mind that the theoretical models used to derive such empirical
potentials contain many simplifying assumptions and therefore cannot be expected to have
the same accuracy as that is seen in fully quantum mechanical calculations. The application
of such analytical classical models do however allow for the extremely fast and efficient
calculation of the energy and forces within an atomic system, when compared to ab initio
derived energies and force. This results in large systems being able to be modeled involving
millions of atoms when modern parallel computers are used.
12 H. Van Swygenhoven and P.M. Derlet Ch. 81

2.5. Nanocrystalline sample construction

As in experiment, the structural and mechanical properties of computer generated nc mate-


rials can be strongly dependent on the way in which the sample is constructed. For exam-
ple, the issue of whether or not, the metallic nc GB is amorphous or not, can depend on the
sample preparation method: more generally, the extent to which the computer generated nc
state is from equilibrium will depend strongly on the preparation method, in turn affecting
the nc mechanical properties [47].

2.5.1. Geometrical construction


In the modeling of nc systems, samples are often constructed by beginning with an empty
simulation cell with fully 3D periodic boundary conditions, and choosing randomly a num-
ber of seed positions [48]. The number of seed positions is determined from the simulation
cell size and the desired characteristic grain size. From each seed position, an FCC lattice
is constructed geometrically with an orientation chosen either randomly or according to a
texture [49] and/or orientation connectivity requirements [50]. At a point where atoms from
one grain center are closer to the center of another grain, construction is halted. Eventu-
ally construction will cease throughout the entire sample, resulting in a three dimensional
granular structure according to the Voronoi construction [51]. At this stage atom pairs,
each atom originating from a different crystallite, are inspected and where there is a near-
est neighbor distance of less than 0.2 nm, one atom is removed. Molecular statics is then
performed to relax any local high potential energy configuration that may exist, followed
by constant pressure MD at room temperature to further relax and equilibrate the structure.
The constant pressure MD is performed using the ParrinelloRahman Lagrangian with
orthorhombic simulation cell geometry conditions.
When the number of grains in the simulation cell is sufficiently large, such a Voronoi tes-
sellation procedure has the advantage of producing a log-normal grain size distribution [23]
a size distribution that is commonly found in experimental bulk nc materials [52]. For
computational reasons, early 3D simulation work generally employed a small number of
grains and to avoid grain centers that are too close together, the grain centers where chosen
to give a relatively narrow grain size distribution [5355]. Other 3D work has chosen the
grain centers to be at face-centered locations within the simulation cell resulting in iden-
tical rhombic dodecahedron grain shapes [2527,5658]. Such an approach has also been
used in 2D nc simulation work in which the grain centers have been chosen to be at the
vertices of a triangular lattice, resulting in a columnar GB with a given texture along the
columns [5962]. The advantage of such a 2D geometry is the ease with which large grain
sizes can be simulated, whereas the disadvantage is that the reduced dimensionality allows
for only a restricted class of GBs, a restriction in available slip systems, the preclusion of
the fully 3D aspect of dislocation nucleation [47,63]. Indeed, by virtue of the imposed peri-
odicity in the columnar direction, the dislocation cores will interact strongly with their im-
ages, producing immediately a kink-free dislocation of infinite extent along the columnar
direction. Since the interaction between the propagating dislocation with the surrounding
GB environment has recently been identified both experimentally [6466] and computa-
tionally [21] as key elements of the operating slip mechanism, results obtained from 2D
simulations have to be taken with special care.
2.5 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 13

(a) (b)

(c)
Fig. 4. Atomic positions of a triple junction region for the (a) as-prepared, (b) annealed and (c) deformed samples.
For both the annealed and as-prepared samples, migration of the triple junction to a more equilibrium configura-
tion is observed. Figure is taken from Ref. [24].

The Voronoi tessellation procedure produces planar GB and triple junction geometries
that are expected to be energetically unfavorable. Indeed MD simulations performed at
elevated temperatures or under a uniaxial tensile load have revealed that certain triple
junction geometries introduced by the Voronoi construction relax via GB/TJ migration
to a more energetically favorable geometry [24]. Fig. 4 shows an example of TJ in which
the starting configuration (a) is far from that containing the energetically favorable 120
GB plane angles, and which upon performing MD relaxes to a configuration in which
the GB plane angles converge to 120 at the triple junction intersection (b). The limi-
tations of such geometrical Voronoi constructions motivate the usage of tessellation data
derived from 3D grain growth simulations in which an initially geometrical construction is
allowed to evolve. Such modeling techniques involve phase field approaches [67] or ver-
tex tracking techniques [68] producing networks that can be employed to construct more
realistic atomistic nc samples. To date such an approach has not been employed to con-
struct MD samples but offers a realistic alternative to the standard Voronoi constructure
procedure. Procedures have been developed where experimental grain size and texture dis-
tributions can be used as input to develop more realistic GB networks suitable for MD
simulation [69].
TEM analysis of experimental nc structures evidences the presence of many special GBs
including coincidence site lattice (CSL) boundaries, perfect and general = 3 boundaries
or low-angle pure tilt boundaries [70,71] suggesting possible discrepancies between the
14 H. Van Swygenhoven and P.M. Derlet Ch. 81

type of GB structures incorporated in the simulations and those present in experimental


samples. Recent simulation work has demonstrated that by modifying the Voronoi con-
struction to contain clusters of grains with given configurations of coincident site lattice
(CSL) and low-angle tilt boundaries, the simulated bulk deformation properties can be
changed [50]. For example the low-angle cluster samples can accommodate deformation
by rearrangement of their lattice dislocation network, whereas in the case of the CSL sam-
ple, the potential for strain accommodation is restricted by the high stability of the CSL
boundaries with a resulting reduction in bulk plastic strain for a given load. Such clusters
of special boundaries are expected to induce collective grain activity, a phenomenon that
has been observed in MD simulations [72,73].

2.5.2. Cluster compaction


An experimental nc synthesis method used widely is the inert gas condensation method,
in which free clusters are collected, and sintered under high pressure and temperature.
At first sight such a technique can also be employed to produce nc samples. Individual
crystallites can be brought together within the 3D periodic simulation cell and a hydrostatic
load applied within the ParinelloRhaman framework. MD or Monte Carlo can then be
applied to allow the system to relax for a given pressure [7476]. In the case of MD,
typical simulation times are 100200 ps. As in experiment, this often results in a porous
structure existing throughout the simulation cell, the extent and degree of which can depend
sensitively on the precise simulation procedure [74].

2.5.3. Quenching from the melt


Another technique used in the preparation of computer generated nc systems involves em-
bedding small FCC crystalline configurations within a highly disordered structure above
the melting point of the model material. Such seeds can be at random positions and ran-
dom crystallographic orientations. The corresponding atoms are held fixed whilst the free
atoms of the disordered structure evolve in position at the elevated temperature using MD.
At some point the systems temperature is cooled down, with the fixed crystallites acting
as seeds for solidification, subsequently growing and meeting up at some point with those
atoms of neighboring crystallites, forming a GB interface region and network [7779]. The
structure of the GB in terms of the degree of atomic order obtained in such a way will de-
pend sensitively on the cooling rate and the pressure at which the simulation is performed.
In particular since simulation cooling rates are always very high due to the short-time re-
striction of the MD simulation technique, the simulated cooling down procedure actually
corresponds to an extremely fast quenching procedure when compared to experimental
cooling rates. It is therefore likely that highly energetic interface structures are produced
such as metallic amorphous grain boundary structures. The observation of such structures
in metallic FCC nc structures [79] led to an early and controversial debate on the nature of
GB interfaces in the nc environment.

2.6. Atomic visualization

To capture the essential properties of an nc system, a large number of grains must be in-
cluded within the atomic configuration. This in turn involves a large number of atoms, often
2.6 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 15

numbering in the millions. Therefore, it is obvious that inspection at the atomic scale re-
quires appropriate visualization methods in which filters can be applied to the atomic data.
Several visualization methods have been developed providing a level of detail in the struc-
tural characterization of the GB that simply cannot be achieved through experiment. These
methods facilitated the observation of atomic scale processes related to thermal processes
during sample annealing or to the deformation mechanisms during plastic deformation.
In addition to an atom position, a number of physical and structural properties can be
calculated and assigned to it such as its medium range order, positional disorder, energy,
local stress, etc. Using these quantities, one can develop a classification scheme allowing
for instance the consideration of only those atoms that constitute the GB region. Of course,
different classification schemes can lead to different views on the GB region [47]. For
example, if atoms that are locally non-FCC are classified as GB atoms, this will portray a
GB as a confined region with a thickness that is smaller than when the same GB is defined
as consisting of those atoms having a local stress higher than some critical value [80]. It has
been demonstrated that different patterns and degrees of order with different length scales
are obtained when different visualization criteria are used for GB atoms, causing concern
about fast conclusions regarding the disorder in nanosized GBs, when only one criterion
such as energy is used [47,78]. This demonstrates the importance of using a variety of
schemes, separately or in conjunction, to investigate the GB structure resulting from the
computer synthesis methods, and to clearly state those used when quoting calculated GB
properties.
In what follows we summarize the main visualization methods that are currently used
in the study of deformation behavior of nc metals, using an example of a dislocation that
has propagated half-way across an nc grain (see Fig. 5). In Fig. 5, the viewing direction is
parallel to the normal of the slip plane of the visualized full dislocation. For this example
which is taken out of a simulation at 300 K the atomic visualization is greatly improved
when thermal averages of the atomic positions and atomic quantities are used [81]. This is
found to be particularly the case for the local stress quantities [82,83].

2.6.1. Local energy


The simplest classification scheme is to view atoms according to their local potential en-
ergy. For a second-moment tight-binding or embedded-atom potential, the local potential
energy is given by
1
Ei = F [i ] + V (rij ).
2
j

For the nc system, GB regions can be to some extent identified by viewing only those
atoms with energies greater than a certain threshold, for instance exceeding the cohesive
energy by a value equaling the latent heat of melting. Such a method for the identification
of the GB region must be used with caution, since one is by definition considering only
those atoms in a high energy configuration, and therefore naturally biasing the probed GB
structure to more disordered configurations. In the past such a criteria has been used and
the calculation of the pair-distribution function of this selection of atoms has lead to the
incorrect conclusion that FCC nc GB are amorphous [79].
16 H. Van Swygenhoven and P.M. Derlet Ch. 81

Fig. 5. Six different methods of visualization of a dislocation segment within a nanocrystalline environment.
Atoms are shaded according to (a) potential energy, (b) coordination, (c) medium range order, (d) position disor-
der, (e) hydrostatic pressure and (f) centro-symmetric parameter.

Fig. 5(a) displays the atomic positions of those atoms with a cohesive energy of approx-
imately 0.1 eV higher than the FCC crystalline energy. Using this criterion the GB and
partial dislocation core are visible. Close inspection of the GB and TJ regions reveals an
interface thickness that contains many FCC atoms; and also a dislocation core region that is
quite diffuse. Moreover, the stacking fault that remains behind the partial dislocation core
cannot be identified a more negative energy tolerance would be needed to identify those
nearest neighbor coordinated HCP atoms that constitute the {111} stacking fault plane.
Such a modification to the energy cutoff tolerance would in turn lead to a more extended
and diffuse visualized GB network region.

2.6.2. Atomic coordination


One can also identify the GB region using the coordination number of each atom. In the
calculation of the coordination, the neighbor distance is generally defined as the mid-way
2.6 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 17

Table 1
The local symmetry classification list used for the MRO visualization method

PFCC FCC environment up to 4th nearest neighbor


GFCC FCC environment up to 1st nearest neighbor
PBCC BCC environment up to 4th nearest neighbor
GBCC BCC environment up to 1st nearest neighbor
PHCP HCP environment up to 4th nearest neighbor
GHCP HCP environment up to 1st nearest neighbor
OT12 12 coordinated atom without the above symmetries
OT8 8 coordinated atom without the above symmetries

distance between first and second nearest neighbor distances of the appropriate crystalline
structure. For example, an atom within the FCC environment has 12 nearest neighbors,
and it is expected that a certain percentage of the atoms belonging to the GB region will
not have a coordination of 12. An underlying problem of this method is that most atoms
belonging to the GB will retain their 12 coordination, and therefore the method is biased
in picturing only the less ordered highly energetic local atomic GB environments. Fig. 5(b)
displays the atoms according to their coordination with light grey representing twelve co-
ordinated atoms, and darker greys non-twelve coordinated atoms, and we see that the GB
region is clearly identifiable, as well as the full dislocation. An important limitation of
the atomic coordination is that it is unable to distinguish between HCP and FCC and thus
unable to visualize the stacking fault defect between the leading and tailing partial dislo-
cations.

2.6.3. Common neighbor analysis


Atomic visualization of grain and GB structures has been greatly facilitated by a medium
range order analysis of all atoms within the sample, which ascribes a local crystallinity
class to each atom [84]. This is performed by selecting the common neighbors of a pair
of atoms separated by no more than a second nearest neighbor distance, and introducing
a classification scheme for the nearest-neighbor bond pathways between the two atoms.
Since each crystalline symmetry has a unique topological signature, when all second near-
est neighbor bond permutations are enumerated, a local symmetry label can be assigned
to each atom see Table 1. In this classification scheme, for a given crystal symmetry
there are two groups: perfect and good, where perfect represents an atom situated
in a particular crystallographic environment up to four neighbor shells away and good
represents an atom situated in a particular crystallographic environment up to only the
first nearest neighbor shell. This difference is useful in identifying those atoms within a
crystalline environment close to a GB region or within the crystalline grain [85]. More
commonly these two groupings have been combined to define one crystallinity in order to
easily identify the entire crystalline grain regions.
This local atomic classification scheme allows the GB network and structure to be eas-
ily identified. For examples of such atomic visualization of the GB we refer to Ref. [86],
in which both high-angle and low-angle general GBs are shown. A significant advantage
of such a local crystallinity analysis is that (111) HCP planes represent twin planes, and
18 H. Van Swygenhoven and P.M. Derlet Ch. 81

two neighboring parallel (111) HCP planes represent an intrinsic stacking fault. The vi-
sualization of the twin planes has allowed for the easy identification of GBs containing
structural units of a symmetric boundary [86]. In the case of stacking fault defects, this
approach has given evidence for partial dislocation activity under uniaxial tensile loading
conditions [87]. Fig. 5(c) demonstrates the power of this visualization technique, where the
full dislocation and surrounding GBs are clearly evidenced.

2.6.4. Positional disorder


An additional classification concerned with GB atoms indicates whether or not an atom
is positionally disordered; an atom is regarded as positionally disordered when its loca-
tion cannot be attached to a lattice site of the nearby FCC grains [48]. At first sight this
may be considered a non-trivial task since in principle both the orientation and position of
the neighboring FCC lattices need to be determined. However when the local crystallinity
from the previous section is already known for the surrounding atoms, the task becomes
straightforward. If a GB atom has at least one nearest neighbor atom that is FCC, then that
GB atom must be at a lattice site of the FCC grain to which the neighboring FCC atom
belongs to. If all nearest neighbor atoms are non-FCC, then that GB atom will be position-
ally disordered with respect to the nearby FCC lattices. Such a definition is approximate
but generally appropriate for the fully dense and locally relaxed GB interfaces considered
in the present work [48]. Inspection of Fig. 5(d) demonstrates that the positionally disor-
dered atoms (black) can delineate the GB region, however it ignores much of the detail of
the local GB structure and is therefore very limited. Most importantly, such a visualization
method cannot resolve the dislocation core structure of the leading and trailing partial, nor
the stacking fault defect connecting them.

2.6.5. Local stress and strain


To calculate the global stress tensor within a computer simulation the virial theorem has
been generally used, which in the thermodynamic limit of large V and N , represents the
true bulk homogeneous stress. To investigate the spatial dependence of the stress field
within the nc sample, one generally applies the virial theorem directly to each atom. It is
however known that for such a volumetric partition, momentum is no longer conserved,
leading to non-negligible artifacts such as oscillatory behavior in strongly inhomogeneous
systems. A systematic approach has been developed that can represent local stress more
accurately [88,89] via
  
1 1 1
= mv v
i + F (rij )rij lij ,
 2 2
j

where  is the volume of some representative partition element,


i is unity if atom i
is within the volume element and zero otherwise, and lij is the fraction of the length of
the bond between atoms i and j lying within the volume element. The above equation
rigorously satisfies conservation of linear momentum. In the present work we choose the
volume element to be a sphere centered on each atom, and define the resulting stress of
that sphere as the local stress of the central atom. The radius of the sphere is taken as
0.4 nm and contains approximately 19 atoms, which is generally less than the range of
the empirical potential used. A thermal average is performed over 1 ps which typically
3 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 19

spans a few atomic vibrational periods. Rather than the full stress tensor, the two scalar
invariants of the stress tensor are directly visualized: the local hydrostatic pressure and
local maximum resolved shear stress. A positive value for the hydrostatic pressure rep-
resents compression and negative dilatation. Past work has shown that on average, FCC
metallic GBs are under a net tensile load although large variations between positive and
negative hydrostatic pressure occur within the GB regions whereas the grain interior is un-
der compressive load [24,83]. Fig. 5(e) displays the atoms shaded according to the local
hydrostatic pressure where the darkest shades represent atoms with a tensile pressure less
than 1 GPa, and the lighter shades a compressive pressure greater than 1 GPa. Fig. 5(e)
demonstrates that the full dislocation structure is difficult to visualize using this shading
scheme. This can be somewhat improved by changing the range of the shading scheme at
the expense of visualizing the GB region. Past work has shown that the local stress measure
becomes useful in identifying local regions within the GB that are under compression or
tension [82,83]. Moreover, the local stress becomes a sensitive indicator of the GB struc-
tural changes that occur due to the arrival and absorption of a lattice dislocation see
Fig. 13 in Section 3.2.

2.6.6. Centro-symmetric parameter


The centro-symmetric parameter is defined as

P = |R i + R i+6 |2 ,
i=1,6

where R i and R i+6 are the vectors corresponding to the six pairs of opposite nearest neigh-
bors in the distorted FCC lattice, giving numerical values of P that allow the identification
of atoms within the crystalline lattice, stacking fault and dislocation core region environ-
ments. This visualization technique was developed to visualize the atomic scale plasticity
occurring in nano-indentation MD simulations [90] and is particularly suited to visualizing
the full dislocation structure as is shown in Fig. 5(f).

2.6.7. Free volume


Free volume within a computer generated sample can easily be calculated by imposing a
fine mesh of grid points over the simulation cell [91]. For each mesh point, the shortest
distance to a nearby atom is calculated, if this distance is above a chosen value, then the
grid point is defined as being associated with free volume. The threshold distance must
be chosen to be at least half the nearest neighbor distance of the appropriate crystalline
lattice, to avoid inclusion of free volume arising from the interstitial regions. With this
information, the connectivity of the free volume regions (above a chosen threshold) leads
to an easy identification of the free volume content within a GB. See Fig. 22 in Section 4.2,
where free volume within a GB region has been identified.

3. Atomistic simulations of deformation in bulk 3D nanocrystalline metals


The use of large scale molecular dynamics to study the mechanical properties of nc ma-
terials provides a detailed picture of the atomic-scale processes during plastic deforma-
tion at room temperature and has served as an inspiration for dedicated experiments [12],
20 H. Van Swygenhoven and P.M. Derlet Ch. 81

where processes responsible for the plastic strain can be studied with an atomic resolu-
tion. Such processes have been classed as intergranular interface dominated processes such
as GB sliding [53,92,93], GB migration [24,94] and GB diffusion in the form of Coble
creep [56,95], and intragranular deformation processes involving dislocation activity with
corresponding GB mechanisms that accommodate the associated slip [21,22,2527,5963,
82,9698]. In the absence of defects within the grain interior it has been found that GBs
can act as both sources and sinks for partial or full dislocations [92]. Moreover, the sur-
rounding grain boundary environment can significantly affect the motion of a dislocation
as it propagates through the grain.
In particular atomistic simulation has revealed the following processes [21,92,99]:
Dislocation nucleation at the grain boundary is generally in regions where there is a
high hydrostatic pressure concentration prior to the nucleation which is relieved after
emission.
The nucleation and subsequent emission of the dislocation is accompanied by atomic
shuffles and usually stress-assisted free volume migration within the grain boundary.
The leading and trailing partial dislocations that would constitute a full dislocation need
not be nucleated in the same grain boundary region, nor for that matter, in the same grain
boundary.
Nucleation and propagation are separate processes often partials, whether leading or
trailing, are nucleated but do not propagate.
Grain boundary structure such as misfit regions or ledges can hinder dislocation propa-
gation to an extent that depends on the specific geometrical conditions and the Burgers
vector.
When a dislocation is pinned, i.e. cannot pass by a grain boundary region, the depinning
mechanism is thermally activated. Dislocation propagation between such pinning sites
can be considered athermal.
The nature of the dislocation activity, whether it is only partial dislocation, full disloca-
tion or twin migration can be understood in terms of an empirical potentials generalized
stacking fault energy curve.

3.1. The role of the generalized stacking fault energy curve

A wide range of atomistic simulations covering both two- and three-dimensional nc config-
urations has revealed that depending on the type of metal, either full dislocations or partial
dislocations are predominantly seen. For example, simulations for Cu and Ni exhibit only
partial dislocation activity, whereas for Al full dislocation activity is seen [99]. Whether
or not partial or full dislocation activity is predominant, has important consequences since
the former results in the creation of a stacking fault energy defect that transects the entire
grain. A rule-of-thumb argument that in the past has often been used to rationalize these
observations is that metals with a low stacking fault energy have a tendency towards only
partial dislocation activity since the cost in energy of leaving an extended stacking fault
is relatively low [58,100]. Whilst consistent with Cu which has a low stacking fault en-
ergy equal to 38 mJ/m2 , it fails to explain why in Ni with a stacking fault energy equal to
3.1 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 21

Fig. 6. General stacking fault curves for magnetic and non-magnetic Ni, Cu and Al. Figure taken from Ref. [104].

137 mJ/m2 partial dislocations are seen and in Al with a also a high stacking fault energy
of 146 mJ/m2 , full dislocations are seen [21].
The stacking fault energy is part of a more general energy surface concept referred to as
the generalized stacking fault energy (GSFE) curve. Formally this is defined as the energy
dependency of rigidly shearing a crystal at a (111) plane along a [112] slip direction [101]
and historically gives information about the shear properties of a perfect crystal and there-
fore insight into a materials ideal shear strength as originally proposed by Frenkel [102]
and more recently by Rice to study the nature of dislocation nucleation at a crack tip [103].
Although this quantity is not experimentally accessible, the GSFE curve has been the fo-
cus of considerable attention in the modeling of the mechanical strength of bulk crystalline
materials. Fig. 6 displays such GSFE curves derived from density functional theory calcu-
lations for Cu, Ni and Al [104]. We see that as a function of rigid displacement, the energy
of the system passes through a maximum that is referred to as the unstable stacking fault
energy.
Recent work has shown that both the stable and unstable stacking fault energies play a
role in determining the nature of slip activity observed in the simulations of plasticity in
nc GB networks [99]. Specifically, given the nucleation of a partial dislocation at the GB
and its propagation into the grain, the likelihood that a second trailing partial is nucleated
within the timeframe of a simulation depends on the ratio of the stable to unstable stacking
fault surface energy density for the empirical potential used in the simulation. If this ratio
is close to one, as it is in Al, then the likelihood of observing full dislocations within the
nanosecond time frame of a classical molecular dynamics simulation is high. If the ratio
is low as it is in Cu and Ni, the likelihood of seeing a full dislocation in a simulation is
reduced. The underlying argument for this trend is that upon nucleation and propagation
of a leading partial, the energy barrier that in part controls the probability of nucleating
a trailing partial is a function of the difference between the unstable and stable stacking
fault energy. If the ratio of these two quantities is close to one, then the barrier is small,
and trailing partial dislocation nucleation is likely to be observed. We note that this does
not constitute a criterion for full dislocation nucleation at a GB; rather it is a statement
that can rationalize all published simulation results on the nature of slip activity within
the nc environment [99]. Indeed, in addition to the applied stress, it is expected that the
22 H. Van Swygenhoven and P.M. Derlet Ch. 81

Fig. 7. Idealized generalized stacking fault, twin fault and twin migration energy curves for Al. Figure is taken
from Ref. [82].

detailed local conditions of the GB region will play a crucial role in whether or not partial
dislocation nucleation occurs. For example, dislocation nucleation is often seen to occur
in regions within the GB that contain strong local stress variations, which are relieved
upon nucleation. Atomistic simulation has revealed that atomic scale GB processes such
as atomic shuffling and free volume migration occur before and/or after partial dislocation
nucleation and facilitating local stress relaxation [96,105]. Whether the observed stress
relief is transferable to experimental conditions or whether this is related uniquely to the
purity of the GBs remains an open question. The fact is that upon dislocation nucleation
at a particular GB, the associated stress relief within that boundary makes it unlikely that
the nucleation of the trailing partial dislocation will be seen within the timeframe of a
simulation unless the ratio between an empirical potentials unstable to stable stacking
fault is very close to unity. The latter is the case for Al, for which the nucleation of the
second partial is very likely within the time frame of the simulation, irrespective of the
condition of the GB.
With the creation of a stacking fault defect that transects the entire grain, other slip
mechanisms become possible. In addition to the nucleation of a trailing partial on the same
{111} slip plane as the leading partial, there is also the possibility that the trailing partial
nucleates at a neighboring {111} plane resulting in the creation of a twin fault. Such a
process can also be modeled using the rigid slip model resulting in a generalized twin
fault energy (GTFE) curve. Fig. 7 displays the combined GSFE and GTFE curves for an
Al empirical potential [106] and demonstrates that the energy barrier to twin faulting is
higher than that of full dislocation activity. This gives an indication that for this model Al
material, twin faulting is possible but less likely than full dislocation activity. Deformation
twinning was first observed by molecular dynamics (MD) simulations of 2D Al columnar
structures [60]. In simulations of 3D nc Cu twin formation was observed at larger strains
as a competing process to partial dislocation activity [26].
Until here, we have considered only nc-GB networks that contain defect-free grains.
Similar arguments can also be help in the understanding slip mediated plasticity in
3.1 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 23

Fig. 8. An example of a grain in which (1) extended partial dislocation, (2) twin fault and (3) full dislocation
activity was observed. Figure is taken from Ref. [97].

grains containing transecting twin faults geometrically introduced before deforma-


tion [82,97,107]. For example, if a partial dislocation is nucleated at a pre-existing twin
plane, then the energy versus rigid displacement curve can be influenced significantly due
to the presence of the twin. For Al (Fig. 7), the so-called unstable twin migration energy
is significantly less than the unstable stacking fault energy indicating that if pre-existing
twins existed, twin migration would be the dominant slip activity. Case studies of atomistic
simulations of nc materials within and with out pre-existing twins will be given in detail in
the next two sections.
The above discussion on dislocation nucleation in terms of such rigid displacement en-
ergy curves raises a question regarding the extrapolation of the existence of extended par-
tial dislocations as a deformation mechanism towards experiments, since the observation
of extended partials might be influenced by the short time effect inherent to MD sim-
ulations. In particular, the atomistic simulations of nc Ni exhibit predominantly partial
dislocation activity which, as a result, leave a dense stacking fault network throughout
the nc grain structure a prediction that is not in agreement with experiment. However,
one expects that when much longer simulation times comparable to experiment could be
achieved at lower strain rates, atomic scale processes could build up local stress intensi-
ties at the intersection between the stacking fault defects and the surrounding GB. This, in
turn, could facilitate nucleation of a trailing partial that upon propagation would remove
the stacking fault, or create a twin laminar fault that itself might grow via twin migra-
tion. Indeed it becomes plausible that the dislocation activity observed at the time scale
resolution appropriate to experiment will be a mixture of the three types of slip systems
represented in the generalized energy curve, the relative importance being heavily depen-
dent also on the applied stress pattern and the GB structures. That all mechanisms are
possible is shown in Fig. 8 where one particular Al grain underwent all three types of
slip [97].
24 H. Van Swygenhoven and P.M. Derlet Ch. 81

3.2. Case study one: dislocations in a defect free grain

An important result of atomistic simulations for a variety of FCC metals and micro-
structural nc geometries has been that the plasticity is mainly due to propagating disloca-
tions that have nucleated at grain boundaries [25,92]. As discussed in the previous section,
the nature of the slip observed in MD is in part due to the GSFE curve of the empirical
potential employed and therefore the classification according to full or partial dislocations
should be taken with care. Simulations however have been important in revealing localized
atomic scale GB processes that occur before, during and after dislocation nucleation, prop-
agation and absorption. These processes are driven predominantly by a combination of the
applied stress and local stress relaxation within the GBs. Moreover GB sliding which to a
varying extent is always present, causes discrete atomic activity within the GB regions in
the form of atomic shuffling and localized free volume migration resulting in changes in
local stress, particularly at GB ledges and misfit regions that can lead to optimal conditions
for the nucleation of slip within the neighboring FCC lattice.
An example of the entire lifetime of a full dislocation in nc Al atomic configuration with
a mean grain size of 10 nm has been documented in detail in Ref. [21]. Using the MRO
atomic visualization, Figs 912 show the same atomic section over a period of time start-
ing just before dislocation nucleation until absorption of the remaining lattice dislocation
segment. For the central grain, the plane normal to the figures is the {111} plane in which
slip occurs. Fig. 9 shows the earlier stages of the nucleation of a partial dislocation which
is nucleated between 0 and 0.5 ps at a triple junction, remaining localized in this region
for nearly 7 ps, at which time a second partial is nucleated (see Fig. 11 at 7.5 ps). The
nucleation of such a partial dislocation is associated with pre- and/or post-cursor activity
within the GB that can involve free volume migration between the nucleation region and a
neighboring triple junction or region of misfit [96,105].
Between the nucleation of the leading and trailing partial, the leading partial fluctuates
in size by bowing out into the grain and retracting a number of times see Fig. 10. With
the nucleation of the trailing partial at another triple junction region, the full dislocation
then begins to propagate further into the grain until at 10 ps one end of the dislocation
becomes pinned at the GB. Locally this region of the GB contains a step/ledge structure
in an orientation relative to the leading partial dislocations Burgers vector that makes it
difficult to deposit the partial dislocation into the GB. As a result the dislocation bows
out into the grain, via the creation of kinks piling up at both ends of the dislocation. As a
result the leading partial dislocation can finally overcome the ledge structure, and deposit
itself into the GB at 11.25 ps, resulting in a depinning and further propagation of the full
dislocation into the grain.
Inspection of Figs 11 and 12 reveals that the propagation and shape of the dislocation
depends sensitively on where it is within the grain and the surrounding GB structure. For
example the splitting distance between the leading and trailing partial is a strong function
of the local stress within the grain and therefore variations in stress within a grain will
lead to variations in the splitting distance along the dislocation line [108]. The type of
GB influences how easily a dislocation propagates by depositing itself into the GB. For
example the GB on the right in Figs 11 and 12 is under a tension, and deposition of the
full dislocation into the GB occurs without any difficulty. Here, propagation appears to be
3.2 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 25

Fig. 9. Snaps shots of the life of a full dislocation nucleation of a leading partial dislocation.

athermal, which is not the case when pinning/depinning occurs [21]. Final absorption of the
full dislocation in the GBs in the lower part of Fig. 12, after 18.25 ps, also demonstrates that
the local structure of GB can result in only partial absorption, where the small dislocation
segment in the lower left triple junction region at 18.25 ps remains for 5 ps before it is
completely absorbed.
Fig. 13 now displays the local hydrostatic pressure as defined in Section 2.6.5 for the
same atomic configuration as in Figs 912, where in (a) the atomic section is taken just
before nucleation of the leading partial and in (b) the atomic section is taken just after
final absorption of the full dislocation. A change in the local stress can be observed as a
result of the slip. For instance, the high local compressive pressure related to the pinning
region has subsequently been removed, whereas the triple junction region where nucle-
ation has occurred now contains new regions of high compressive pressure. Furthermore
in other regions of the GB new compressive pressure regions have been created as a result
of the dislocation activity. Similar changes can be seen in the local deviatoric or maximum
resolved shear stress. Such high stress regions can themselves be catalysts for the future
nucleation of dislocations and although such events have not yet been observed in the time
frame of the simulation, it cannot be excluded that such structural changes might constitute
mechanisms that facilitate slip transmission into neighboring grains.
Due to the difference in the unstable and stable stacking fault, similar simulations per-
formed for nc Ni [99] have, until now, only observed the nucleation of a leading partial
which in the case detailed in Figs 912 would result in the creation of a stacking fault that
transects the entire grain.
26
H. Van Swygenhoven and P.M. Derlet
Fig. 10. Fluctuating leading partial dislocation before nucleation of trailing partial.

Ch. 81
3.2 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 27

Fig. 11. Snaps shots of the life of a full dislocation nucleation of a trailing partial dislocation and the creation
of a full dislocation which is pinned at one end to a grain boundary ledge structure.

Fig. 12. Snaps shots of the life of a full dislocation athermal propagation of a full dislocation through the grain
until eventual complete absorption.
28 H. Van Swygenhoven and P.M. Derlet Ch. 81

Fig. 13. Atomic section (a) before and (b) after the existence of a full dislocation, where atoms are shaded
according to their local hydrostatic pressure.

3.3. Case study two: a grain containing twin defects

Many experimental synthesis techniques produce nc FCC structures (Al, Cu, and Ni) with
grown-in twin structures. The general planar fault energy curves (Fig. 14) show that nu-
cleation of partial dislocations on planes adjacent to the pre-existing twin plane becomes
energetically more likely for Al, so that twin migration can be expected to occur. For Cu
the barrier for twin migration is similar to the unstable stacking fault energy, suggesting
that twin migration will not dominate over partial mediated dislocation activity. The GSFE
curves for Ni suggest an intermediate regime.
A range of dedicated simulations on samples with pre-existing twins have been per-
formed to test the suggestions coming from the GSFE curves [82,97,107]. Fig. 15 displays
an nc sample containing fifteen grains with each grain containing a number of twin planes,
identified as a dark grey HCP {111} plane transecting the entire grain. By inserting such
defects one is faced with a number of new degrees of freedom such as the Schmid factor of
the particular {111} plane and the number of twins per grain. In the deformation of such a
sample, twin migration was indeed observed and the nucleation of the partial dislocations
on the plane adjacent to the twin plane occurred in a similar way as to what was observed
for defect-free grains: i.e. in regions of high local stress within the GB and atomic scale
processes involving shuffling and free volume migration are seen before, during and after
partial dislocation nucleation.
In nc Al samples [82] twin migration was found to be the dominant process and com-
pared to the defect free sample with the same GB network, the number of dislocation events
was enhanced. Twins existing on {111} planes with a high Schmid factor contributed pre-
dominantly to the observed plastic strain. Moreover, when the number of twins was re-
duced, the amount of plastic strain reduced accordingly. Fig. 16 displays a cross-section in
which a twin plane migrated in the central grain and a twin migration is occurring (lower
right hand corner). When twin planes were along planes with a low Schmid factor, then the
migration activity was reduced [82].
Similar simulations performed for nc Cu and nc Ni revealed that twin migration was
not the dominant slip mechanism [82,97,107]. When all geometrically constructed twin
3.3 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 29

(a) (b)

(c)
Fig. 14. General planar fault energy curves as predicted from empirical potentials for Al, Cu and Ni. Figure is
taken from Ref. [97].

Fig. 15. The twinned nc Al sample with an average grain diameter of 12 nm. Figure is taken from Ref. [82].
30 H. Van Swygenhoven and P.M. Derlet Ch. 81

Fig. 16. Cross-section of a twinned grain. The (111) planes, indicated by white, show the original position of the
neighbor twin-boundary samples. Figure is taken from Ref. [82].

Fig. 17. Dislocations moving in regions constrained by twin boundaries in the nc Cu sample containing twin on
low Schmid factor 111 planes. Figure is taken from Ref. [107].

planes where chosen to be on high Schmid factor planes, extended partial dislocations on
planes parallel to the twin planes were observed. However, when all twins where chosen
to be on low Schmid factor planes, the same loading conditions resulted in less plastic
strain and extended partial dislocation activity occurred on slip planes with higher Schmid
factors resulting in the dislocations intersecting the pre-existing twin plane. Thus twin
planes can act as an obstacle for the propagation of extended partial dislocations across
the entire grain thereby restricting the total amount of plastic strain see Fig. 17. In these
simulations, increasing the applied load eventually did result in transmission of slip across
the twin faults albeit via a [100] slip system. For bi-crystalline configurations pure screw
dislocations can be transmitted through coherent twin boundaries [109,110]. The geometry
in these studies is however such that the impinging dislocation is a straight line, making
4 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 31

the extrapolation towards an nc environment not obvious. More recently, MD simulations


have revealed that step structures between twin planes existing in GBs with misorientations
close to a pure twin boundary can act as sources for full dislocation nucleation within nc
Al [23] an important result when one considers that heavily twinned polycrystalline Cu
with a mean grain size of 400 nm exhibits the high strength characteristic of nc materials
but with an increased ductility [111].
In summary, the presence of grown-in twins in nc FCC metals can change the deforma-
tion mechanism, but the effect is not the same for all FCC metals and can be explained
by the relative values of the extrema of the generalized planar fault energy curves. For the
case of Cu and Ni, which have twin migration energy barriers that are not so different from
their partial dislocation energy barriers, twin boundary migration will not become dom-
inant. For the case of Al, where the twin migration energy barrier is considerably lower
than the unstable stacking fault energy, twin migration becomes the favorable deformation
mechanism for twin planes with a high Schmid factor.

4. Experimental-computational synergy tools

The development of computational tools that measure experimentally accessible material


properties of computer generated nc samples constitutes a powerful approach in developing
more realistic atomic configurations, as well as providing a mechanism through which to
interpret experimental results without the use of oversimplifying models.

4.1. X-ray diffraction

X-ray diffraction (XRD) peak profile analysis has played a central role in the characteri-
zation of experimental nc microstructure, since it allows determination of the mean grain
size and the root-mean-square (RMS) internal strain. Williamson and Hall demonstrated
that the integral width (IW) of a 2 diffraction peak can be decomposed into a contribution
arising from a finite scattering volume and a contribution arising from local fluctuations in
strain. In terms of the scattering vector, s = 2 sin /, the IW can be written as

s = sSize + sStrain ,
where s = 2 cos /. Here sSize = 1/LSize , where

LSize is the
length that characterizes
the coherent scattering volume, and sStrain = 8  2 s/5. Here  2  is the root-mean-
square (RMS) strain. In the present context, LSize may be taken to represent the character-
istic grain size of the nc system. Thus the above equation demonstrates that if the IWs are
plotted as a function of their corresponding scattering angles, a linear relation is expected
where the y-intercept is equal to the inverse of LSize and the gradient proportional to the
RMS strain.
For nc materials, experiments have revealed rather high RMS strains (particularly for the
electrodeposited nc Ni samples) [52] suggesting large internal stresses exist within the mi-
crostructure. Such high internal stresses have been confirmed in stress dip-test experiments
revealing an effective internal stress of 1.4 GPa in electrodeposited nc Ni [65]. To investi-
32 H. Van Swygenhoven and P.M. Derlet Ch. 81

gate the possible origins of such high RMS strains, an algorithm has been created [80] to
calculate two-theta X-ray diffraction profiles from computer generated nc samples.
To calculate the elastic scattering intensity from a known atomic configuration, we con-
sider the N atom sample to be immersed entirely in a monochromatic plane wave X-ray
beam. Within the kinematical limit, the asymptotic scattered wave function for a given
scattering vector k is given by

=
scatt (k) fi exp(i k r i ).
i
Here fi is the X-ray atomic form factor for atom i. The scattering intensity as a function
of k may then be calculated from magnitude squared of the above equation. To obtain
diffraction peaks of comparable quality to that of experiment, a very large number of atoms
would be required. In the nc regime however, one may exploit the spherical symmetry of
an non-textured nc structure and perform a spherical average of the scattering intensity of
a given configuration. That is,
 2   sin krij 
1
I (k) =
dk
fi exp(i k r i ) = |f | N +
2
,
2 krij
i i,j,i
=j

where k = 2 sin / and rij = | ri r k |. In the above equation, all atoms are assumed
the same type giving fi = f . The final summation over atomic pairs can be recast into
a one dimensional integral weighted by the atomic configurations global pair-distribution
function, giving finally
rc  
sin kr rc cos krc sin krc
I (k) = 1 + dr l(r) + 4
0 kr k2 k3
rc
sin kr rc cos krc
1+ dr l(r) + 4 ,
0 kr k2
where l(r) is the inter-atomic pair correlation function, rc the continuum cut-off distance
which is typically slightly larger than the grain size, and the bulk number density. The
final approximation is valid when rc , which will generally be the case for the grain
sizes considered.
Fig. 18 shows two-theta diffraction spectra for two computer generated nc Ni samples
with respective grains sizes of 5 and 12 nm, using the geometrical Voronoi construction
technique outlined in Section 2.5.1. Close inspection [see Fig. 18(b)] reveals that for
the small grain size, the peak width is indeed greater. Fig. 19 now shows the so-called
WilliamsonHall plot for the calculated X-ray diffraction spectra and compares this to ex-
perimental data for electrodeposited nc Ni, which has a mean grain size of 30 nm. What
becomes immediately clear is that with increasing grain size, the calculated Williamson
Hall plot becomes flat when the grain size is larger than 5 nm. Extrapolating this result
to the experimental grain size of 30 nm, indicates that the geometrical approach of con-
structing nc samples would produce an nc structure with negligible RMS strain and no
anisotropy a result quite different from experiment.
Experimentally, from the large RMS strain and the WH anisotropy a dislocation content
of 4.9 1015 m2 could be calculated [112,113], values that are not really supported by
4.1 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 33

Fig. 18. (a) Calculated X-ray diffraction spectra of two computer generated nc Ni structures with mean grain
sizes of 5 and 12 nm. (b) Blow up of the 111 and 200 peaks.

Fig. 19. A WilliamsonHall plot of the Ni5 and Ni12 samples in terms of the scattering magnitude. Figure is
taken from Ref. [80].

TEM work. To check for possible sources of WilliamsonHall anisotropy, a certain lattice
dislocation content has been introduced in the nc Al MD samples by deforming the config-
urations to large strains followed by short unloading times during which the dislocations
could not all be absorbed in the GBs [114]. Fig. 20 displays the resulting WilliamsonHall
plot arising from four samples (where DIS0 is the as-prepared sample and DIS1 to DIS3
contain an increasing lattice dislocation content due to the increasing amount of plastic de-
formation experienced by the sample). Clearly, the introduction of lattice dislocations into
the grains increases the anisotropy, qualitatively reproducing the experimental anisotropy
seen in Fig. 20 for electrodeposited nc Ni. Since simulations where performed for nc Al
and the experimental results are for nc Ni, quantitative differences are expected due to the
lower elastic anisotropy of Al.
34 H. Van Swygenhoven and P.M. Derlet Ch. 81

Fig. 20. The calculated WilliamsonHall plot as a function of increasing dislocation content.

In the configuration DIS3 15% of the 100 grains contained a lattice dislocation that
was sitting well inside the grain, whereas the majority of the grains contained full or partial
dislocations close to the GBs. The most interesting observation from these calculations was
that even when the diffraction pattern is calculated from the same sample but excluding all
grains with full lattice dislocations, the WH anisotropy and the RMS strain as pictured in
Fig. 20 for DIS3 survived. In other words, dislocations that were incompletely absorbed in
the GBs also contribute to the overall shape of the WH plot and the average RMS strain.
Such an observation conforms with the TEM results on as-prepared and samples deformed
at room temperature as well as with the in-situ XRD experiments.

4.2. Phonons

Inelastic neutron scattering experiments have revealed that FCC nanocrystalline structures
exhibit anomalous bulk phonon properties. The phonon or vibrational density of states
(VDOS) can be calculated using two different approaches. In the first instance, the aver-
age VDOS for the grain interior and grain boundary regions is obtained via the Fourier
transform of the velocityvelocity auto-correlation function, the latter of which is given by
 1 

v(0) v(t)
= vi (0) vi (t),
N
i
and obtained directly from finite temperature MD simulations. In the above equation, the
averages are performed separately over the grain interior and grain boundary regions to
obtain the associated VDOS via the Fourier transform [85]. Although this method is rela-
tively efficient and yields energy spectra that can be compared to experiment, it does not
allow for an efficient investigation of the VDOS for one or a few atoms. To do this, the
local VDOS is calculated directly from the imaginary part of the onsite Greens phonon
function [115].
Fig. 21(a) displays the average VDOS for the interior of nc grains and Fig. 21(b) the
average grain boundary VDOS. Fig. 21(a) confirms that the vibrational properties of the
4.2 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 35

Fig. 21. Nanocrystalline Ni VDOS for (a) grain interior and (b) grain boundary region where the standard devi-
ation derived from an environment ensemble is shown in grey.

Fig. 22. High-angle GB with viewing angle parallel to the GB normal where atoms are shaded according to their
VDOS oscillator strength evaluated at phonon energies: (a) 44.791 meV and (b) 44.318 meV.

interior of the FCC grains differs little from that of the perfect lattice, whereas Fig. 21(b)
shows that the VDOS of the GB is fundamentally different [85]. In the high frequency
regime of the GB VDOS [Fig. 21(b)], a singular structure is observed in the grain bound-
ary VDOS. Detailed inspection of the grain boundary structure reveals that such singular
structure represents high-frequency localized modes at the interface region [115].
Fig. 22 displays the atomic structure of a grain boundary with atoms shaded according to
their local VDOS oscillator strength at two phonon energies at which VDOS singular peaks
are observed to occur. Here dark (light) shades of grey represent the maximum (minimum)
oscillator strength. Inspection of Fig. 22 indicates that the corresponding vibrational modes
are well localized in the grain boundary and in both cases correspond to regions of reduced
coordination [115].
The present work establishes that such anomalous behavior arises from the low- and
high-frequency vibrational properties of the grain boundary region [85,115], while within
the grain interiors, the vibrational properties remain similar to that of bulk crystalline be-
36 H. Van Swygenhoven and P.M. Derlet Ch. 81

havior. Moreover simulation has revealed that the low frequency non-quadratic behavior
may arise from a reduced effective dimensionality in the grain boundaries since the VDOS
is expected to scale at low frequencies as d1 , where d is the spatial dimension. On the
other hand, in the high-frequency regime, the anomalous behavior arises from localized
vibrational modes within the grain boundary region.
Knowledge of the grain boundary VDOS allows one to directly calculate the ther-
mal mean-square-displacement (MSD) of grain boundary atoms within the harmonic
approximation. Use of the VDOS displayed in Fig. 21 results in a thermal MSD of
85 106 nm2 . On the other hand, by calculating the XRD profile from atomic co-
ordinates derived from both instantaneous and averaged atomic configurations one can
extract bulk DebyeWaller factors and therefore bulk values for the MSD. Doing so gives
42 106 nm2 for the 12 nm and 46 106 nm2 for the 5 nm grain size sample,
which are not so different from the calculated MSD (43 106 nm2 ) and also the
experimental MSD (42 106 nm2 ) for bulk crystalline FCC Ni [116]. The smaller
values (when compared to the grain boundary) and relative insensitivity to grain size of
the XRD derived bulk MSD values reflects the fact that the Bragg peaks present in the
calculated XRD patterns primarily probe the nc grain interiors, and not that of the grain
boundary [80].

5. Discussion and concluding remarks

The most important suggestion resulting from the many MD simulations studies that have
been performed on nc FCC metals is that even in the absence of dislocation sources in
the grain interior, dislocations still play an important role in the deformation mechanism.
Indeed MD suggests that GBs act as dislocation sources as well as dislocation sinks, sug-
gesting a slip mechanism that does not build up a dislocation network and does not leave
dislocation debris. Using atomic scale visualization methods such as local crystallinity, lo-
cal hydrostatic pressure and maximum resolved shear stress to explore discrete atomic
activity, the importance of local grain boundary structure in terms of both the nucle-
ation/propagation and absorption of the dislocation has been demonstrated. Meanwhile,
many of the characteristic aspects of the suggested dislocation mechanism have been sup-
ported by experiments.
Already in the early beginning of TEM investigations of nc metals, the lack of a dislo-
cation network in the grain interior had been noticed in as-prepared as well as in deformed
samples. However to observe a dislocation mechanism without debris in-situ investiga-
tions had to be performed. It was soon clear that due to the high internal strain typical for
the experimental nc samples it was nearly impossible to follow with in-situ TEM the nu-
cleation and propagation of a dislocation in a grain. A few in-situ experiments have been
able to detect dislocation activity to some extent [117121]. The TEM sample restrictions
however did not allow the determination of where the dislocations were nucleated, which
for instance could have been at the surface of the thin TEM foil instead of at a GB, nei-
ther had it been possible to determine the Burgers vector of the dislocations. More recent
HR-TEM investigations have also confirmed dislocation storage at liquid nitrogen temper-
5 Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 37

ature [122] in electrodeposited nc-Ni, whereas no defect accumulation was observed at


room temperature.
In-situ X-ray powder diffraction performed with a special detector allowing the mea-
surement of a diffraction pattern over 60 at once, allowed for the first time to demonstrate
that in nc metals there is a plastic deformation mechanism operating that produces a similar
amount of peak broadening as is the case for polycrystalline metals during ongoing defor-
mation, however that upon unloading was fully recoverable [123125]. In coarse grained
structures, the permanent build up of peak broadening during deformation is related to the
building up of a dislocation network. The recoverable peak broadening therefore was the
first indication for the presence of a pervasive dislocation mechanism during plastic de-
formation and this in the absence of dislocation storage. Further in-situ experiments have
demonstrated that when this experiment is performed at temperatures below 300 K, the
peak broadening becomes irreversible, however when the temperature is raised back to
300 K during unloading, most of the peak broadening recovers [64]. Such an observation
can be understood in terms of the suggestions from MD that not only dislocation nucle-
ation but also dislocation propagation plays a role. Indeed, when performing MD at lower
temperatures [81], it has been demonstrated that propagation can be hindered at GB ledges
and that pinningdepinning is a thermally activated process. That propagation plays also
an important role has been experimentally confirmed by the high values for both the in-
ternal and effective stress measured during stress reduction experiments together with the
existence of negative creep [65]. Stress relaxation tests performed on electrodeposited nc-
Ni allowed to assign an activation volume to the rate limiting deformation mechanism of
the order of 1020b3 [126], which is at least a factor of 10 lower then the usual values
measured for coarse grained metals. However, MD simulations do not clarify whether the
rate limiting process is nucleation or propagation, since from the atomistic aspect a similar
volume could be assigned to both mechanisms [21].
In other words, the dislocation mechanism without debris suggested by MD simula-
tions is quite well supported by experiments. A full extrapolation of the MD results towards
the experimental regime however raises several major concerns: the first is related to the
non-similarity between the experimental and computational GB network, the second is re-
lated to the temporal and spatial restrictions inherent to the simulation technique, which
does not allow the determination of the rate limiting process.
All computational 3D samples that have been used for deformation studies differ from
the experimental samples in at least three ways: in XRD characteristics such as RMS
strain and WilliamsonHall anisotropy, in impurity content and in GB network geomet-
rical characteristics. Electrodeposited nc Ni has on average an RMS strain measured by
XRD of about 0.4% and the WilliamsonHall plot exhibits a characteristic anisotropy.
In-situ XRD studies have shown that only minor changes in RMS strain and Williamson
Hall anisotropy are observed between the as-prepared and plastically deformed electrode-
posited nc-Ni samples [114,124], changes which are moreover predominantly related to
the microplastic regime and not to the fully plastic regime. However, when calculating
XRD patterns from MD samples, the RMS strain is only of the order of 0.05% with the
WilliamsonHall plot being linear and therefore not exhibiting any type of anisotropy when
the average grain size is above 5 nm. In other words, one has to take into account that the
initial configurations considered in MD are different from the as-prepared experimental
38 H. Van Swygenhoven and P.M. Derlet Ch. 81

samples. Extensive calculations of XRD patterns of computational nc configurations have


demonstrated that a WH anisotropy similar to that what is experimentally measured can
be obtained from computational nc samples that have been deformed to large strains dur-
ing which several lattice dislocations have been nucleated from, and absorbed in GBs. In
other words, in order to obtain similar XRD characteristics between the experimental and
the simulated samples, the GB network of the computational sample needs to have more
pronounced stress intensities with a long range character extending into the grain interior.
As the XRD calculations demonstrate, such stress intensities can be obtained by means of
incomplete absorption of lattice dislocations. This in turn would mean that the GB char-
acteristics of Voronoi constructed samples differ substantially in terms of local stress dis-
tributions. However it is not excluded that the different characteristics measured by XRD
have to be related to other parameters such as for instance the lack in atomic impurities or
nanoscale precipitates within the GBs, which is no doubt another major difference between
experimental and MD samples. This in turn brings into question the role of the impurities
in dislocation nucleation and propagation mechanisms, a topic that has until now hardly
been addressed, predominantly due to the missing atomic potentials able to deal with light
element impurities such as oxygen, phosphor and sulfur. It is however to be expected, as
was discussed in Section 3.2, that the local relief of stress upon emission of a dislocation,
or the atomic activity in the GB related to the absorption of a dislocation might be to a
great extent influenced by impurity content in the GBs.
Other possible sources of important differences between experimental and computa-
tional samples could be connectivity among different grain orientations, Schmid factor
distributions among neighboring grains or localized textured clusters of grains. Such mi-
crostructural parameters are expected to play an important role in the nucleation and espe-
cially propagation of the dislocations. For instance, in 3D samples with general high angle
GBs, slip transmission from one grain towards another is usually not observed, except for
some special cases [127]. MD samples in which special grain clusters are repeated with
the GB network have however demonstrated the importance of these clusters in the overall
plasticity [50]. Especially in the latter case, where clusters of small angle GBs demon-
strated the tendency to form larger grains, which when extrapolating towards experiments
would mean the formation of a bimodal structure for which enhanced plasticity has been
seen experimentally [128].
The second major concern when extrapolating MD results towards experiments is of
course the timescale restriction inherent to MD. All MD simulations have been performed
at very high strain rates that cannot be compared with experimental conditions, except in
exceptional cases such as shock loading [129131]. Such high strain rates are necessary
in order to obtain enough plasticity during restricted simulation times, so that the char-
acteristics of deformation mechanism can be explored. MD simulations have been used
to study for instance yield or flow stress versus grain size [26] or to derive deformation
maps [58]: however these results have to be taken with extreme care. As demonstrated in
Section 2.1 the flow stress is clearly strain rate dependent, making extrapolation of simu-
lation stressstrain curves towards experimental stressstrain curves a precarious domain.
The same uncertainties are valid concerning the suggested deformation maps, where a crit-
ical grain size is suggested below which no dislocation activity is possible. For example,
when a uniaxial tensile stress of 1.2 GPa is applied to an nc-Al sample with a mean grain
Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 39

Fig. 23. Comparison of dislocation activity in Al for two 15 grain samples with an average grain size of 5 nm
under 1.2 and 1.5 GPa tensile stress. The insets show the formation of a full dislocation where (a) is a snapshot
from just after the nucleation of the leading partial and (b) shows the nucleation of the trailing partial.

size of 5 nm, a strain rate of 1 108 /s is seen and no dislocations are observed. However,
after the applied tensile stress is increased to a value of 1.5 GPa, the strain rate of the latter
increases to 1 1010 /s and full dislocations are observed, as can be seen from Fig. 23.
What has to be drawn as a conclusion from these results is that the competition between
the nucleation of a leading dislocation, the nucleation barrier for a trailing dislocation, the
additional amount of structural relaxation in the GBs accompanying nucleation and ab-
sorption processes is heavily influenced by the applied stress level in the simulation and
therefore MD simulations alone cannot alone be used for extracting rate limiting nucleation
criteria as a function of stress or grain size. This picture becomes even more complicated
when one realizes that the amount of structural relaxation and atomic activity observed dur-
ing nucleation/propagation and absorption of dislocations in an MD simulation might be
unrealistically high, due to the lack of a realistic impurity content within the GB. The high
strain rate/short time restrictions inherent to MD make it impossible to determine the true
rate-limiting processes, and therefore, atomistic simulations are at present not suited for
setting up a deformation map analogous to that used in constitutive plasticity for nanocrys-
talline structures, where the suggested mechanisms are quantified in terms of applied stress
and grain size.
Nevertheless, atomistic simulation has been very fruitful in, on the one hand confirming
some of the earlier dislocation theories of the 5060s, and on the other hand, posing new
questions especially concerning the interaction between dislocations and grain boundaries.

Acknowledgements
The authors wish to acknowledge the financial support of the Swiss National Science Foun-
dation and the 6th Framework European Union Program (NANOMESO).
40 H. Van Swygenhoven and P.M. Derlet

References
[1] F.R.N. Nabarro, Theory of Crystal Dislocations, Clarendon, 1967.
[2] J.P. Hirth, J. Lothe, Theory of Dislocations, second ed., John Wiley & Sons, 1982.
[3] J. Weertman, J.R. Weertman, Elementary Dislocation Theory, Macmillan, 1966.
[4] A.H. Cottrell, Theory of Crystal Dislocations, Gordon, 1964.
[5] J. Friedel, et al., Dislocations, Pergamon, 1964.
[6] W.T. Read, Dislocations in Crystals, McGrawHill, 1953.
[7] L.P. Kubin, et al., Solid State Phys. 2324 (1992) 455472.
[8] B. Devincre, L.P. Kubin, Mater. Sci. Eng. A (1997) 234236.
[9] K.W. Schwarz, J. Appl. Phys. 85 (1) (1999).
[10] M.C. Fivel, available from: http://www.gpm2.inpg.fr/axes/plast/MicroPlast/ddd/.
[11] H. Gleiter, Prog. Mater. Sci. 33 (4) (1989) 223315.
[12] H. Van Swygenhoven, J.R. Weertman, Mater. Today 9 (5) (2006) 2431.
[13] E. Arzt, Acta Mater. 46 (16) (1998) 56115626.
[14] R.W. Siegel, G.J. Thomas, Ultramicroscopy 40 (3) (1992) 376384.
[15] G.J. Thomas, R.W. Siegel, J.A. Eastman, Scripta Metall. Mater. 24 (1) (1990) 201206.
[16] J.P. Hirth, Metall. Trans. 3 (12) (1972) 30473067.
[17] C.W. Price, J.P. Hirth, Mater. Sci. Eng. 9 (1) (1972) 15.
[18] R.Z. Valiev, V.Y. Gertsman, O.A. Kaibyshev, Phys. Status Solidi A Appl. Res. 97 (1) (1986) 1156.
[19] D. Frenkel, B. Smit, Understanding Molecular Simulation, Academic, 2002.
[20] M. Parrinello, A. Rahman, J. Appl. Phys. 52 (12) (1981) 71827190.
[21] H. Van Swygenhoven, P.M. Derlet, A.G. Froseth, Acta Mater. 54 (7) (2006) 19751983.
[22] A.G. Froseth, P.M. Derlet, H. Van Swygenhoven, Acta Mater. 52 (20) (2004) 58635870.
[23] A.G. Froseth, P.M. Derlet, H. Van Swygenhoven, Scripta Mater. 54 (3) (2006) 477481.
[24] A. Hasnaoui, H. Van Swygenhoven, P.M. Derlet, Acta Mater. 50 (15) (2002) 39273939.
[25] J. Schiotz, F.D. Di Tolla, K.W. Jacobsen, Nature 391 (6667) (1998) 561563.
[26] J. Schiotz, K.W. Jacobsen, Science 301 (5638) (2003) 13571359.
[27] J. Schiotz, et al., Phys. Rev. B 60 (17) (1999) 1197111983.
[28] Y. Mishin, et al., Phys. Rev. B 59 (1999) 33933407.
[29] A.F. Voter, J. Chem. Phys. 106 (11) (1997) 46654677.
[30] A.F. Voter, Phys. Rev. Lett. 78 (20) (1997) 39083911.
[31] M.R. Sorensen, A.F. Voter, J. Chem. Phys. 112 (21) (2000) 95999606.
[32] A.F. Voter, Phys. Rev. B 57 (22) (1998) 1398513988.
[33] Y.S.A. Mishin, B.P. Uberuaga, A.F. Voter, Phys. Rev. B 75 (2007).
[34] H. Jnsson, G. Mills, W. Jacobsen, Classical and Quantum Dynamics in Condensed Phase Simulations,
World, 1998.
[35] T. Zhu, J. Li, S. Yip, Phys. Rev. Lett. 93 (2004).
[36] T. Zhu, et al., Proc. Natl. Acad. Sci. USA 104 (9) (2007) 30313036.
[37] M.W. Finnis, J.E. Sinclair, Philos. Mag. A 50 (1) (1984) 4555.
[38] F. Cleri, V. Rosato, Phys. Rev. B 48 (1) (1993) 2233.
[39] M.S. Daw, M.I. Baskes, Phys. Rev. B 29 (12) (1984) 64436453.
[40] P.M. Derlet, S.L. Dudarev, Prog. Mater. Sci. 52 (23) (2007) 299318.
[41] S.L. Dudarev, P.M. Derlet, J. Phys. Condens. Matter 17 (44) (2005) 70977118.
[42] F.H. Stillinger, T.A. Weber, Phys. Rev. B 31 (8) (1985) 52625271.
[43] J. Tersoff, Phys. Rev. B 39 (1989).
[44] D.G. Pettifor, et al., Comput. Mater. Sci. 23 (14) (2002) 3337.
[45] D.G. Pettifor, I.I. Oleinik, Phys. Rev. B 65 (17) (2002).
[46] M.W. Finnis, Interatomic Forces in Condensed Matter, in: A.P. Sutton, R.E. Rudd (Eds.), Oxford Series on
Materials Modeling, Oxford University Press, 2003.
[47] P.M. Derlet, A. Hasnaoui, H. Van Swygenhoven, Scripta Mater. 49 (7) (2003) 629635.
[48] P.M. Derlet, H. Van Swygenhoven, Phys. Rev. B 67 (1) (2003).
[49] H. Van Swygenhoven, M. Spaczer, A. Caro, Nanostruct. Mater. 10 (5) (1998) 819828.
[50] A.G. Froseth, H. Van Swygenhoven, P.M. Derlet, Acta Mater. 53 (18) (2005) 48474856.
Atomistic Simulations of Dislocations in FCC Metallic Nanocrystalline Materials 41

[51] G. Voronoi, J. Reine Angew. Math. 134 (1908) 198287.


[52] F. Dalla Torre, H. Van Swygenhoven, M. Victoria, Acta Mater. 50 (15) (2002) 39573970.
[53] H. Van Swygenhoven, P.A. Derlet, Phys. Rev. B 64 (22) (2001).
[54] H. Van Swygenhoven, M. Spaczer, A. Caro, Acta Mater. 47 (10) (1999) 31173126.
[55] H. Van Swygenhoven, et al., Phys. Rev. B 60 (1) (1999) 2225.
[56] V. Yamakov, et al., Acta Mater. 50 (1) (2002) 6173.
[57] V. Yamakov, et al., Philos. Mag. Lett. 83 (6) (2003) 385393.
[58] V. Yamakov, et al., Nat. Mater. 3 (1) (2004) 4347.
[59] V. Yamakov, et al., Acta Mater. 50 (20) (2002) 50055020.
[60] V. Yamakov, et al., Acta Mater. 51 (14) (2003) 41354147.
[61] V. Yamakov, et al., Acta Mater. 49 (14) (2001) 27132722.
[62] V. Yamakov, et al., Nat. Mater. 1 (1) (2002) 4548.
[63] P.M. Derlet, H. Van Swygenhoven, Scripta Mater. 47 (11) (2002) 719724.
[64] S. Brandstetter, et al., Appl. Phys. Lett. 87 (23) (2005).
[65] S. Van Petegem, et al., Appl. Phys. Lett. 89 (7) (2006).
[66] Y.M. Wang, A.V. Hamza, E. Ma, Acta Mater. (2006) 27152726.
[67] C.E. Krill, L.Q. Chen, Acta Mater. 50 (12) (2002) 30573073.
[68] D. Weygand, Y. Brechet, J. Lepinoux, Philos. Mag. B Phys. Condens. Matter Statist. Mech. Electron. Opt.
Magn. Prop. 78 (4) (1998) 329352.
[69] D. Gross, M. Li, Appl. Phys. Lett. 80 (5) (2002) 746748.
[70] K.S. Kumar, H. Van Swygenhoven, S. Suresh, Acta Mater. 51 (19) (2003) 57435774.
[71] W.M. Straub, et al., Nanostruct. Mater. 6 (58) (1995) 571576.
[72] A. Hasnaoui, H. Van Swygenhoven, P.M. Derlet, Phys. Rev. B 66 (18) (2002).
[73] A. Hasnaoui, H. Van Swygenhoven, P.M. Derlet, Science 300 (5625) (2003) 15501552.
[74] R. Meyer, et al., Phys. Rev. B 68 (10) (2003).
[75] M. Hou, V.S. Kharlamov, E.E. Zhurkin, Phys. Rev. B 66 (19) (2002).
[76] Q. Hou, et al., Phys. Rev. B 62 (4) (2000) 28252834.
[77] P. Keblinski, et al., Nanostruct. Mater. 9 (18) (1997) 651660.
[78] P. Keblinski, et al., Scripta Mater. 41 (6) (1999) 631636.
[79] S.R. Phillpot, D. Wolf, H. Gleiter, J. Appl. Phys. 78 (2) (1995) 847861.
[80] P.M. Derlet, S. Van Petegem, H. Van Swygenhoven, Phys. Rev. B 71 (2) (2005).
[81] H. Van Swygenhoven, Nat. Mater. 5 (11) (2006) 841.
[82] A. Froseth, H. Van Swygenhoven, P.M. Derlet, Acta Mater. 52 (8) (2004) 22592268.
[83] M. Samaras, et al., Phys. Rev. B 68 (22) (2003).
[84] J.D. Honeycutt, H.C. Andersen, J. Phys. Chem. 91 (19) (1987) 49504963.
[85] P.M. Derlet, et al., Phys. Rev. Lett. 87 (2001) 205501.
[86] H. Van Swygenhoven, D. Farkas, A. Caro, Phys. Rev. B 62 (2) (2000) 831838.
[87] H. Van Swygenhoven, A. Caro, D. Farkas, Scripta Mater. 44 (89) (2001) 15131516.
[88] J. Cormier, J.M. Rickman, T.J. Delph, J. Appl. Phys. 89 (7) (2001) 4198.
[89] J. Cormier, J.M. Rickman, T.J. Delph, J. Appl. Phys. 89 (1) (2001) 99104.
[90] C. Kelchner, S.J. Plimpton, J.C. Hamilton, Phys. Rev. B 58 (1998) 1108511088.
[91] S. Van Petegem, J. Kuriplach, Acta Phys. Pol. A 107 (5) (2005) 769775.
[92] H. Van Swygenhoven, Science 296 (5565) (2002) 6667.
[93] H. Van Swygenhoven, A. Caro, Phys. Rev. B 58 (17) (1998) 1124611251.
[94] A.J. Haslam, et al., Acta Mater. 52 (7) (2004) 19711987.
[95] D. Wolf, et al., Acta Mater. 53 (1) (2005) 140.
[96] H. Van Swygenhoven, P.M. Derlet, A. Hasnaoui, Phys. Rev. B 66 (2) (2002).
[97] A.G. Froseth, P.M. Derlet, H. Van Swygenhoven, Adv. Eng. Mater. 7 (12) (2005) 1620.
[98] P.M. Derlet, H. Van Swygenhoven, A. Hasnaoui, Philos. Mag. 83 (3134) (2003) 35693575.
[99] H. Van Swygenhoven, P.M. Derlet, A.G. Froseth, Nat. Mater. 3 (6) (2004) 399403.
[100] X.L. Wu, Q.Y. Zhu, Appl. Phys. Lett. 90 (2007).
[101] V. Vitek, Philos. Mag. 18 (1968) 773786.
[102] J. Frenkel, Z. Phys. 37 (1926) 572.
[103] J. Rice, J. Mech. Phys. Solids 40 (1992) 239271.
42 H. Van Swygenhoven and P.M. Derlet

[104] C. Brandl, P.M. Derlet, H. Van Swygenhoven, Phys. Rev. B 76 (2007) 054124.
[105] P.M. Derlet, H. Swygenhoven, A. Hasnaoui, Philos. Mag. 83 (2003) 35693575.
[106] Y. Mishin, et al., Interatomic potential for Al and Ni from experimental data and ab initio calculations, in:
Mat. Res. Soc. Symp. Proc., Mater. Research Society, 1999, pp. 533540.
[107] A.G. Froseth, P.M. Derlet, H. Van Swygenhoven, Appl. Phys. Lett. 85 (24) (2004) 58635865.
[108] A.G. Froseth, P.M. Derlet, H. Swygenhoven, Acta Metall. 52 (2004).
[109] Z.H. Jin, et al., Scripta Mater. 54 (6) (2006) 11631168.
[110] J.K. Chen, G. Jr. Chen, Philos. Mag. A 78 (2) (1998) 405422.
[111] M.W. Chen, et al., Science 300 (5623) (2003) 12751277.
[112] T. Ungar, A. Revesz, A. Borbely, J. Appl. Crystallogr. 31 (1998) 554558.
[113] T. Ungar, et al., Acta Mater. 46 (10) (1998) 36933699.
[114] S. Branstetter, P.M. Derlet, S. Van Petegem, H. Van Swygenhoven, Acta Mater. 56 (2008) 165176.
[115] P.M. Derlet, H. Van Swygenhoven, Phys. Rev. Lett. 92 (3) (2004).
[116] L.M. Peng, et al., Acta Crystallogr. Sect. A 52 (1996) 456470.
[117] M. Dao, et al., Acta Mater. 54 (20) (2006) 54215432.
[118] R.C. Hugo, et al., Acta Mater. 51 (7) (2003) 19371943.
[119] C.J. Youngdahl, et al., Scripta Mater. 44 (89) (2001) 14751478.
[120] M. Ke, et al., Nanostruct. Mater. 5 (6) (1995) 689697.
[121] K.S. Kumar, et al., Acta Mater. 51 (2) (2003) 387405.
[122] X.L. Wu, E. Ma, Appl. Phys. Lett. 88 (23) (2006).
[123] Z. Budrovic, et al., Appl. Phys. Lett. 86 (23) (2005).
[124] Z. Budrovic, et al., Science 304 (5668) (2004) 273276.
[125] H. Van Swygenhoven, et al., Rev. Sci. Instrum. 77 (1) (2006).
[126] Y.M. Wang, A.V. Hamza, E. Ma, Appl. Phys. Lett. 86 (24) (2005).
[127] C.B.E. Brandl, P.M. Derlet, H. Van Swygenhoven, Mechanisms of slip transfer through a general high
angle grain boundary in nanocrystalline aluminum, 2007, submitted for publication.
[128] Y.M. Wang, et al., Nature 419 (6910) (2002) 912915.
[129] M.A. Shehadeh, et al., Appl. Phys. Lett. 89 (17) (2006).
[130] E.M. Bringa, A. Caro, E. Leveugle, Appl. Phys. Lett. 89 (2) (2006).
[131] E.M. Bringa, et al., Science 309 (5742) (2005) 18381841.

Das könnte Ihnen auch gefallen