Acta Materialia 53 (2005) 4825–4838

www.actamat-journals.com

Transition of deformation mechanisms and its connection to

grain size distribution in nanocrystalline metals
B. Zhu, R.J. Asaro *, P. Krysl, R. Bailey
Department of Structural Engineering, University of California, San Diego, La Jolla, CA 92093, United States

Received 27 April 2005; received in revised form 23 June 2005; accepted 25 June 2005
Available online 1 September 2005

Abstract

A polycrystalline constitutive theory is developed based on the recently proposed model of Asaro, Krysl, and Kad (AKK) for
deformation mechanisms in nanocrystalline metals and the extended aggregate Taylor model of Asaro and Needleman (AN).
The AKK model is particularly concerned with mechanisms including the emission of perfect and partial dislocations, along with
deformation twins from grain boundaries, and grain boundary sliding, and with the grain size, strain rate, and temperature depen-
dent transition between these mechanisms. Thus of specific concern is the effect of grain size distribution as well as the mean grain
size on overall nanocrystalline aggregate response. Grain size distributions are simulated with log-normal distributions and incor-
porated into the AN aggregate model in the form of distributions of the volume fraction of grains with various sizes. Numerical
results for the case of electrodeposited Ni are shown to be in good agreement with available experimental data.

2005 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Keywords: Nanostructured metals; Grain boundary emission of dislocations; Grain boundary sliding; Grain size distribution; Mechanical prope-
rties; Nanocrystalline modelling

1. Introduction and larger, size range plastic deformation occurs via

When the grain size of polycrystalline metals transits
through the micrometre down to the nanometre there
are accompanying transitions in the mechanisms of
inelastic deformation as well as significant changes in
constitutive properties including, inter alia, levels of
strength, strain rate sensitivity, and strain hardening.
There is direct experimental evidence for these transi-
tions (see e.g. [1–4]), theoretical evidence vis-à-vis
molecular dynamics simulations of nanocrystalline
deformation [5–8], as well as suspicions that arise from
what is known about the mechanisms of plastic defor-
mation in crystalline metals. For example, in face cen-
tered cubic (fcc) metals with grain sizes in the micron,
*
Corresponding author. Tel.: +1 858 534 6888; fax: +1 858 534
6373.
E-mail address: rasaro@san.rr.com (R.J. Asaro).
the generation and motion of intragranular slip, i.e., dis-
location motion. This process is evidently shut off at
grain sizes somewhat below a micron. This is readily
understood by simply noting that the crystallographic
shear stresses required to generate and move dislocation
segments that exist within the well characterized net-
works which evolve during plastic flow are on the order
of s
Gb/‘, where G is the shear modulus, b the magni-
tude of the Burgers vector, and ‘ is the segment length.
But if dislocations are to be confined to the intragranu-
lar space – which might be taken as the definition of a
grain – ‘ must be less than the grain diameter, d. In fact
simulations of the operation of Frank–Reed sources
would suggest ‘ 6 d/4  d/3 (see e.g. [9]). This would
lead to the conclusion that s P (3  4)G(b/d), or
s/G P (3  4)(b/d). For pure Ni, since G
82 GPa, then
if d
1 lm, s P 82 MPa, which is reasonable. If, how-
ever, d
30 nm then s P 3280 MPa, which is too large

1359-6454/$30.00
2005 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.actamat.2005.06.033
4826 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838
by at least a factor of nearly 3! Not surprisingly, exper-
imental evidence shows that, at grain sizes in this nano-
crystalline range, grains seem free of dislocations in their
interior (see e.g. [10,11]). Likewise molecular dynamic
simulations of deformation of such nanocrystalline
polycrystals do not reveal intragranular dislocation
activity as part of the deformation process [5–7].
As importantly, at the nanocrystalline grain size
scale, common fcc metals display an increasingly large
sensitivity to strain rate, a property that affords them
ductility even though they have large strength. For later
reference, and for perspective, we note that at a grain
size of 20 nm the value of m from Fig. 1 is on the order
of m
0.035 and 1/m
28, as would appear in a power
law relation between strain rate and stress, i.e., as in a
relation of the form
_ / r1=m . This may be compared
to values of 1/m for fcc metals with more traditional
grain sizes, also as shown in Fig. 1, of 50–100 or so. Lit-
erature data, compiled in Fig. 1, include data for nano-
crystalline metals and Cu with nano-scale twins [12].
Recently, Asaro et al. [13] developed a simple, yet
compelling, mechanistic model for the most likely sce-
nario for the anticipated transitions in deformation
mechanisms that appears to be quite consistent with
experimental observations that exist to date, and which
as we show herein leads to a complete constitutive the-
ory amenable to computational analysis. The model is
based on the notion, which is shown to follow naturally
from dislocation and partial dislocation mechanics, that
at grain sizes below, say, 200–300 nm deformation oc-
curs via grain boundary dislocation emission; at grain
sizes below, say, 20–30 nm deformation may occur in-
stead by the emission of partial dislocations or deforma-
tion twins, and at even finer grain sizes deformation
occurs by grain boundary sliding. These mechanisms,
Grain Size (nm)
1000 100 50 20
0.04
0.03
m
0.02
0.01
Nanotwins in 500 nm grains
Literature data
0.00
0.0 0.1 0.2
d-1/2 (nm-1/2)
Fig. 1. Strain rate sensitivity exponent vs. grain size for nanocrystal-
line and ‘‘traditional’’ grain size fcc metals.
and their characteristic grain sizes, are influenced by
temperature, lattice spacing, and stacking fault energy,
as illustrated below. In fact stacking fault energy, inter
alia, is revealed to be a particularly important factor
in determining deformation mechanism and strength le-
vel. Moreover, it is shown that the proposed mecha-
nisms lead to an inherent strain rate sensitivity that is
associated with the various mechanisms and that is also
part of deciding which mechanism is dominant at a par-
ticular imposed strain rate.
Asaro and Suresh [12] have recently explored mecha-
nisms involved in the nucleation of either perfect or par-
tial dislocations as well as deformation twins from grain
boundaries for the specific purpose of rationalizing the
above mentioned trends of increasing strain rate sensi-
tivity in nanocrystalline fcc metals. Their analysis re-
veals the influence of both the intrinsic stacking fault
energy and the so-called unstable stacking energy intro-
duced by Rice [14] in his analysis of dislocation emission
from a crack tip. Asaro and Suresh [12] indeed demon-
strate how at grain sizes below 20 nm, or so, the activa-
tion volume for such emission processes falls to low
values on the order of 10–20b3 and, accordingly, the
strain rate sensitivity exponent, 1/m becomes on the or-
der of 20–30 or less. We will not make specific use of
their results in what follows except to note this fact
found both experimentally and theoretically as, for
example, from their analysis.
It is evident, therefore, that there is a very strong sen-
sitivity of mechanisms and overall response to grain size
and this will influence the response of nanostructured
polycrystals. This also means that in a given material
sample the distribution of grain sizes is vital to establish-
ing response. For example, and most particularly, we
will explore data obtained for electrodeposited pure Ni
that will explain that in a typical polycrystal with an
average grain size (by number count) of say 20 nm, the
much fewer grains by number count in the size range
>20 nm account for at least 50% of the volume fraction
– they, therefore, can dominate the behavior. In typical
nanocrystalline polycrystals, therefore, it is vital to pre-
cisely document the distribution of grain size as well as
to have an analytical (i.e., theoretical) framework for
interpreting the deformation mechanisms responsible
for the observed deformation. It is the development of
such a theoretical framework, a procedure for account-
ing for grain size distribution, and the analysis of exist-
ing data, that is the main goal of this series.
This paper is designed to be Part I of a series which ex-
plores the sensitivity of nanocrystalline aggregate re-
sponse to grain size distribution as well as to strain
rate, and temperature; the plan of Part I is as follows.
0.3
The Asaro–Krysl–Kad (AKK) model is reviewed and ex-
tended in the next section, where experimental data are
also described to illustrate its consistency. In Section 3,
the model is rendered into a constitutive theory for
 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838
nanocrystalline aggregates, and in Section 4 some exam-
ples of calculations are described. We adapt the ex-
tended Taylor model of Asaro and Needleman [15] for
this purpose and note that, within the context of that
aggregate model, grain distributions are to be expressed
as volume fractions as described below. Grain size distri-
butions, as obtained for electrodeposited nanocrystal-
line Ni by Dalla Torre et al. [16], are analyzed in
detail and our constitutive model is used to analyze
experimental data for pure nanocrystalline Ni. Conclu-
sions follow in Section 5. Part II of this series will be
concerned with a more extensive statistical analysis of
available data for nanocrystalline Ni and Pd as well as
Cu and Al.
2. Background: the AKK model
Fig. 2 will serve as a point of departure in reviewing
the model of Asaro–Krysl–Kad [13]. The figure illus-
trates a process of emission of a dislocation from a grain
boundary into the interior of a grain. Note that the fig-
ure describes details associated with the emission of a
partial dislocation, but it can also be used to explain
the result of the emission of a perfect dislocation. As ex-
plained by AKK, as the segment is emitted into the
grain it creates two trailing segments in the ‘‘side grain
boundaries’’, and in the case of a partial dislocation a
stacking fault within the grain. The energy (per unit
length) of these segments is taken as 1/2Gb2 for the per-
fect dislocation, and 1/6Gb2 for the partial dislocation,
respectively. The explanation for the latter value lies
simply in the fact that for a Schockley partial dislocation
in a FCC crystal, the magnitude of the Burgers vector is
p
bpartial = 1/ 3bperfect. As is evident, b ” bperfect is the
magnitude of the perfect Burgers vector. Now in the
case of the emission of a perfect dislocation, the mini-
mum required resolved shear stress is that required to
perform the work of creating the two residual segments
Fig. 2. Emission of a partial dislocation from an existing dislocation at
a grain boundary. As the leading partial dislocation enters the grain it
produces partial segments along the side boundaries.
4827
in the side boundaries, or sbd dx ¼ 2ð12Gb2 Þdx. This leads
to the remarkably simple result, s/G = (b/d). The d 1
scaling of stress level derives simply from the fact that
the area over which work can be performed by the ap-
plied shear stress itself scales with d for a given dx. As
noted by AKK, this leads to forecasted shear stresses
that are too high for grain sizes less than, say 20–
30 nm for typical fcc metals for which data exists.
For the case of the emission of a partial dislocation,
there is on the one hand a reduced requirement for work
associated with the residual segments in the side bound-
aries (owing to their lesser energy per unit length), but
now the additional requirement of creating a stacking
fault with an energy C per unit area. The analysis of
AKK is not repeated here, but the result for the emis-
sion criterion is:
ða  1Þ ~
fðsms =GÞbsð1Þ =jbj þ ðsmz =GÞbð1Þ
z =jbjg ¼ C þ 1=3b=d;
a
ð1Þ
~  C=Gb. Thus if we define
where a ” d/deq and C
sðaÞ  sms bsð1Þ =jbj þ smz bzð1Þ =jbj ð2Þ
we obtain
ða  1Þ ~
sðaÞ =G
1=3b=d þ C; ð3Þ
a
which is AKK
s equation (7). We also recall that deq is
defined as the equilibrium spacing of Schockley partials
in the absence of applied stress and is given as,
1
deq ¼ ðGb2 Þ=C. ð4Þ
12p
The coordinate system of the dislocation
s slip system is
such that m is the unit normal to the slip plane, s is unit
vector along the direction of the perfect Burgers vector,
b, and z is the third of a right-handed unit triad and lies
in the slip plane as well. AKK explore the effects of
material parameters such as stacking fault energy, mod-
ulus, and lattice parameter, inter alia, and successfully
explain some key trends described in the existing
literature.
An example of experimental data is shown to illus-
trate the consistency of these predictions, and those to
follow, and what is known experimentally. Fig. 3 shows
stress vs. strain data for tension for electrodeposited Ni
with a mean grain size of approximately 20 nm. In the
first place, we note that for this Ni the ultimate strength
at the highest imposed strain rate (_
¼ 4:5  103 s1 ) is
approximately rult
1800 MPa. If we take r
2s at
yield and just after, the above with data typical of Ni
(see Table 2), suggests that rult
2 · 1115 MPa =
2230 MPa, which is of the order of the observed ultimate
strength but is somewhat larger than the experimental
value of 1800 MPa. We will reconcile this difference by
analyzing the polycrystal with a suitable grain size
4828 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838

initiation of either perfect or partial dislocations from

Fig. 3. Tensile behavior of nanocrystalline pure Ni with average grain
size of 20 nm. Electrodeposited nanocrystalline Ni was obtained in the
form of 80 mm · 80 mm foils, 200 lm thick, from Goodfellow Metals.
Dog-bone shaped tensile specimens were cut, via EDM, with gauge
dimensions 500 lm · 200 lm. The gauge length was 12 mm. Tests were
conducted quasi-statically in a mechanically driven Instron test system
at 20 C at the strain rates indicated.
grain boundaries that requires comment. Such discus-
sion is provided below.
Between the highest strain rate imposed and the low-
est, viz.
_ ¼ 4:5  105 s1 , there is an approximately
200 MPa reduction in flow stress. Thus there is indeed
a significant sensitivity to strain rate which, if analyzed
via a power law as given above, would be estimated in
the range 1/m
25. This order of magnitude will
be used later vis-à-vis a phenomenological power law ki-
netic expression for the shear rate. A value of 1/m
25
is also consistent with the trends shown for the data of
Fig. 1, as noted in Section 1.
We now estimate the grain size, say for Ni, at which
there would be a transition from perfect dislocation
emission to partial dislocation emission by simply equat-
ing (b/d) to Eq. (3) as,
~
ðb=dÞ ¼ 1=3ðb=dÞ þ ~aC; ð5Þ
ða1Þ
where ~a  a
For Ni, ~a
1 and the estimate is that
.
the transition would occur at a grain size of approxi-
mately d
70b, or d
20 nm! The implications of such
transitions for polycrystalline response will be illustrated
below via our simulations.
3. A constitutive model

3.1. Perspective on discreteness

distribution as fitted to experimentally determined size
distributions. We also note the essential absence of
To begin, it is revealing to inquire as to the actual
strain hardening which may be also forecast from the
number of dislocations or stacking faults that would
AKK model as described by Eq. (1). There is, however,
need to be emitted into a typical grain whose dimensions
much more that requires comment concerning the
are on the order of, say 20 or 30 nm. Fig. 4 presents a

Table 1 s
List of faulting systems
Slip Perfect Burgers Leading partial, Trailing partial, d
s = nb
plane vector b+ b
n = no. dislocations/
(1 1 1)

½1 0 1

½1 1 2 ½2 1
1


stacking faults
(1 1 1) ½0

1 1 ½1
2 1 ½
1
1 2
suppose b = 0.3nm, then how many disl's/SF ?
(1 1 1) ½

1 1 0 ½
2 1 1 ½
1 2
1
ð

1 1 1Þ [1 1 0] [2 1 1] ½1 2
1
ð

1 1 1Þ ½0

1 1 ½
1
2 1 ½1
1 2 grain
ð

1 1 1Þ ½

10
1 ½
1 1
2 ½2
1
1

d = 10nm 20nm 30nm
size
ð1

1 1Þ [1 1 0] [1 2 1] ½2 1
1 γ = 0.05 n = 1.5 3 4.5
ð1

1 1Þ ½0

1
1 ½1
1
2 ½1
2
1


γ = 0.08 n = 2.4 4.8 7.2
ð1

1 1Þ ½

1 0 1 ½
2
1 1 ½
1 1 2
ð1 1
1Þ [1 0 1] ½2
1 1 [1 1 2]
ð1 1
1Þ ½0

1
1 ½
1
1
2 ½1
2
1
Fig. 4. How many dislocations or stacking faults are emitted into a
ð1 1
1Þ ½

1 1 0 ½
1 2 1 ½
2 1
1
nanocrystalline grain?

Table 2
Material data used in the simulations
G (GPa) b (nm)
b (nm) C/Gb deq (nm) DF (kJ/mol) v (m3) m c_ 0
29
80 0.249 0.144 1/100 0.5 97.5 1.54 · 10 0.037 103
 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838
simple analysis of this for perspective. The important
point to be illustrated here is that to induce strains that
are less than, say 10%, the number of defects involved in
a given grain is itself quite modest and typically less than
10. This, in turn, illustrates the rather discrete nature of
the deformation process that occurs in grains of this
size. We will, nonetheless, view the process in what fol-
lows as occurring in a continuous manner despite its
obviously discrete nature. For analyzing polycrystalline
regions, this is viewed as representing suitable averaging.
In other words, we take the view that the discrete slip
increments given grains undergo over a short time scale
on the order of 106 s can be time-stretched over much
longer times, provided the average ensemble
s strain
increment is the same in both cases. This means that
in the analysis, and at any given moment, the number
of defects emitted will be non integer, yet the average
plastic strain in a representative polycrystalline aggre-
gate will be equal to what would have been produced
by discrete slip events in a discrete group of grains.
3.2. Dislocation and partial dislocation slip systems
The slip systems associated with perfect dislocations
are the usual 12 (or 24) fcc systems involving the octahe-
dral {1 1 1} planes, and the face diagonal Æ1 1 0æ directions
lying within those planes. The choice of 12 or 24 is simply
one of allowing the slipping rate to possess algebraic sign
or to be strictly positive, respectively. In the latter case,
slip systems describing slip in positive and negative direc-
tions are required. Fig. 5 illustrates the kinematics asso-
ciated with partial dislocations, which are themselves
associated with parent perfect dislocations. The parent,
i.e., perfect, dislocation Burgers vector is denoted as ~b.
Partial systems are associated with each parent fcc slip
system, and for our purposes we may restrict attention
to slipping along only a single sense along each Æ1 1 0æ
Fig. 5. Partial dislocation slip systems.
4829
direction. This is because each perfect slip system natu-
rally defines two partial systems, one for slip along the
parent Æ1 1 0æ direction and the second for slipping in
the anti direction. Thus slip in what may be referred to
as the + and  directions, in any {1 1 1} plane, can be de-
scribed by 24 partial systems. Note that partial slip
events occur in particular sequences, e.g., slipping over
þ
the perfect Burgers vector, ~b, occur in the sequence, ~ b

followed by ~b . Slipping in the direction ~b on the other
 þ
hand occurs in the sequence ~b followed by ~b . Now
we recall that slip events involving partial dislocations,
i.e., the emission of stacking faults into the grains, in-
volve emitting only the leading partial dislocation.
Therefore, from the perfect slip system, (m, b) we obtain
the two partial slip systems, (m, b+) and (m, b). Table 2
lists the 24 such partial slip systems, each associated with
a perfect slip system.
The restriction of allowing only the ‘‘leading’’ partial
dislocation to be emitted derives from several rationales.
First, we recall from the dislocation analysis above that
on energetic grounds the emission of a single partial dis-
location is favored over the emission of a perfect dislo-
cation at sufficiently small grain sizes, Secondly, and
entirely consistent with this view, are the results of a ser-
ies of molecular dynamics simulations performed by
Van Swygenhoven and co-workers [5–7], which show
the same phenomena as summarized below.
Before reviewing this, however, the complete list of
partial dislocation, i.e., stacking fault based, slip systems
is provided (see Table 1).
These will provide the basis for developing the kine-
matics of slip via the emission of partial dislocations
from grain boundaries. We will not include, in what fol-
lows, the specific mechanism of deformation twinning
as, for example, analyzed by Asaro and Suresh [12],
but note that in the context of the present analysis the
criteria for twin emission would be similar to that used
here for partial dislocation emission.
3.3. Perspectives from molecular dynamics simulations
The simulations performed by Swygenhoven and co-
workers, noted above, have provided much insight into
the process of stacking fault emission in sufficiently
small grains. Fig. 6 illustrates an example of stacking
fault emission in Ni taken from their simulations. The
stacking fault is indicated in orange and the grain
boundary atoms are colored as blue. The fault is shown
at a particular time as it passed through the grain as it,
in fact, did in this simulation. There are several aspects
to the computed phenomenology that are directly rele-
vant to the development of a continuum theory for such
deformation.
The partial dislocations observed in such simulations
are generally observed to transit entirely across individ-
ual grains, yet on occasion multiple faults in a grain will
4830 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838

on the a fault system within a given grain during the

time period T, where each contributes a strain D(i),
then the average strain rate they contribute during that
time interval would be,
nðaÞ
1 X
p
_ ¼ DðiÞ . ð6Þ
T i¼1
Next realize that, as described by Table 2, there are
some 24 distinct fault systems that are, in general, acti-
vated in parallel. The total strain rate is then, following
the example of relation Eq. (6),
1 Xn
_ p ¼ nðaÞ DðaÞ ; ð7Þ
T a¼1 g
where, after examining Fig. 4, we identify D(a) as
1
DðaÞ ¼ bðaÞ mðaÞ . ð8Þ
d
In Eq. (8), b(a) is the Burgers
vector of the ath fault sys-
tem, and m(a) is the normal to the plane that fault acts
Fig. 6. Emission of a stacking fault from molecular dynamics [5].
on. At this point the approximation is made of taking
the time average over a single nano-grain, or an ensem-
interact and appear to block each other. Such occur-
ble average over many such nano-grains, to obtain,
rences may, in fact, give rise to at least modest levels
of strain hardening. When faults are preferred (over Xn
1
_ p ¼ n_ gðaÞ bðaÞ mðaÞ . ð9Þ
for example grain boundary sliding which is preferred d
a¼1
at very fine grain sizes), only the leading partials are ob-
served, as described and rationalized above. The leading The above relation is to be calibrated vis-à-vis the data
edge of faults, i.e., the partial dislocations themselves, is shown in Fig. 3.
generally not perpendicular to the ‘‘side boundaries’’ as To this end we recall the data shown in Fig. 3 which
is readily apparent in the figure. This suggests that there reveals a clear rate sensitivity. We assume that this arises
is resistance to motion of the partial dislocation within from the rate dependent process of producing what we
the side boundaries. This, in turn, may derive from the have referred to as the side boundary partial dislocation
need to create segments of partial dislocations in the side segments whose energy was described as, 16Gb2 above in
boundaries. Imagine for example, that the partial dislo- addition to the nucleation of dislocations as described,
cation were observed to be tangent to the side boundary. for example, by Asaro and Suresh [12]. At sufficiently
This would mean the partial dislocation would exert a low rates of deformation we assume that the need for
line tension force on the boundary equal to its energy such segments is precluded. Thus, we postulate a kinetic
per unit length, which is as noted above, equal to law for n_ ðaÞ as
1
Gb2partial . This leads to the same picture as outlined 8 n o
2 < ðaÞ ~ 1=m
above for the required shear stress for partial motion. ðaÞ n_ 0 s =G~ ðaÞ
aC ~ > 0;
if sðaÞ =G  ~aC
n_g ¼ g ð10Þ
But this particular simulated event shows an angle much :
0 otherwise;
greater than 0 which, of course, indicates that less work
is being performed at these side boundaries. This has di- where
rect relevance to the discussion associated with Fig. 2
and the effect of the first term in Eq. (3) and with a pos- 1
gðaÞ ðnÞ ¼ b=d þ f ðnÞ. ð11Þ
sible, albeit partial, explanation of the rate sensitivity 3
seen in Fig. 3. Some additional explanation is required at this stage of
development. In Eq. (10), the rate of generation of
3.4. A continuum theory for deformation via the emission faults, n_g ðaÞ , represents the rate of generating faults in
of stacking faults the entire grain; accordingly we can define the rate
n_ ðaÞ ¼ d1n_ gðaÞ as that per unit length of grain boundary.
With the perspectives offered above, a continuum the- If, however, faults are preferentially generated at grain
ory for deformation of nanocrystalline metals, via the boundary triple points, there may well be a grain size
emission of stacking faults, can be constructed as fol- dependence to this generation rate. Now, fault genera-
lows. Suppose that n(a) stacking faults have been emitted tion is only possible if the driving stress is sufficient to
 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838 4831

do the work of creating the associated stacking faults

and this is reflected by the fact that n_ gðaÞ , and n_ ðaÞ would
vanish if sðaÞ =G  ~ aC~ 6 0. At sufficiently low rates of
deformation, fault generation would ensue as long as,
as noted sðaÞ =G  ~ aC~ > 0, but at high deformation rates
ðaÞ
s =G  ~ ~
aC would have to be a significant fraction of
g(a); how significant would, of course, depend on the
strain rate sensitivity parameter, 1/m. As noted below,
during the discussion of calibration, the numerology
associated with the data of Fig. 3 suggests that
1=m  Oð20Þ or Oð30Þ; for that matter the same data
suggests that n_ 0  Oð5  102 Þ  Oð5  101 Þ. The inter-
sections of emitted faults represent impediments to con-
tinued fault propagation, and this represents a form of
strain hardening. To account for this, the phenomeno-
logical term f(n) P has been included. Here, and as an
example, n  na¼1 nðaÞ and f(n) describes an increase in
the resistance to generating and fully emitting faults as
faults accumulate within a given grain. Finally, let
b(a) = b(a)s(a), i.e., let s(a) be a unit vector along b(a) and
b(a) is the magnitude of the partial dislocation Burgers
Fig. 7. A composite model for SF emission and GBS.
vectors, that we immediately note are all crystallograph-
ically equivalent in a fcc crystal. Thus, since bðaÞ ¼
b, for
all systems we can write,
a rationale for the transition between dislocation in-
Xn X n
duced deformation and GBS. This will turn out, not
_ p ¼ n_ ðaÞ

bsm ¼ c_ ðaÞ sm; ð12Þ
a¼1 a¼1
surprisingly, to lead to a picture that is strain rate depen-
dent. Fig. 7 illustrates the scheme, consistent with their
where model. They construct what they refer to as the macro-
8 n ðaÞ o
<1
~ 1=m scopic shearing rate as,
bn_ 0 s =G~ aC
if sðaÞ =G  ~
aC~ > 0;
c_ ðaÞ ¼ d g ðaÞ
ð13Þ
: 6btD
0 otherwise; c_ ¼ sinhðv
s=kT Þ expðDF =RT Þ; ð15Þ
d
and where we may define a reference shearing rate as
c_ 0  d1n_ 0

b so that Eq. (13) becomes, where tD is a typical lattice vibrational frequency
8 n (
1013), v is an atomic volume (taken as b3 by them),
o
~ 1=m
< ðaÞ and DF is an activation energy for localized lattice/grain
ðaÞ c_ 0 s =G~ ðaÞ
aC
if sðaÞ =G  ~
aC~ > 0;
c_ ¼ g ð14Þ boundary diffusion. We further propose to interpret this
:
0 otherwise; shearing rate as an effective plastic strain rate within the
framework of an isotropic flow theory, in particular
3.5. A composite model for grain boundary sliding J2-flow theory. qInffiffiffiffiffiffiffiffiffiffiffi
Eq. ffi(15),
s is an effective shear stress
defined as
s ¼ 12r0ij r0ij , so that
sc_ represents the rate
As noted above, when the grain size is sufficiently of working per unit volume of material. J2-flow theory
small deformation may occur via grain boundary sliding assumes that the plastic part of the rate of deformation,
(GBS). This was described in the AKK model via the Dps, is directed so that Dpsir 0 , or given the definition of
s
phenomenological kinetic relations proposed by Conrad that
and Narayan [17]. Fig. 7 above illustrates a composite
Dps ko
s=or. ð16Þ
scenario for such dual deformation mechanisms. This
model assumes that that as individual nanocrystalline The flow law for this part of the deformation rate then
grains deform by the emission of stacking faults, they becomes,
may also flow past each other via sliding along their mu-
1
tual boundaries. The result would be the formation of Dps ¼ c_ r0 =
s. ð17Þ
shear zones, highlighted in ‘‘red’’ in the figure. As noted, 2
Conrad and Narayan [17] have recently provided a cor- At this point we assume that the total plastic rate of
relation between a simple phenomenological model of deformation is given as the sum of what is contributed
grain boundary sliding in nanostructured metals and by SF emission and by GBS, i.e., we take for the rate
available data – we use that model to attempt to provide of deformation due to SF emission
_ p ) Dpsf . Thus
4832 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838

O
the total plastic part of the rate of deformation is given
s ¼ s_  X  s þ s  X ¼ L : D þ D  s þ s  D . ð26Þ
as,
In Eqs. (24) and (25) the two Jaumann stress rates are
Dp ¼ Dpsf þ Dps . ð18Þ
O O Xn

These specifications for the plastic parts of the deforma- s ¼ s þ bðaÞ c_ ðaÞ ; ð27Þ
a¼1
tion rate will now be inserted into a finite deformation
kinematic and full constitutive framework. where
bðaÞ ¼ WðaÞ  s  s  WðaÞ ; ð28Þ
3.6. Constitutive framework

The material is assumed to flow through the partial
dislocation slip systems and via GBS, whereas elastic
deformation occurs via deformation of the lattice itself
(with material embedded on it). Rigid body rotations
are included in this lattice part. Hence as in [18] or in
the context of crystalline slip, in [19], the deformation
gradient is decomposed as,
F ¼ F  Fp ;
p
where F* and F are the lattice and plastic parts of F,
respectively. Note that the crystallographic vectors
introduced earlier, i.e., s and m convect with the lattice
so that in the deformed state they become,
s ¼ F  s; m ¼ m  F1 .
Note that as s and m are originally orthogonal, so are s*
and m*. Using Eq. (19) to form the velocity gradient we
uncover its lattice and plastic parts, which can be ex-
pressed in
 X
n
1 the divergence theorem,
F_  F1  F_  F1 ¼ Dp þ Xp ¼ c_ ðaÞ s m þ c_ r0 =
s;
a¼1
2 Z
3.7. An aggregate model
We follow here the construction of Asaro and Nee-
dleman [15] and imagine a collection of grains all of
which are characterized by a constitutive law such as gi-
ven in Eq. (24). The aggregate occupies a volume V
bounded by a surface Sext. The region is subjected to
all around displacement conditions that are taken to
ð19Þ
be such as to cause homogeneous deformations in what
is assumed to be an initially homogeneous sample. Aver-
aging theorems that rely on strict equilibrium and com-
patibility are available, but here we will have to relax the
requirement of point-wise equilibrium as follows.
Introduce the nominal (1st Piola–Kirchhoff stress) as
ð20Þ
n ¼ F1  s. ð29Þ
Note that n is related to the force, dP, transmitted
through a surface element m dS in the reference state
via dP = m Æ n dS. Now let F be a deformation gradient
that satisfies compatibility in each grain. Then using
XZ Z I
ð21Þ n : F dV ¼ n : F dV ¼ DT  u dS þ T  u dS.
V g Vg S int

where all terms have meanings as described above. Now ð30Þ

define
In Eq. (30), Vg represents the volume of a grain, Sint
ðaÞ 1 1 represents the internal interfaces between grains (i.e.,
P ¼ fs m þ m s g and WðaÞ ¼ fs m  m s g.
2 2 grain boundaries), and DT is the traction difference
ð22Þ between the grain boundaries. Note that if point wise
Then equilibrium held, DT = 0. Note also that an identity of
the form Eq. (30) holds with F or n, or both replaced
X
n
1 X
n
Dp ¼ c_ ðaÞ PðaÞ þ c_ r0 =
s and Xp ¼ c_ ðaÞ WðaÞ ; by their corresponding rates since the equilibrium equa-
a¼1
2 a¼1 tions expressed in terms of n are, nij,j = 0 and in rate
ð23Þ form n_ ij;j ¼ 0 in the absence of body forces. Here, as typ-
ical, commas denote differentiation and the summation
where we have explicitly made the assumption that convention is used.
GBS, expressed within a J2-flow theory framework, con- The homogeneous displacement boundary conditions
tributes nothing to the plastic spin. As shown in [15], the to be imposed take the form u ¼ F
 x on Sext. Substitut-
resulting constitutive relation for a crystal is given as ing this into the last integral, and rearranging terms, in
Xn   Eq. (30) gives
O 1 0
s¼L:D RðaÞ c_ ðaÞ  L : c_ r =
s ; ð24Þ
2 XZ Z I
a¼1
n : F dV  DT  u dS ¼ f xT dSg : F.

where g Vg S int
S ext
ðaÞ ðaÞ ðaÞ ðaÞ ðaÞ ðaÞ
R ¼L:P þb ; b ¼W ssW ; ð25Þ ð31Þ
 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838 4833

The basic assumption underlying Asaro and Needle- subject to statistical uncertainty that we specifically ad-
man
s [15] averaging model is that the contributions dress here. For grain size distributions, the numbers of
arising from any lack of equilibrium across the grain grains of various diameters in nanocrystalline materials
boundaries will be small compared to the integral taken can usually be well represented by a log-normal distribu-
over the external surface, an approximation difficult to tion function,
precisely measure in general. Invoking this assumption, "  2 #
however, yields 1 1 lnðD=D0 Þ
P ðDÞ ¼ exp  ; ð40Þ
8 9 ð2pÞ1=2 Dr 2 r
XZ <I =
n : F dV ¼ xT dS : F.
ð32Þ where the symbol D is used for the grain diameter and
V : ;
g g
S ext D0 and r are constant parameters describing the median

in and shape parameters (i.e., the standard deviation
In particular, if we specify F to have the uniform F of
each grain R 1 lnD) of the distribution, respectively, and
P ðDÞ dD ¼ 1. The arithmetic mean size D
can be
XZ 0
calculated [20] as
n : F dV ¼ V


n : F; ð33Þ  
g Vg

1 2
D ¼ D0 exp r . ð41Þ
where 2
Z
1 X X ~, the variance of D, is the average value of the quantity
r
n¼

n dV ; V ¼ V g. ð34Þ
2 , and is calculated from
ðD  DÞ
V g Vg g

is trivially the average of F in the aggregate, ~ ¼ D20 expðr2 Þðexpðr2 Þ  1Þ.

r ð42Þ
Since F
the average Kirchhoff stress is Commonly, the arithmetic mean grain size is often gi-


ven by material suppliers or can be obtained from the

s ¼ F n. ð35Þ
experimental measurement; if we assume the variance
To put the constitutive description in terms of the ~ is known, then the parameters, D0 and r which de-
r
nominal stress and rate of change of the deformation scribe the log-normal distribution can be obtained by
gradient we need only recall from Eq. (29) that inverting Eqs. (41) and (42), which yields
1
sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n_ ¼ F1  s_ þ F_  s; ð36Þ D
4
D0 ¼ ; ð43Þ
and from the definition of the Jaumann rate of Kirch- ~þD
r
2
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi!
hoff stress that,
r
~
O r ¼ exp ln þ1 . ð44Þ
s_ ¼ s þX  s  s  X. ð37Þ
2
D
These two expressions allow the constitutive relation, Most of the relevant available literature considers
Eq. (24), to be rephrased as that the mean grain size, internal strains and texture ob-
n_ ¼ K : F_  B;
_ ð38Þ tained from TEM and XRD well describe the structure
of the sample material (e.g. [16]). However, there is usu-
and if we designate individual grains with the super- ally considerable uncertainty associated with both of
script g, the constitutive relation for the grain g these methods (see e.g. the discussion in [21]). For exam-
becomes, ple, the question of whether a true characteristic re-
ðgÞ ðgÞ gion(s) of the sample was investigated for TEM can be
n_ ðgÞ ¼ KðgÞ : F_  B_ . ð39Þ
routinely raised. Therefore, an uncertainty analysis on
We recognize that when any such relation is constructed the grain size distribution is necessary, in order to pro-
for an individual grain that it is most naturally done vide a more accurate description of the actual behavior
with respect to a set of axes that are native to the crys- of the material, especially with respect to interpreting
talline grain in question, i.e., on crystal axes belonging experimental specimen to specimen scatter.
to that grain itself. The uncertainty analysis of this paper is conducted as
follows. We assume that the arithmetic mean grain size
3.8. Grain distributions is given, and what is uncertain is the variance of the
grain size distribution. According to Eqs. (43) and
Experimental techniques available for the full charac- (44), we obtain the parameters (the median and vari-
terization of structure at the nano-scale have been pri- ance) of the log-normal distributions for a certain arith-
marily based on transmission electron microscopy metic mean grain size, using reasonable standard
(TEM) and X-ray diffraction, from which the average deviations. An example is shown in the left side figure
grain size and grain size distributions can be estimated, of Fig. 8. In this the mean grain size is taken to be
4834 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838

20 nm, and the variance varies from 50 to 200. As noted 3.9. Calculational procedure
below, this mean grain size is consistent with that mea-
sured for electrodeposited Ni as tested by Dalla Torre 3.9.1. Boundary conditions
et al. [16] and ourselves. Denote the axes of a fixed Cartesian ‘‘laboratory’’
However as mentioned earlier, for the current analy- frame by ei and the cube axes in the grain g by ai; for
sis of dislocation emission and grain boundary sliding, example, if a tensile test were being set up the tensile axis
the volume-weighted grain distribution is considered to could be the e2 axis. For a FCC crystal the e2 axis would
be more directly relevant than the number-weighted fall somewhere within the standard triangle whose ver-
grain distribution as described by Eq. (40). If one as- texes are the [1 0 0], [1 1 0], [1 1 1] crystal directions. The
sumes each grain in the sample has the same shape, a initial orientation of the ‘‘laboratory’’ axes on with re-
sphere for example, then the total volume of the sample spect to the cube axes of a the typical grain g are de-
is scribed via the transformation,
Z 1 ðgÞ
ei ¼ UðgÞ  ai . ð47Þ
V ¼ kD3 P ðDÞ dD; ð45Þ
0 Note that this involves describing the orientation of
where k is a constant describing the shape of the grains each grain in a polycrystalline aggregate by the specifi-
(e.g., k ¼ 43p213 for a sphere), and the volume-weighted cation of its transformation tensor, U(g). Methods for
grain size distribution is then simply constructing such distributions are discussed in detail
by Asaro and Needleman [15]. For the calculations
kD3 P ðDÞ presented herein, we have taken the polycrystal as
Pv ¼ . ð46Þ
V being isotropic and thus initially composed on a uni-
We can then convert the log-normal distribution to a formly random distribution of grain orientations. Sup-
volume fraction-weighted distribution according to Eq. pose that there are N such grains, and further suppose
(46). The right side figure in Fig. 8 shows the volume- that each has the same initial volume, i.e., each occu-
weighted grain distributions for the above example, pies the same initial volume fraction of the aggregate.
whereas the left side figure illustrates the frequency dis- Then when averages are taken as specified in Eq. (34)
tribution. These distributions are illustrated via different for example the aggregate
s constitutive response is de-
forms below and in specific connection to the simulated scribed by,
experimental results.
The resulting volume-weighted grain distributions are n
_ ¼ K
_  B;

:F
_ ð48Þ
then used in the aggregate model of Asaro and Needle-
where
man [15]. For efficiency, the volume-weighted distribu-
tions are grouped into histograms with suitable 1 XN XN XN
subintervals, and the grains inside the aggregate are as- n
_ ¼ n_ ðgÞ ;
¼ 1
K KðgÞ ;
_ ¼ 1
B
ðgÞ
B_ .
N g¼1 N g¼1 N g¼1
signed the different grain sizes according to the relative
proportions described by the histograms. ð49Þ

Fig. 8. Grain size distributions based on an assumed mean grain size of 20 nm with to different variances.
 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838 4835

To use the tensile test as an example, we let the tensile and that F
_ 22 be specified, that is
axis be the e2 ‘‘laboratory’’ axis; thus the line of material
particles initially along the e2 axis remains so aligned so F
_ 22 ¼ specified stretching rate. ð54Þ
that,
For plane strain loading the third of Eqs. (52) is re-

 e2 ¼ F
22 e2 ;
F ð50Þ placed by
and
F
33 ¼ 0 ) F
_ 33 ¼ 0. ð55Þ
F
12 ¼ F
32 ¼ 0; ð51Þ
Other types of boundary conditions are easily specified
so that tensile fiber remains along the e2 axis. by following the example of tensile loading described
The remaining boundary conditions for axisymmetric above.
loading are:
T
1 ¼

n11 ¼0)
n_ 11 ¼ 0 on x1 ¼ constant; 3.9.2. Grain distributions for nanocrystalline Ni
T
1 ¼ n

31 ¼0)n
_ 31 ¼ 0 on x3 ¼ constant; The grain size distributions represented in Fig. 8 are
shown again in different form in Fig. 9 as functions of


T3 ¼ n
33 ¼0)
n_ 33 ¼ 0 on x3 ¼ constant;
ð52Þ the variance of the distribution. In Fig. 10 are shown


T2 ¼
n12 ¼0)n
_ 12 ¼ 0 on x1 ¼ constant; the two distributions as measured and reported by
T
2 ¼

n32 ¼ 0 ) n
_ 32 ¼ 0 on x3 ¼ constant; Dalla Torre et al. [16] for electrodeposited Ni and
our theoretical fit using the log-normal distribution
F
21 ¼ 0 ) F
_ 21 ¼ 0. give by Eq. (40) above. In the simulations described
Note that Eqs. (50) and (51) also require that below, we use these distributions to describe the grain
size along with the material data listed in the next
F
_ 12 ¼ F
_ 32 ¼ 0; ð53Þ subsection.

0.07 0.06
Volume weighted Number weighted
Number weigthed Volume weighted
0.06 Variance = 50 0.05
Variance = 100
Variance = 150
0.05 Variance = 200
0.04
Frequency
Frequency

0.04
0.03
0.03
0.02
0.02

0.01 0.01

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
(a) Grain size(nm) (b) Grain size (nm)

Fig. 9. (a) Comparison of the distributions with different variance and (b) histogram of the distribution with mean = 20 nm and variance = 100.

70 60 0.5
Percentage of grains per 10 nm

Batch A Batch B
Number of grains per 2 nm

Number of grains per 2 nm

60 50
0.4
50
40
0.3
40
30
30
0.2
20
20
0.1
10 10

0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
(a) Grain size (nm) (b) Grain size (nm) (c) Grain size (nm)

Fig. 10. (a), (b) Number-weighted grain size distributions according to Dalla Torre et al. [16]; (c) a simulated fit to the histogram of (b) ([16] Batch B).
4836 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838

3.10. Material data
The applications below are carried out for the case of
nanocrystalline Ni whose properties are listed in Table 2
along with the grain size distributions described in the
previous subsection.
The phenomenological hardening term, f(n), in Eq. (11)
is neglected in the simulated cases presented below.
4. Numerical simulations
Fig. 11 shows the simulated results using the grain
size distributions described above for uniaxial tension
at the three different strain rates we performed our uni-
axial quasi-static tension tests for electrodeposited nano-
crystalline Ni. The experimental uniaxial stress vs. strain
curves from Fig. 3 have been superposed for compari-
son. Note that as described above, the mean grain size
here was taken from the best fit to the data of Dalla
Torre et al. [16] (i.e., their batch B) with a mean grain
size of 20 nm and our fitted variance of 100. The simu-
lated curves are clearly below the experimental curves.
The right side figure in Fig. 11 shows the contributions
from the three deformation modes, viz. perfect and par-
tial dislocation emission, and grain boundary sliding, to
the overall tensile deformation. With this distribution,
the deformation is forecast to primarily involve the
emission of perfect dislocations.
The left side figure in Fig. 12 shows another simu-
lated stress vs. strain curve, but with an assumed mean
grain size of 23 nm and variance of 70. Note that now
the agreement with experiment is much better except
for an apparent larger ultimate tensile stress as simu-
lated than what the experimental curves appear to dis-
play. This apparent discrepancy is discussed in the
next section, but here we claim the agreement is in fact
quite good using a grain size distribution that nearly
equally fits the Dalla Torre et al. [16] data as does that
used to obtain the results shown in Fig. 11. Note also

2000 0.2
Strain rate 4.5e-3
Strain rate 4.5e-4
Strain rate 4.5e-5
1500 0.15 Full dislocation emission
Partial dislocation emission
GBS
Stress (MPa)

Shear strain

1000 0.1

Strain rate 4.5e-3

500 Strain rate 4.5e-4 0.05
Strain rate 4.5e-5
Numerical prediction
Experiment
0 0
0 1 2 3 4 0 0.05 0.1 0.15
Strain (%) Accumulated effective plastic strain

Fig. 11. Computed stress vs. strain response (left side figure) for nanocrystalline Ni with average grain size of 20 nm and variance of 100 as in
Eq. (42). Contribution to the overall plastic strain arising from the three deformation mechanisms considered. All simulations are carried out at 20 C.

2000
0.15 Strain rate 4.5e-3
Strain rate 4.5e-4
Strain rate 4.5e-5
1500 Full dislocation emission
Partial dislocation emission
0.1 GBS
Stress (MPa)

Shear strain

1000

Strain rate 4.5e-3 0.05

500 Strain rate 4.5e-4
Strain rate 4.5e-5
Numerical prediction
Experiment
0 0
0 1 2 3 4 0 0.05 0.1 0.15
Strain (%) Accumulated effective plastic strain

Fig. 12. Computed stress vs. strain response (left side figure) for nanocrystalline Ni with average grain size of 23 nm and variance of 60 as in Eq. (42).
Contribution to the overall plastic strain arising from the three deformation mechanisms considered but for a simulated nanocrystal with a mean
grain size of 10 nm and a variance of 50. All simulations are carried out at 20 C.
 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838
that the predicted trend in strain rate sensitivity is in
good accord with experiment. This agreement has been
accomplished using a value of 1/m in complete accord
with the data of Fig. 1, viz. 1/m
27.
The right side figure in Fig. 12 shows the computed
contributions to the overall tensile plastic strain for a
simulated uniaxial tension test for a nanocrystal with a
mean grain size of 10 nm and a variance of 50. Note that
with this smaller grain size there is noticeably more con-
tribution from deformation modes including partial dis-
locations (i.e., stacking fault emission) and grain
boundary sliding. Of course, as the temperature is in-
creased we expect contributions from modes such as
grain boundary sliding to increase further [17,22]. It
should be recalled from Eq. (5) that, given the material
properties of Ni, the grain size where we expect a tran-
sition from perfect to partial dislocation emission is
approximately 20 nm – this is what these simulations
show, at least at ambient temperatures like 20 C.
In Part II of this series we explore, in far more detail,
the influence of grain size distribution, material proper-
ties, temperature, and strain rate on material response.
This is also analyzed vis-à-vis stress and strain state
via example simulations of plane strain vs. uniaxial ten-
sion and compression deformation.
5. Discussion and conclusions
The examples presented above have illustrated the
main focus of this discussion, viz. the vital influence of
grain size distribution on the response of nanocrystalline
metals. As an example, we have shown that for electro-
deposited Ni with a mean grain size in the range of
20 nm (typical for materials that have been reported
on recently) the response is sensitive to grain size distri-
bution, yet not simply to the mean grain size, but also to
the spread about the mean grain size as measured by the
variance about the mean, as defined in Eq. (42). This
illustrates, inter alia, that future experiments need to
be diligent with respect to documenting grain size distri-
bution, but also need to develop methodology for pre-
paring nanocrystalline metals with better controlled
grain sizes that allow for more definitive discrimination
amongst the various potential mechanisms.
The comparison between the simulated tensile stress
vs. strain curves is seen to be quite good except that
the experimental curves display limited ductility and ter-
minate after quite modest plastic strains. This is clear in
the curves shown in Fig. 3 which are repeated on the left
side of Figs. 11 and 12. The experimentally observed fail-
ure mode involved abrupt necking and was ductile in that
the fracture surface was dimpled. These results will be de-
scribed in further detail in Part II. Similar curves in the
literature likewise often display limited elongation to
failure owing to the very high strength and low levels
4837
of strain hardening. In fact, it is most probable that most
recorded stress vs. strain response is terminated prema-
turely due to the onset of fracture brought on by the pres-
ence of small defects in the specimens.
Finally, we note that a comparison of the right side
figures in Figs. 11 and 12 shows that when the grain size
is at the transitional grain size range of 20 nm and just
below, as estimated for nanocrystalline Ni, the expected
deformation modes indeed begin to change. What this
means, however, is that at a grain size of 20 nm, and
with a grain size distribution typical of electrodeposited
Ni, observation of faults or twins would be anticipated
to be relatively rare events. In fact, as the variance of
the grain size distribution increases beyond say 100, as
we have defined this parameter, it is unlikely that any
deformation twinning or fault emission would be ob-
served at all. At mean grain sizes below 15 nm such
modes become prevalent enough to be anticipated via
TEM observation but only in the smallest of the grains
in the distribution. Again this shows the need for mate-
rials synthesized with more uniform grain sizes than are
currently available. With grain sizes below 15 nm, grain
boundary sliding is also seen to contribute to the overall
deformation as seen by the right side of Fig. 12.
Part II of this series will explore the effect of grain size
distribution, strain rate, and temperature in far more de-
tail and use additional data for nanostructured Ni and
Pd, along with the more limited data available for Cu
and Al.
Acknowledgments
R.J.A. and P.K. thank the National Science Founda-
tion NIRT initiative, specifically under Grant #0210173.
References
[1] Gleiter H. Nanocrystalline materials. Prog Mater Sci
1989;33(4):223–315.
[2] Wang YM, Ma E. Strain hardening, strain rate sensitivity, and
ductility of nanostructured metals. Mater Sci Eng A 2004;375–
77:46–52.
[3] Gertsman VY, Hoffman M, Gleiter H, Birringer R. The study of
grain-size dependence of yield stress of copper for a wide grain-
size range. Acta Metall Mater 1994;42:3539.
[4] Ebrahimi F, Bourne GR, Kelly MS, Matthews TE. Mechanical
properties of nanostructured nickel produced by electrodeposit-
ion. Nanostruct Mater 1999;11(3):343–50.
[5] Van Swygenhoven H, Spaczer M, Caro A. Microscopic descrip-
tion of plasticity in computer generated metallic nanophase
samples: a comparison between Cu and Ni. Acta Mater
1999;47:3117–26.
[6] Van Swygenhoven H, Derlet PM. Grain-boundary sliding in
nanocrystalline fcc metals. Phys Rev B 2001;64. 224105-1,5.
[7] Van Swygenhoven H, Caro A, Farkas D. A molecular dynamics
study of polycrystalline fcc metals at the nanoscale: grain
boundary structure and its influence on plastic deformation.
Mater Sci Eng A 2001;309:440–4.
4838 B. Zhu et al. / Acta Materialia 53 (2005) 4825–4838
[8] Hasnaoui A, Van Swygenhoven H, Derlet PM. On non-equilib-
rium grain boundaries and their effect on thermal and mechanical
behavior: a molecular dynamics computer simulation. Acta Mater
2002;50:3927–39.
[9] Hirth JP, Lothe J. Theory of dislocations. New York (NY): Mc-
Graw-Hill; 1968.
[10] Kumar KS, Suresh S, Chisholm MF, Horton JA, Wang P.
Deformation of electrodeposited nanocrystalline nickel. Acta
Mater 2003;51:387–405.
[11] Kumar KS, Van Swygenhoven H, Suresh S. Mechanical behavior
of nanocrystalline metals and alloys. Acta Mater
2003;51(19):5743–74.
[12] Asaro RJ, Suresh S. Mechanistic models for the activation volume
and rate sensitivity in metals with nanocrystalline grains and
nano-scale twins. Acta Mater 2005;53(12):3369–82.
[13] Asaro RJ, Krysl P, Kad B. Deformation mechanism transitions in
nanoscale FCC metals. Philos Mag Lett 2003;83(12):733–43.
[14] Rice JR. Dislocation nucleation from a crack tip – an analysis
based on the Peierls concept. J Mech Phys Solids
1992;40(2):239–71.
[15] Asaro RJ, Needelman A. Texture development and strain
hardening in rate dependent polycrystals. Acta Metall
1985;33:923–53.
[16] Dalla Torre F, Van Swygenhoven H, Victoria M. Nanocrystalline
electrodeposited Ni: microstructure and tensile properties. Acta
Mater 2002;50:3957–70.
[17] Conrad H, Narayan J. On the grain size softening in nanocrys-
talline materials. Scr Mater 2000;42:1025–30.
[18] Lee EH. Elastic–plastic deformation at finite strains. J Appl Mech
1969;36:1.
[19] Asaro RJ, Rice JR. Strain localization in ductile single crystals. J
Mech Phys Solid 1977;25:309–38.
[20] Evans M, Hastings N, Peacock B. Statistical distributions. 3rd
ed. New York (NY): Wiley; 2000.
[21] Krill CE, Birringer R. Estimating grain-size distributions in
nanocrystalline materials from X-ray diffraction profile analysis.
Philos Mag A 1998;77(3):621–40.
[22] Nieman GW, Weertman JR, Siegel RW. Mechanical
behavior of nanostructured metals. Nanostruct Mater
1992;1:185–90.