Professional Documents
Culture Documents
www.actamat-journals.com
Received 28 January 2003; received in revised form 8 April 2003; accepted 8 April 2003
Abstract
We use massively parallel molecular dynamics simulations of polycrystal plasticity to elucidate the intricate dislo-
cation dynamics that evolves during the process of deformation of columnar nanocrystalline Al microstructures of grain
size between 30 and 100 nm. We analyze in detail the mechanisms of dislocation–dislocation and dislocation–twin
boundary reactions that take place under sufficiently high stress. These reactions are shown to lead to the formation
of complex twin networks, i.e. structures of coherent twin boundaries connected by stair-rod dislocations. Consistent
with recent experimental observations, these twin networks may cause dislocation pile-ups and thus give rise to
strain hardening.
2003 Acta Materialia Inc. Published by Elsevier Science Ltd. All rights reserved.
1359-6454/03/$30.00 2003 Acta Materialia Inc. Published by Elsevier Science Ltd. All rights reserved.
doi:10.1016/S1359-6454(03)00232-5
4136 V. Yamakov et al. / Acta Materialia 51 (2003) 4135–4147
thesis processes, [4,5,11–13] show that at grain to be considered, revealed complete development
sizes above 10 nm high dislocation densities of deformation twins in the high-SFE metal. Taken
(usually above 1015 m⫺2), comparable to the dislo- together, these simulations [17,20,21] suggest that
cation density in work-hardened coarse-grained deformation twinning may accompany the slip pro-
metals [14], are observed. On the other hand, no cesses in nanocrystalline fcc metals. This sugges-
dislocations were found in ball-milled Cu-Zr with tion has found experimental support in a recent
a grain size under 100 nm [15]. study by Wang et al. [24] who report that twinning
In addition to such experimental studies, recent is an important deformation mode in nanocrystal-
computer simulations have demonstrated that dis- line and ultrafine grains (⬍ 300 nm) of Cu.
locations may, indeed, play a role in the defor- The purpose of this paper is to study the inter-
mation of nanocrystalline fcc metals [16–21]. In play between deformation twinning and dislo-
particular, we have recently demonstrated a con- cation–slip processes in nanocrystalline fcc metals
nection between the grain size and the type and columnar microstructures of grain size between 30
structure of the existing dislocations in terms of a and 100 nm. In particular we focus on the reactions
length-scale competition between the grain diam- between slip dislocations and twin boundaries.
eter and the dislocation splitting distance (i.e. the Reactions between dislocations on different slip
size of the stacking fault (SF), that connects the systems, including the well-known process of
partial dislocations to form extended perfect cross-slip and the formation of Cottrell–Lomer
dislocations) [19]. Our conclusion was that when locks, are also observed and analyzed. Twin-
the grain size is less than the splitting distance, boundary networks, which are structures of coher-
only partial dislocations can be nucleated from the ent twin boundaries connected by stair-rod dislo-
GBs, i.e. single partials which produce SFs as they cations, are shown to form as an outcome of these
glide through the grain until they become incorpor- reactions in grains of 70 and 100 nm in size. It is
ated into the GBs on the opposite side [16–18]. suggested that these formations may lead to the
These SFs remain in the grains as planar defects. strain hardening seen in low-SFE coarse-grained
After the grain interior has been transected by a metals, thus offering an alternative explanation to
number of SFs, further dislocation propagation the experimentally observed Hall–Petch behavior.
cannot take place. The dislocation propagation pro- Molecular dynamics (MD) and molecular statics
cess can therefore operate only during the initial simulations of dislocations and twinning in fcc
stages of deformation and thus differs significantly metals performed to date have been limited to sin-
from the usual slip mechanism observed in coarse- gle crystals. For example, there have been studies
grained materials. The conclusion of these simula- of dislocation–dislocation reactions between differ-
tions was that for the smallest grain sizes the domi- ent slip systems, such as cross-slip in Ni [25] and
nant deformation mechanism must involve GB- Cu [26,27], annihilation of screw dislocations in
mediated processes, such as GB sliding [16–18] Cu [28] and Cottrell–Lomer lock formation and
and GB diffusion [22,23]. By contrast, when the destruction [29]. Dislocation–twin boundary inter-
grain size is larger than the dislocation splitting actions have also been studied recently by atom-
distance, complete nucleation of perfect dislo- istic simulations, but again only in single crystals
cations from the GBs can take place, thereby initi- [30]. These studies agree well with the theoretical
ating slip deformation [19]. analysis for dislocation reactions in a single crys-
In addition to dislocation activity, the very early tal [31,32].
stages of twinning, in terms of the emission of Our columnar model microstructures offer the
coupled partial dislocations from the GBs that pro- possibility to extend these simulation studies on
duce extrinsic SFs, have been observed in a low dislocation and twinning processes in single crys-
stacking-fault energy (SFE) metal, Cu, in some of tals to polycrystalline materials, while preserving
the earlier simulations [17]. Large-scale simula- most of the characteristics of fully three-dimen-
tions [20,21] on a columnar microstructure [19], sional (3D) models. The limitations of the colum-
allowing much larger grain sizes and plastic strains nar model come mainly from the inability, due to
V. Yamakov et al. / Acta Materialia 51 (2003) 4135–4147 4137
atoms in a local hcp environment and those in an limiting dynamical processes be compared with all
fcc environment. A single line of hcp atoms thus other relevant characteristic stresses and time
represents a twin boundary; two adjacent hcp lines scales in the system.
indicate an intrinsic stacking fault and two hcp Concerning the present simulations, in spite of
lines with an fcc line between them represents an the large tensile stresses (of 2.3 GPa) and strain
extrinsic stacking fault. The atoms that were found rates (~107 s⫺1) under which deformation is
to be neither in the fcc, nor in the hcp environment observed, typical dislocation-glide velocities (of
were marked as ‘disordered’ atoms. They are about 500 m/s) are well below the sound velocity
mostly seen in the GBs (see Fig. 1) and in the dis- (of 3664 m/s in the [100] direction for this poten-
location cores. Thus a single spot of disordered tial [28], compared to the experimental value of
atoms is most likely to be a core of a dislocation 3050 m/s). More importantly, however, the
or a vacancy. If this core is connected to a SF, then resolved shear stresses, of the order of 1 GPa on
we see a partial dislocation. In addition a lattice the (111) slip planes of our system, are well below
circuit analyses [31,32] on (110) plane (the x–y the theoretical shear strength, s th = 2.84 GPa esti-
plane in our columnar model in Fig. 1) was used mated by first-principles calculations for Al at 0
to identify the edge component of the dislocation’s K [41].
Burgers vector (due to the [110]-columnar The issue of the high strain rates in MD simula-
geometry of the model, the dislocation line is tions was also recently addressed in high-tempera-
always perpendicular to (110) plane as explained ture deformation simulations of nanocrystalline Pd
in the Section 3). To identify the screw component [23]. These simulations revealed that, in spite of
we use a cross-section of the dislocation’s glide the extremely high strain rates ( ⬎107 s⫺1), the
plane, as defined by the edge component and the Coble-creep equation [42] describing GB diffusion
dislocation line, where the atoms’ displacements at creep in coarse-grained materials was quantitat-
the dislocation core show the dislocation shift. An ively reproduced. Most importantly, the activation
example of such a cross-section to analyze an energy for the creep rate was found to agree com-
extended dislocation was given in Ref. [19]. pletely with that obtained from separate simula-
All of our simulations were carried out under tions of self-diffusion in bicrystalline GBs, i.e. in
constant tensile loads of 2.3–2.5 GPa applied along the absence of applied stress [33]. These simula-
the x-axis (see Fig. 1) and at a temperature of T tions demonstrate that the artificially high strain
= 300 K. (The melting point for this interatomic rates present in MD simulations, and the necessar-
potential was estimated [36] at 940 K, compared to ily high applied stresses, need not change the
the experimental value of 933 K.) The Parrinello– underlying GB physics. This is also supported by
Rahman constant stress technique [37] combined the experimental work of Pond and Glass [43] on
with the Nose–Hoover thermostat [38] and an inte- high-velocity deformation of Al single crystals
gration time step of 0.5×10⫺15 s was used to achi- showing that the deformation modes of Al at strain
eve the constant stress–constant temperature rates of up to 106 s⫺1 do not differ significantly
dynamics of the system. from those at slow-rate static testing.
It is important to be aware of the fundamental
limitations inherent to this model. Apart from
being limited to systems which are relatively small 3. Reactions between dislocations on different
in size (even in the columnar geometry considered slip systems
here), by their very nature MD simulations are
restricted to very high strain rates and, hence, To investigate the complex dislocaton dynamics
stress. For example, a strain of 1% occurring in 10 taking place in our nanocrystalline microstructures,
ns of simulation time corresponds to a strain rate we will start by analyzing several well-known dis-
of 106 s⫺1. It is therefore imperative that for each location reactions, such as cross-slip and formation
type of simulation, the characteristic stresses and of stair-rod locks as they are observed in our simul-
time scales associated with the deformation rate- ations. Although these are well-known processes,
V. Yamakov et al. / Acta Materialia 51 (2003) 4135–4147 4139
their mechanisms have not been directly seen A or B node of the Thompson triangle in Fig. 2.
before in simulations of polycrystalline materials. These 60° dislocations produce a shear defor-
More importantly, the crystallographic analysis of mation along the glide plane equal to the edge
these processes will help us to understand the more component of their total Burgers vector, which lies
complicated dislocation–twin reactions taking in the deformation plane (x–y plane). These dislo-
place in our system at larger strain. cations are commonly seen in our model because
they are needed to accommodate the resolved shear
3.1. Crystallographic analysis of the dislocations stress on the glide planes.
and cross-slip The second type of dislocation is a pure screw
BA. Being common to both glide planes, (111)
Two types of intrinsically dissociated dislo- and (111̄), it can dissociate on either one of them,
cations can exist on each of the two glide planes as Bd + dA in the (111) plane or as Bg + gA in
of each grain in our columnar microstructure. the (111̄) plane. Given these two possibilities of
Using the well-known Thompson crystallographic dissociation, this pure screw dislocation can pro-
notation [44] to represent the possible combi- duce cross-slip between the (111) and (111̄) planes.
nations of dislocations on these glide planes (Fig. Because its Burgers vector is along the columnar,
2), we identify these two types as following. z axis and, as defined in Section 2, perpendicular
The first type consists of the 60° CA and CB to the deformation plane, the BA screw dislocation
dislocations on the (111) plane and the 60° DA and does not contribute to the deformation and, thus,
DB dislocations on the (111̄) plane. They all dis- its nucleation is a rare event in our simulations.
sociate into a pure-edge partial, Cd or Dg combined Nevertheless, because of its dissociation into par-
with a 30° partial, dX or gX, where X is either the tials which do have a component in the defor-
mation plane, the BA dislocation can help in
accommodating a local shear stress (on the scale
of its splitting distance) in the [1̄1̄2] or [112] direc-
tions for the (111) and (111̄) planes, respectively.
Such an event is captured in the two snapshots
in Fig. 3. The figure shows a hexagonal grain of
45 nm size (grain 2 in Fig. 1) transected by a defor-
mation twin (the region between the two parallel
lines of hcp atoms starting from the triple junction
on the left and ending at the GB on the right side
of the grain). Parallel to this twin, but slightly
below it, an intrinsic SF (seen as two adjacent hcp
planes) originates at the GB on the right and ends
Fig. 2. Thompson notation in the unfolded Thompson’s tetra- with a single Shockley partial near the center of
hedron [44] for the dislocation–dissociation reactions possible the grain. Ahead of this SF, a moving extended
on the (111) and (111̄) glide planes of our columnar fcc micro- dislocation is marked by the circle. Whereas in Fig.
structure. Both the Thompson notation (such as CA for Burgers 3a this dislocation, being dissociated on the (111)
vector of a perfect dislocation, and Ad for Burgers vector of a plane (parallel to the twin) is gliding towards the
partial dislocation) and the corresponding set of indices for the
same directions are presented. The sense of the directions are left GB, in Fig. 3b taken only 2 ps later, this dislo-
indicated by [ ⬎. The dislocation lines are always restricted to cation has changed its plane of dissociation to
be along BA which is the columnar, z, axis in our microstruc- (111̄) and is now approaching the twin. As shown
ture. The deformation takes place in the x–y plane (see also in the inset in Fig. 3, this process is possible only
Fig. 1) which here, for clarity, is shown as a plane perpendicular because the total Burgers vector of the dissociated
to BA and crossing it in the middle. The dashed gd line con-
necting the middle points of the two Thompson triangles rep- dislocation BA is along the line common to the
resents the 1/6[110] stair-rod dislocation. The directions seen as two slip planes, i.e. the columnar direction. This
underneath the deformation plane are also given as dashed lines. allows for the dissociation to take place on both
4140 V. Yamakov et al. / Acta Materialia 51 (2003) 4135–4147
1
dC ⫹ Dg→dD / Cg ⫽ [110]. (5)
3
The core of this dislocation is visualized by the
disordered atoms (shown as open circles) at the
connecting point of the two SFs in Fig. 4a, forma-
tion (3); this core is larger than the core of the
Cottrell–Lomer lock (formation (2)) because of the
twice larger Burgers vector of this dislocation,
leading to an about four times larger stored energy.
Formation (4) in Fig. 4a is the product of a more
complicated reaction between a partial dislocation
and a SF which will be analyzed in Section 5.
Fig. 5. Successive snapshots taken at 2 ps time intervals, illustrating the process of splitting of a stacking fault by an approaching
dislocation (marked by the circled region in both snapshots, see text). The process has taken place in grain 1 from Fig. 1 at about
5% strain. The atoms in (a) and (b) are marked as in Fig. 3a,b, and in the closer views (c) and (d) are marked as in Fig. 4a. The
distance between the incoming dislocation and the stacking fault is also given in (a) and (b).
at the points where the extrinsic SFs transform into employed here, have carefully estimated the
intrinsic ones). Fig. 5c further captures the moment resolved shear stress required to split a SF to be
when the incoming dislocation crosses the intrinsic 930 MPa. This value is consistent with our esti-
SF and starts to split it into two fragments. In the mate for the resolved shear stress on the (111)
next stage, revealed in Fig. 5d, the splitting of the planes (see Section 2). This confirms that the split-
intrinsic SF is completed, producing two more par- ting of the SF is not due to a dislocation reaction
tial dislocations gliding apart. Thus the perfect between the incoming dislocation and the SF (as
crystal lattice is restored, allowing the incident dis- Fig. 5c may suggest), but due to a long-range elas-
location to pass through. tic–field interaction. The fact that the incoming dis-
More careful analysis show that the process of location preserves its structure after the interaction,
splitting of a SF can take place even without the supports this conclusion.
help of an incident dislocation if the resolved shear
stress on the SF plane is large enough. Recently
Bulatov et al. [45], using an atomistic model of a
Frank–Read source and the same EAM potential
V. Yamakov et al. / Acta Materialia 51 (2003) 4135–4147 4143
remove (Fig. 6a) or add (Fig. 6b) one lattice plane location, Ag, gliding on the SF plane (see Eq. (6)).
to the twinned region. As seen in Fig. 4a, in its glide Ag has started to
The combination of gd and the remaining partial transform the intrinsic SF into an extrinsic SF in
dC is sometimes considered as a dissociated Frank the same way as a twinning dislocation broadens
partial dislocation, gC [56]. Recent HRTEM obser- a twin (see Fig. 6b). Finally, the gliding partial has
vations by Medlin et al. [54] and Foiles and Medlin joined the partial ending the SF; together they have
[30] revealed that dissociated Frank dislocations at formed the double Shockly partial seen at the bot-
the coherent twin boundaries, as seen in Fig. 6a,b, tom end of formation (4).
are common in Al. Moreover, using atomistic This example illustrates that the reactions
simulations they showed that an initial compact between dislocations and a coherent twin boundary
Frank partial at the twin boundary is unstable as it may take place also between a dislocation and a
dissociates into a stair-rod and a Shockley partial SF, resulting again in an intrinsic–extrinsic trans-
dislocations separated by a SF from the twin formation of the SF or in cutting off an intrinsic
boundary. At elevated temperatures or in the pres- SF. However, this type of a cutting off of a SF
ence of vacancies this dissociated couple can climb (not shown here) is different from the one shown
along the twin boundary and grow the twin by a in Fig. 5, in that the incident dislocation will form
non-conservative, thermally activated process a stair-rod lock with one piece of the SF while the
[30,54]. other piece, together with the newly formed partial,
Our simulations show that dissociated Frank Ag, forms a perfect dislocation which glides away.
partials can appear from the interaction between a The type of the resulting stair-rod dislocation
gliding perfect dislocation and a twin boundary depends on the type of incident Shockley partial.
(see Eq. (6) and Fig. 6a,b) and actually activate the Eq. (6), which provides a gd dislocation, describes
twinning–untwinning processes by the conserva- only one possible reaction. Another possibility
tive glide of the emitted twinning dislocation—a arises if the leading partial was dC instead of Ad
possibility recently analyzed by Müllner and (i.e. a pure-edge instead of a 30° partial), leading
Solenthaler [53]. A subsequent climb of the to the reaction:
remaining dissociated Frank partial has not been
AC→Ad ⫹ dC→Ag ⫹ dD / Cg ⫹ gD. (7)
observed. However, this is not surprising in our
simulations because: (1) we do not have sources As discussed in connection with Eq. (4), the stair-
of vacancies that can facilitate the climb at low rod, dD/Cg, has a higher stored excess energy; i.e.
temperatures as in the TEM observations of Foiles such a reaction is energetically less favorable than
et al. [54], where they suggest that the 400 KeV the one in Eq. (6). In fact, our simulations indicate
beam is able to produce the necessary vacancies to that a pure-edge partial is usually repelled from the
assist the climb; (2) the time scale of 5 min per twin boundary and that, therefore the above reac-
micrograph during which climb by a few nm is tion (Eq. (7)) takes place only under high stress.
observed gives a climb rate too slow to be repro- An example of the second type of dislocation–
duced in an MD simulation where the time scale twin reaction, which results in a cross-slip event
is in the range of 1 ns. at the twin boundary without leaving a stair-rod
Returning to Fig. 4a, we can now explain forma- dislocation behind, is shown in Fig. 7a,b. In this
tion (4) in terms of a dissociated Frank partial as case the incident dislocation (see Fig. 7a) has a
expressed by Eq. (6). In this case the leading par- Burgers vector parallel to the twin plane. Cross-
tial, Ad of the approaching dissociated dislocation slip at the twin boundary transforms the incident
CA (one on the right in formation (4)) reacts with dissociated dislocation into a dissociated dislo-
the SF of the other dissociated dislocation, instead cation in the twin plane. This creates two twinning
with one of its partials (as in formations (2) and dislocations gliding in opposite directions (Fig.
(3) when a stair-rod lock is formed). Reacting with 7b). The result is a twinning or untwinning process
the SF, the incident partial, Ad, has been dis- (depending on the sense of the Burgers vectors) in
sociated into a stair-rod, gd, and another partial dis- which the incident dissociated dislocation is fully
V. Yamakov et al. / Acta Materialia 51 (2003) 4135–4147 4145
between coarse-grained and nanocrystalline [21] Yamakov V, Wolf D, Phillpot SR, Gleiter H. Acta
materials is expected to be only qualitative. In prac- Mater 2002;50:5005.
[22] Keblinski P, Wolf D, Gleiter H. Interface Science
tice, the yield stress of nanocrystalline metals is 1998;6:205.
much higher, following the Hall–Petch relation. [23] Yamakov V, Wolf D, Phillpot SR, Gleiter H. Acta
Mater 2002;50:61.
[24] Wang Y, Chen M, Zhou F, Ma E. Nature 2002;419:912.
Acknowledgements [25] Rao S, Parthasarathy TA, Woodward C. Phil Mag A
1999;79:1167.
VY, DW and SRP are supported by the US [26] Vegge T, Rasmussen T, Leffers T, Pedersen OB, Jacobsen
Department of Energy, BES-Materials Science KW. Phil Mag Lett 2001;81:137.
under contract W-31-109-Eng-38. VY is also [27] Zhou SJ, Preston DL, Lomdahl PS, Beazley DM.
grateful for support from the DOE/BES Compu- Science 1998;279:1525.
[28] Swaminarayan S, LeSar R, Lomdahl P, Beazley D. J
tational Materials Science Network (CMSN). We
Mater Res 1998;13:3478.
are grateful for grants of computer time on the [29] Bulatov V, Abraham FF, Kubin L, Devincre B, Yip S.
Cray-T3E at the John von Neumann Institute for Nature 1998;391:669.
Computing in Jülich, Germany and on the Chiba [30] Foiles SM, Medlin DL. Mat Sci Eng A 2001;319-321:102.
City Linux cluster at ANL. [31] Hirth JP, Lothe J. Theory of dislocations, 2nd edn. John
Wiley & Sons Inc, 1992.
[32] Weertman J, Weertman JR. Elementary dislocation theory.
References New York, Oxford: Oxford University Press, 1992.
[33] Keblinski P, Phillpot SR, Wolf D, Gleiter H. Phil Mag
[1] Birringer R, Gleiter H, Klein HP, Marquardt P. Phys Lett A 1999;79:2735.
A 1984;102:365. [34] Keblinski P, Phillpot SR, Wolf D, Gleiter H. Phys Rev
[2] Gleiter H. Acta Mater 2000;48:1. Lett 1996;77:2965.
[3] Karch J, Birringer R, Gleiter H. Nature 1987;330:556. [35] Wolf D. Grain boundaries: structure. In: Cahn R, editor.
[4] Nastasi M, Parkin DM, Gleiter H, editors. Mechanical The encyclopedia of materials: science and technology.
properties and deformation behavior of materials having Pergamon Press; 2001. p. 3597.
ultra-fine microstructures. Kluwer; 1993. [36] Ercolessi F, Adams JB. Europhys Lett 1994;26:583.
[5] Siegel RW. Mater Sci Forum 1997;851:235. [37] Parrinello M, Rahman A. J Appl Phys 1981;52:7182.
[6] Yip S. Nature 1998;391:532. [38] Melchionna S, Ciccotti G, Holian BL. Mol Phys
[7] Koch CC, Suryanarayana C. Nanocrystalline materials, 1993;78:533.
chapter 6. In: Li JCM, editor. Microstructure and proper- [39] Jonsson H, Andersen HC. Phys Rev Lett 1988;60:2295.
ties of materials, vol 2. Singapore: World Scientific Pub- [40] Dahmen U, Hetherington CJD, O’Keefe MA, Westmacott
lishing; 2000. p. 380–5. KH, Mills MJ, Daw MS, Vitek V. Phil Mag Lett
[8] Nieh TG, Wadsworth J. Scripta Metall 1991;25:955. 1990;62:327.
[9] Scattergood RO, Koch CC. Scripta Metall 1992;27:1195.
[41] Ogata S, Li J, Yip S. Science 2002;298:807.
[10] Wang N, Wang Z, Aust KT, Erb U. Acta Metall Mater
[42] Coble RL. J Appl Phys 1963;34:1679.
1995;43:519.
[43] Pond R, Glass CM. In: Shewnion PG, Zackay VF, editors.
[11] Wunderlich W, Ishida Y, Maurer R. Scripta Metall
Response of metals to high velocity deformation. New
Mater 1990;24:403.
York: Interscience Publishers; 1961. p. 145.
[12] Huang JY, Wu YK, Ye HQ. Acta Mater 1996;44:1211.
[44] Thompson N. Proc Phys Soc 1953;66B:481.
[13] Islamgaliev RK, Chmelik F, Kuzel R. Mat Sci Eng ; A
1997;A234-236:335. [45] de Koning M, Cai W, Bulatov VV. Phys Rev Lett, in press.
[14] Cottrell AH. Dislocations and plastic flow in crystals. [46] Mahajan S, Chin GY. Acta Met 1973;21:173.
Oxford: Clarendon Press, 1961. [47] Mahajan S, Chin GY. Acta Met 1974;22:1113.
[15] Morris DG, Morris MA. Acta Metall Mater 1991;39:1763. [48] Remy L. Scripta Met 1977;11:169.
[16] Schiotz J, Di Tolla FD, Jacobsen KW. Nature [49] Remy L. Metall Trans A 1981;12A:387.
1998;391:561. [50] Hartley CS, Blachon DLA. J Appl Phys 1978;49:4788.
[17] Schiotz J, Di Tolla FD, Jacobsen KW. Phys Rev B [51] Hartley CS. Mat Sci Eng 1986;81:543.
1999;60:11971. [52] Solenthaler C. Mat Sci Eng A 1990;125:57.
[18] Swygenhoven HV, Spaczer M, Caro A. Acta Mater [53] Mullner P, Solenthaler C. Mat Sci Eng A 1997;230:107.
1999;47:3117. [54] Medlin DL, Carter CB, Angelo JE, Mills MJ. Phil Mag
[19] Yamakov V, Wolf D, Salazar M, Phillpot SR, Gleiter H. A 1997;75:733.
Acta Mater 2001;49:2713. [55] Rusakov GM, Litvinov VS. Phys Met Metall 2000;89:445.
[20] Yamakov V, Wolf D, Phillpot SR, Mukherjee A, Gleiter [56] Frank FC. Proc Phys Soc 1949;62A:202.
H. Nat Mater 2002;1:45. [57] Ashby MF. Phil Mag 1970;21:399.