You are on page 1of 6

Scripta mater.

42 (2000) 1025–1030
www.elsevier.com/locate/scriptamat

ON THE GRAIN SIZE SOFTENING IN NANOCRYSTALLINE


MATERIALS
Hans Conrad and Jagdish Narayan
Materials Science and Engineering Department, North Carolina State University,
Raleigh, NC 27695-7907, USA
(Received December 17, 1999)
(Accepted December 29, 1999)

Keywords: Grain boundaries; Thermally activated; Hardness; Atomic

Introduction

Nanocrystalline materials with grain sizes in the range of 1–100 nm have been found to exhibit novel
and often improved physical properties [1– 4]. They are thus of considerable interest from both
scientific and technological viewpoints. In the present paper we consider the influence of grain size on
the flow stress (or corresponding hardness) of nanocrystalline materials, with focus on the grain size
softening (i.e., the so-called negative or inverse Hall-Petch effect) which has been observed at grain
sizes ⬍⬃50 nm [5–12].
The effect of grain size on the hardness of nanocrystalline materials has been reported to have the
form shown schematically in Fig. 1 [5–15]. At the larger grain sizes the hardness increases with
decrease in grain size according to the well-known Hall-Petch relation (with the possibility that the
Hall-Petch constant kH-P may decrease as a critical grain size dc is approached). At the critical grain size
dc the hardness either remains relatively constant with further reduction in grain size, or it decreases,
i.e., grain size softening occurs. The implication by some investigators is that the grain size softening
also obeys a d⫺1/2 relationship, giving a so-called “negative or inverse Hall-Petch effect”. We take issue
with the concept that the grain size softening fits a negative Hall-Petch d⫺1/2 relationship, but rather
develop an alternate mechanism and expression.
Referring to Fig. 1, the Hall-Petch behavior at d ⬎ dc has been explained by several dislocation
models: (a) the grain size work hardening model by Conrad [16 –18], (b) the grain boundary ledge
model by Li [19] and (c) the geometric accommodation and statistically stored dislocation model by
Ashby [20]. As shown by the composite model proposed by Meyers and coworkers [21, 22] these
dislocation models can lead to a decrease in kH-P as the grain size approaches the nanometer size scale.
However, they do not give a negative slope. In contrast, the line tension model by Koch and Scattergood
[23] gives first a reduction in kH-P as the grain size is decreased and then a reversal as the grain size
approaches the effective dislocation core cut-off radius. It thus appears that their model is able to
explain the behavior over the entire range of grain sizes depicted in Fig. 1. However, their model is
based on dislocation motion being responsible for plastic flow both at grain sizes larger and smaller than
dc. This differs from the computer simulation results of Shiotz et al [24] and Swygenhoven et al [25],
which give that no dislocation activity occurs below a critical grain size (of the order of 10 nm for
metals) and that plastic deformation then occurs entirely by grain boundary sliding. Of related interest
are the theoretical predictions of the limiting grain size for dislocation activity. Taking the theoretical
1359-6462/00/$–see front matter. © 2000 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved.
PII: S1359-6462(00)00320-1
1026 GRAIN SIZE SOFTENING Vol. 42, No. 11

Figure 1. Schematic of the variation of hardness H with grain size d.

shear strength ␶th to be between 1/4 and 1/16 the shear modulus ␮ [26, 27], one can obtain from the
following the limit to dislocation activity as the grain size is reduced: (a) the dislocation pile-up model
gives d ⬇ ␲b/(␶/ ␮) for n ⫽ 1, the number in the pile-up; (b) the elastic interaction stress between
dislocations gives d ⫽ b/[(2␲(1 ⫺ ␷)(␶/␮)]; and (c) Orowan bowing gives d ⫽ b/(␶/␮). The grain size
limit to dislocation activity obtained taking ␶ ⫽ ␶th is similar in magnitude to that indicated in the
computer simulations [26, 27].
Coble creep [5, 28], grain boundary sliding [8, 24, 25, 29] and grain boundary triple junction activity
[30] have been proposed to account for the plastic deformation without dislocation motion at d ⬍ dc and
the decrease in flow stress with decrease in grain size. In the case of Coble creep, the inverse cube
dependence of the strain rate on the grain size has not been experimentally established and moreover
the observed creep rate in nanocrystalline Cu is two orders of magnitude larger than that predicted [10].
No detailed quantitative model has been developed for triple junction activity. Swygenhoven and Caro
[31] have developed a non-linear viscous grain boundary sliding model to explain their computer
simulation results. However, the strains considered occur in picoseconds. Further, their model is lacking
in details and has not been applied to actual experimental data. The grain boundary sliding model by
Hahn and Padmanabhan [29] is largely phenomenological and contains many adjustable parameters. In
the present paper we present an atomistic model for the grain boundary sliding which can account for
the grain size softening which occurs at d ⬍ dc. The model draws on the computer simulation results
[24, 25, 31], but compares the predicted behavior with experimental data.

Proposed Model

The influence of grain size on the hardness H(⬇3␴, the flow stress) at d ⬎ dc is here considered to
represent the motion of dislocations and is given by the Hall-Petch equation
H ⫽ Hi ⫹ kH-Pd⫺1/2 (1)
where Hi is the lattice friction stress and kH-P the Hall-Petch constant. Dislocation mechanisms which
can lead to this equation are given in [17–20]. For d ⬍ dc, the computer simulations [24, 25, 31] reveal
that dislocation activity in Cu and Ni ceases below a critical grain size dc ⬇ 10 nm and grain boundary
shear becomes the deformation mode, giving grain size softening. The grain boundary shear was
associated with many independent, atomic shear events. Employing the concept of thermally-activated
shear, we propose that the macroscopic shear rate produced by the independent, atomic shear events at
the grain boundaries is given by
Vol. 42, No. 11 GRAIN SIZE SOFTENING 1027

Figure 2. Schematic of the force vs. distance curve for stress-assisted, thermally activated motion of atoms in the grain boundary.
Single hatched area is the work by the effective stress ⌬W ⫽ ␶b2x*; double hatched area is the Gibbs free energy of activation
⌬G ⫽ ⌬F ⫺ ⌬W; the combined single and double hatched area is the Helmholtz free energy of activation (⌬F ⫽ ⌬U ⫺ T⌬S.

␥˙ ⫽ NvAb␷ exp 冋⫺⌬G(␶e)


kT 册 (2)

where Nv is the number of places per unit volume where thermally-activated shear can occur, A the area
swept out per successful thermal fluctuation, b the atomic diameter, ␷ the frequency of vibration and ⌬G
the Gibbs free activation energy, which is a decreasing function of the effective shear stress ␶e ⫽ ␶ ⫺
␶䡩, where ␶ is the applied stress and ␶䡩 a back stress or threshold stress. Reasonable expansions for the
parameters on the right side of Eqn. 1 are: Nv ⫽ ␦/db3, where ␦ ⬇ 3b is the grain boundary width, A ⫽
b2, ␷ ⫽ ␷D(⬇1013 s⫺1) the Debye frequency and ⌬G ⫽ ⌬F ⫺ v␶e, where ⌬F is the Holmholtz free
energy (which may decrease with stress; see Fig. 2) and v ⫽ b3 the activation volume. It is expected
that the value of ⌬F will be approximately that for an atom-vacancy interchange in the lattice or grain
boundary diffusion Qb [32–34]. Inserting the above values into Eqn. 2 and considering both forward and
backward jumps gives

␥˙ ⫽
6b␷D
d
sinh 冉 冊 冉 冊
v␶e
kT
exp ⫺
⌬F
kT
(3)

where sinh x ⬇ x for x ⱕ 0.5 and sinh x ⬇ 1/2 exp x for x ⱖ 2. Eqn. 3 thus gives that ␶e is proportional
to d for small values of v␶e/kT and varies as ln d with large values. Further, dc can be obtained by
equating Eqns. 1 and 3 at constant ␶e. Again, the value of ⌬F should be approximately that for grain
boundary diffusion.

Comparison with Experimental Data.

The validity of Eqn. 3 was checked against available experimental data [5–9, 11, 12, 25, 35–37],
focusing on the values of the activation volume v and the Helmholtz free activation energy ⌬F. It was
assumed that ␶ ⫽ ␴/2 ⫽ H/6 (where ␴ is the tensile stress and H the hardness), the shear rate ␥˙ ⫽ 10⫺3
s⫺1 and ␶e ⬇ ␶, i.e., ␶䡩 ⫽ 0. When hardness or flow stress data were available for d ⬍ dc, the value of
v was derived from the slope of the best fit straight line in a plot of ␶ vs. log d; see for example Fig.
3. In all cases v␶/kT ⱖ 2 so that sinh v␶/kT became the exponential. The value of ⌬F was then derived
from the intercept of this plot, or alternately by inserting the experimental values of ␶ce, dc and the
derived or assumed value of v into Eqn. 3. Reasonable agreement occurred between the two values of
⌬F so obtained. In the case of Cu and Ni, when no hardness data were available for d ⬍ dc, the value
of dc provided by the computer simulations was employed along with the experimental Hall-Petch data
to give the critical stress ␶ce. The value of ⌬F was then derived using Eqn. 3 and assuming v ⫽ b3.
1028 GRAIN SIZE SOFTENING Vol. 42, No. 11

Figure 3. Hardness vs. log d in the grain size softening regime. Data from: Cu [5, 36]; Pd [5]; Ni-P [6, 7]; TiAl [8, 9], Nb77Al23
[11];NiAl3 [12].

The values of v and ⌬F obtained from the experimental data are presented in Table 1. To be noted
is that for the metals Ni and Pd and the one set of data for Cu the values of v and ⌬F are in rather good
agreement with those expected, namely v ⬇ b3 and ⌬F ⬇ Qb. This also appears to be the case for Ni-P.
In the case of the intermetallic compounds TiAl, Nb77Al23 and NbAl3 and the other Cu [36], the derived
value of v is smaller than b3; moreover ⌬F is below that expected for grain boundary diffusion.

Discussion

The agreement between the derived and expected values of v and ⌬F obtained for the metals provides
support that the plastic deformation kinetics at d ⬍ dc is described by Eqn. 3, and that grain boundary
sliding by single atomic shear events is the operating mechanism. This is in accord with the computer
simulation results [23, 24, 31]. The lower values of v and ⌬F for the intermetallics and the one set of
experiments on Cu [36] could result from the high values of the applied stress in these cases and the
form of the force-distance barrier for the atomic displacements; see for example Fig. 2. The higher

TABLE 1
Comparison of Experiment with Theory
␶ec, (GPa) dc (nm) v/b3** ⌬F(kJ/mole)

Data
Materials Expt’l Ref. Expt’l Ref. Value Method Ref. ⌬F Qgb*****

Cu 0.67 31 8**** 25 1.0 Assumed — 93.4 80–100


12.5 32 7 36 0.25 ␶ vs.Ln d 36 87.4 80–100
— — — — 1.2 ␶ vs.Ln d 5 90.8 80–100
Ni 1.4 33 11**** 25 1.0 Assumed — 97.5 106–129
Pd — — — — 0.75 ␶ vs.Ln d 5 90.3 88–106
Ni-P 1.13 7 8 7 1.00 ␶ vs.Ln d 7 97.5 ?
— — — — 0.96 ␶ vs.Ln d 6 100.9
TiAl 1.96 9 23 9 0.24 ␶ vs.Ln d 9 87.8 ?
TiAl — — — — 0.18 ␶ vs.Ln d 8 86.8
(243K) 3.17 9 16 9 0.09 ␶ vs.Ln d 9 86.3
Nb77Al2 1.75 12 35 12 0.58 ␶ vs.Ln d 12 91.7 ?
NbAl3 1.2 11 63 11 0.23 ␶ vs.Ln d 11 85.0 ?

* ␶ec ⫽ shear stress at d ⫽ dc; ** b ⬇ 2.5 Å was taken for the intermetallic compounds; *** ⌬F was derived using Eqn. 3;
**** Computer simulation; ***** From Refs. 32 and 33.
Vol. 42, No. 11 GRAIN SIZE SOFTENING 1029

hardness of this Cu compared to that in [5] could be due to the method by which it was prepared [36],
which introduced tungsten into the grain boundaries.
In the present model the critical grain size dc represents the change from Hall-Petch dislocation
plastic flow at d ⬎ dc, to grain boundary sliding (which becomes easier) at smaller grain sizes. The
change may not be as sharp as indicated by the solid lines in Fig. 1. Rather, it is expected that there will
be a transition region where both dislocation motion and grain boundary sliding will contribute to the
plastic deformation, giving a curve similar to that shown by the dashed line. Also, it is assumed in the
present model that the grain size is uniform with a random orientation and the material is relatively free
of porosity or other imperfections. This is often not the case for real nanocrystalline materials and one
might therefore expect a transition region and in the extreme a departure from the predictions of our
model.

Summary and Conclusions

A thermally-activated grain boundary shearing model based on computer simulation results is devel-
oped for the grain size softening (the so-called inverse or negative Hall-Petch effect) which has been
reported for nanocrystalline materials. The predictions of the model are in accord with experimental
data.

Acknowledgement

The authors wish to thank Dr. Di Yang for assistance in preparation of the manuscript and for partial
support by the NSF Center for Advanced Materials and Smart Structures.

References

1. J. Narayan, Y. Chen, and R. Moon, Phys. Rev. Lett. 46, 1491 (1981).
2. J. Narayan and Y. Chen, Phil. Mag. A119, 475 (1984); U.S. Patent No. 4,376,455 (March 15, 1983).
3. J. Narayan, Y. Chen, and K. Tsang, Phil. Mag. A55, 807 (1987).
4. R. Siegel, E. Hu, D. Cox, H. Goronkin, L. Jelinski, and C. Koch, Nanostructure Science and Technology, International
Technology Research Institute, NSF (1998).
5. A. Chokshi, A. Rosen, J. Karch, and H. Gleiter, Scripta Metall. 23, 1679 (1989).
6. K. Lu, W. Wei, and J. Wang, Scripta Metall. Mater. 24, 2319 (1990).
7. D. Ostrander and U. Erb, Scripta Metall. (1992); G. McMahon and U. Erb, Microstr. Sci. 17, 447 (1989).
8. H. Chang, H. Höfler, C. Altstetter, and R. Averbach, Scripta Metall. Mater. 25, 1161 (1991).
9. C. Alstetter, in Mechanical Properties and Deformation Behavior of Materials Having Ultra-Fine Microstructures, M.
Nastasi, D. Parkin, and H. Glaitar, p. 381, Kluver Academic Publishers, Dordrecht, the Netherlands (1993).
10. G. Nieman and J. R. Weertman, J. Mater. Res. 6, 1012 (1991).
11. D. Kim and K. Okazaki, Mater. Sci. Forum 88 –90, 553 (1992).
12. K. Okazaki, Proc. Thermac ’97 (1997).
13. A. El-Sherik, U. Erb, G. Palumbo, and K. Aust, Scripta Metall. Mater. 27, 1185 (1992).
14. M. Furukawa, Z. Horita, M. Nemoto, R. Valiev, and T. Langdon, Acta Mater. 44, 4619 (1996).
15. C. Sunyanarayan, M. Mukhopadhyay, S. Patankar and F. Froes, J. Mater. Res. 7, 2114 (1992).
16. H. Conrad, in Electron Microscopy and Strength of Crystals, ed. G. Thomas and J. Wasburn, p. 299, Interscience, New York
(1961).
17. H. Conrad, Acta Metall. 11, 75 (1963).
18. H. Conrad, in Ultrafine-Grain Metals, ed. J. Burke and V. Weiss, p. 213, Syracuse University Press, Syracuse, NY (1970).
19. J. C. M. Li, Trans. TMS-AIME. 227, 239 (1963).
20. M. F. Ashby, Phil. Mag. 21, 399 (1970).
21. M. Meyers and E. Ashworth, Phil. Mag. A46, 737 (1982).
1030 GRAIN SIZE SOFTENING Vol. 42, No. 11

22. M. Meyers, D. Benson, and H.-H, Fu, in Advanced Materials for the 21st Century, ed. Y.-W. Chung, D. Dunand, P. Liaw,
and G. Olson, p. 499, TMS, Warrendale, PA (1999).
23. R. Scattergood and C. Koch, Scripta Metall. Mater. 27, 1195 (1992).
24. J. Schiotz, F. Di Tella, and K. Jacobson, Nature 391, 561 (1988).
25. H. van Swygenhoven, M. Spaczer, and A. Caro, Acta Mater. 47, 3117 (1999).
26. A. H. Cottrell, Dislocations and Plastic Flow in Crystals, Oxford University Press (1953).
27. W. R. Tyson, Phil. Mag. 14, 925 (1966).
28. B. Cai, Q. Kong, L. Liu and K. Lu, Scripta Mater. 41, 755 (1999).
29. H. Hahn and K. A. Padmanabhan, Phil. Mag. B.76, 559 (1997).
30. G. Palumbo, U. Erb, and K. Aust, Scripta Metall. Mater. 24, 2347 (1990).
31. H. van Swygenhoven and A. Caro, Appl. Phys. Lett. 71, 1652 (1997); Phys. Rev. 58, 11246 (1998).
32. N. A. Gjostein, Diffusion, p. 241, ASM Metals Park, OH (1973).
33. P. Shewman, Diffusion in Solids, 2nd edn, TMS, Warrendale, PA (1989).
34. J. Narayan, J. Appl. Phys. 53, 8607 (1982).
35. J. Weertman, D. Farkas, K. Hemker, H. Kung, M. Mayo, R. Mitra, and H. van Swygenhoven, MRS Bull Feb., p. 44, (1999).
36. J. Narayan, unpublished research, N.C. State University (1999).
37. G. Hughes, S. Smith, C. Pande, H. Johnson, and R. Armstrong, Scripta Metall. 20, 93 (1986).

You might also like