Sunteți pe pagina 1din 11

Available online at www.sciencedirect.

com

Acta Materialia 61 (2013) 4887–4897


www.elsevier.com/locate/actamat

Sluggish diffusion in Co–Cr–Fe–Mn–Ni high-entropy alloys


K.-Y. Tsai a, M.-H. Tsai b, J.-W. Yeh a,⇑
a
Department of Materials Science and Engineering, National Tsing Hua University, Hsinchu 30013, Taiwan
b
Department of Materials Science and Engineering, National Chung Hsing University, Taichung 40227, Taiwan

Received 27 March 2013; accepted 28 April 2013


Available online 4 June 2013

Abstract

Sluggish diffusion kinetics is an important contributor to the outstanding properties of high-entropy alloys. However, the diffusion
kinetics in high-entropy alloys has never been probed directly. Here, the diffusion couple method was used to measure the diffusion
parameters of Co, Cr, Fe, Mn and Ni in ideal-solution-like Co–Cr–Fe–Mn–Ni alloys. These parameters were compared with those in
various conventional face-centered cubic metals. The results show that the diffusion coefficients in the Co–Cr–Fe–Mn–Ni alloys are
indeed lower than those in the reference metals. Correspondingly, the activation energies in the high-entropy alloys are higher than those
in the reference metals. Moreover, the trend of the normalized activation energy is positively related to the number of composing ele-
ments in the matrix. A quasi-chemical model is proposed to analyze the fluctuation of lattice potential energy in different matrices
and to explain the observed trend in activation energies. Greater fluctuation of lattice potential energy produces more significant atomic
traps and blocks, leading to higher activation energies, and thus accounts for the sluggish diffusion in high-entropy alloys.
Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Keywords: High-entropy alloy; Multicomponent diffusion; Diffusion mechanism; Lattice potential energy fluctuation

1. Introduction lattice distortion and cocktail effects [3]. Among these


effects, the sluggish diffusion effect is very important,
The design of conventional alloys is based on one or two because it leads to exceptional high-temperature strength
principal elements, and some minor elements are added to [11,12], impressive high-temperature structural stability
modify their microstructure and properties. In 1995, a [13,16] and formation of nanostructures [17–19]. More-
novel design concept called high-entropy alloys (HEA) over, in applications demanding slow diffusion kinetics,
was proposed by Yeh et al. [1–4]. HEA are defined as alloys this effect makes high-entropy materials extremely compet-
containing at least five major elements, each of which has a itive. Indeed, HEA and their nitrides have been tested in
concentration range between 5 and 35 at.%, and with high diffusion barrier applications and show remarkable effec-
mixing entropy in their liquid state or high-temperature tiveness [13,16,20–22].
solid solution state [3,4]. HEA have been reported to Despite the significant scientific interest and the techno-
possess many promising properties, such as high hardness logical values of the sluggish diffusion effect, the diffusion
[5–9], outstanding wear resistance [10], excellent high-tem- kinetics in HEA has never been probed directly. The diffu-
perature strength [11,12], good thermal stability [13] and, in sion constants and activation energies of diffusion in HEA
general, good oxidation [10] and corrosion resistance have never been documented. The mechanisms behind this
[14,15]. These outstanding properties are directly related effect also remain unknown. An important difference
to the so-called “core effects” in such multi-principal- between the diffusion in HEA and that in conventional
element design: high entropy, sluggish diffusion, severe alloys is that the surrounding atoms of each lattice site in
the solid solution phases of HEA have much greater variety
⇑ Corresponding author. Tel.: +886 3 5719558; fax: +886 3 5722366.
than those in conventional ones. This is because: (1) in HEA
E-mail address: jwyeh@mx.nthu.edu.tw (J.-W. Yeh).
there is no solvent element that dominates the compositions

1359-6454/$36.00 Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.actamat.2013.04.058
4888 K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897

of solid solution phases; (2) in HEA there are at least five where x0 is the location of Matano interface given by
principal elements in the solid solution phases. Because each Z C1
i
site is surrounded by different atoms, each site has different ðx  x0 ÞdC i ¼ 0 ð3Þ
bond configuration and thus different lattice potential C þ1
i

energy (LPE) [23]. Since vacancy formation and migration On the basis of Eq. (2), determining all the interdiffusion
enthalpies are related to local atomic interactions [23–25], coefficients requires (n  1)2 independent equations gener-
both enthalpies vary from one lattice site to another. Thus, ated from (n  1) independent diffusion couples with a
the variation in LPE can lead to changes in diffusion kinet- common composition point on their diffusion paths. The
ics. Indeed, the differences in LPE have been reported to interdiffusion coefficients can be solved at this composition
cause a local biased atom drift into a low potential region only [31]. Because the interdiffusion coefficients are usually
[26]. Furthermore, the models of random walk with random dependent on concentration, the determination of interdif-
LPE biases have been formulated to analyze one-dimen- fusion coefficients over a wide composition range is a mas-
sional diffusion in alloys [27,28]. However, studies regarding sive work. Furthermore, designing multiple diffusion
the effects of LPE fluctuation on diffusion are limited to bin- couples with diffusion paths intersecting at a common com-
ary systems. Indeed, studying the interactions between sol- position point becomes virtually unfeasible for systems
ute atoms, solvent atoms and vacancies in ternary or higher- with more than three components [32].
order systems is very challenging. In order to simplify these problems, quasi-binary cou-
In this paper, the diffusion phenomenon in a quinary ples conceptually similar to Darken-type couples [33] are
HEA is studied. Since the solutions to the five-element dif- employed. For a quinary system, if the concentration gra-
fusion equation are too complex to obtain, a quasi-binary dients of components 3, 4 and 5 at a specific position x0
approach is used to simplify this problem. Moreover, to are negligibly small, in other words oC1/ox ffi oC2/ox,
avoid the complexity arising from two or more phases, it then Eq. (2) can be simplified to
is important that the quinary alloy system should remain Z C0
e 2 @C 1
1
single-phased in the composition and temperature range ðx0  x0 ÞdC 1 ¼ 2t  D ð4Þ
11
involved in this study. Based on this consideration, the þ1
C1 @x
ideal-solution-like face-centered cubic (fcc) Co–Cr–Fe–
Mn–Ni alloy system is selected [29]. The diffusion parame- This is the same as the equation for a binary system and
ters of each composing element in this HEA are determined can be calculated easily. In order to eliminate the errors
from different quasi-binary diffusion couples. They are also associated with the determination of the location of the
compared with the diffusion parameters of these elements Matano plane, Eq. (4) has been modified by the Sauer–Fre-
in conventional fcc metals. Moreover, to understand the ise method without using the position of the Matano plane
effect of LPE fluctuation on diffusion, a quasi-chemical [34,35]:
" Z x0 Z þ1 #
model is used to analyze the distribution of local interac- e 0Þ ¼
DðC
1
ð1  Y Þ ðC 0  C  Þdx þ Y ðC þ  C 0 Þdx ð5Þ
tion energy in the present HEA and the reference metals. 2tðdC=dxÞx0 1 x0

The effects of the number of principal elements on diffusion where


are revealed and discussed.
C0  C

2. Theoretical background: determination of diffusion Cþ  C0
coefficients from quasi-binary diffusion couples Thus, the quasi-binary interdiffusion coefficients can be
evaluated using Eq. (5).
For interdiffusion in an n-component system, Fick’s sec- It should be mentioned that there is only one interdiffu-
ond law is extended to (n  1) independent partial differen- sion coefficient in a binary system, since the interdiffusion
tial equations and produces (n  1)2 interdiffusion fluxes of the two elements are counterbalanced. In the
coefficients D e n [30,31]:
ij quasi-binary approach, however, two different interdiffu-
Xn1   sion coefficients would be obtained from the concentration
@C i @ e n @C j
¼ D ij ði ¼ 1; 2; . . . ; n  1Þ ð1Þ profiles of components 1 and 2. First, although the differ-
@t j¼1
@x @x
ence between the concentration gradients of components
in which concentration Ci is a function of time t and posi- 1 and 2 at position x0 is negligibly small, the difference
tion x. To estimate the interdiffusion coefficients from Eq. between the concentration profiles could be enlarged as a
(1), Kirkaldy [31] extended the Boltzmann–Matano meth- result of the integration operation in Eq. (5). Secondly,
od from binary to multicomponent form: the total diffusion fluxes of the other three components
Z Ci X
n1 are probably unequal to zero and cause the interdiffusion
ðx  x0 ÞdC i ¼ 2t e n @C j ði ¼ 1; 2; . . . ; n  1Þ
D fluxes of components 1 and 2 uncounterbalanced. As a
ij
þ1
Ci j¼1
@x result, two different interdiffusion coefficients in a single
ð2Þ quasi-binary couple are obtained:
K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897 4889

e2 @C 1 e e 1 @C 2 li ¼ l0 þ RT ln ai ð9Þ
D 11 ¼ J 1–  e
J2 ¼ D 22 ð6Þ
@x @x and Ci = Xi/Vm, where Xi is the atomic fraction of element
i, and Vm is the molar volume of alloy, the relation between
The calculated interdiffusion coefficients could be intrinsic and tracer diffusion coefficients can be obtained:
regarded as intrinsic diffusion coefficients at the Kirkendall  
 @ ln ai  @ ln ci
interfaces, owing to the negligible marker movement in the Dii ¼ Di ¼ Di 1 þ ð10Þ
@ ln X i @ ln X i
present quasi-binary couples, which will be shown in Sec-
tion 4.1. From the diffusion theory, the intrinsic diffusion where the activity coefficient ci = ai/Xi. This equation dem-
is observed relative to the lattice frame of reference, onstrates that the difference between tracer and intrinsic
whereas the interdiffusion is observed relative to the labo- diffusion coefficients comes from the thermodynamic
ratory frame of reference. The intrinsic diffusion coeffi- nonideality of solution.
cients can be related to interdiffusion coefficients by Based on the thermodynamic analysis by Otto et al. [29],
measuring the marker drift velocity. Since the marker the equimolar CoCrFeMnNi alloy is a near-ideal solid
movement can be neglected in the present couples, there solution at >1123 K. In consideration of the fact that only
is no detectable difference between intrinsic diffusion and two binary pairs (Co–Mn and Mn–Ni) give greater contri-
interdiffusion. butions to the deviation from the ideal solution, it is
However, the direct comparison of one element diffusing reasonable that the present quasi-binary couples with
in different alloys should be based on tracer diffusion exper- CoCrFeMn0.5Ni composition at the Kirkendall interface
iments. In order to reveal the correlation of the present dif- should also behave like an ideal solution at the diffusion
fusion coefficients with tracer diffusion coefficients, a well- temperatures, owing to the reduction of Mn concentration.
known approach to relating intrinsic and tracer diffusion This leads to the equality between the tracer and the intrin-
coefficients is described in the following. By assuming that sic diffusion coefficients in the CoCrFeMn0.5Ni HEA. In
vacancies are in equilibrium in the system and that the con- other words, the calculated quasi-binary interdiffusion
tributions of vacancy wind and correlation effects are coefficients have approximately the same values as tracer
neglected, the non-diagonal elements in the matrix of diffusion coefficients, which allow meaningful comparisons
Onsager’s phenomenon diffusion coefficients Lij can be between the present results and other tracer-based diffusion
eliminated, and thus the intrinsic diffusion flux Ji can be results.
written as [36,37]
3. Experimental procedures
@l X
n1
@C j
J i ¼ Lii i ¼  Dij ð7Þ
@x @x In this study, the diffusion couples were designed so that
j¼1
only two of the five composing elements had a concentra-
Moreover, Darken [38] and LeClaire [39] derived the rela- tion gradient across the interface. Each quasi-binary diffu-
tion between tracer diffusion coefficient D* and Lij: sion couple was used to probe the diffusion of two
elements. Table 1 shows the designation of three diffusion
C i Di couples and the compositions of the end members. All
Lii ¼ ð8Þ
RT the end members have a stable fcc structure, which is con-
For the quasi-binary diffusion couples, as mentioned ear- firmed from their X-ray diffraction patterns (not shown), in
lier, the intrinsic flux J1 is only contributed from oC1/ox, the whole temperature range involved in this work.
and thus Eq. (7) can be simplified to the first term on the The end member alloys were fabricated by vacuum arc
right-hand side. In addition, as the chemical potential li melting. Raw materials (with purities >99.9%) were melted
can be expressed in terms of the thermodynamic activity at least five times to make sure they were well mixed in their
ai through the equation liquid state. The ingots were cold rolled and cut into discs
8 mm in diameter and 3 mm thick. These discs were then

Table 1
Compositions of the end members in the diffusion couples.
Couple Alloy Composition (at.%)
Co Cr Fe Mn Ni
Cr–Mn 1 22 29 22 5 22
2 22 17 22 17 22
Fe–Co 3 33 23 11 11 22
4 11 23 33 11 22
Fe–Ni 5 23 24 30 11 12
6 23 24 12 11 30
4890 K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897

Fig. 1. Schematic diagram showing the assembly of the diffusion couples.

pre-annealed at 1273 K for 100 h in an argon-filled quartz particles, and no Kirkendall voids can be seen across the
tube to remove the defects induced by rolling and obtain entire diffusion zone in the couples. The concentration pro-
equiaxed grains with grain size 0.5–1 mm. Annealed discs files obtained from the Cr–Mn couple are shown in Fig. 3
were ground and polished with standard metallurgical pro- as examples. Obvious concentration gradients are seen only
cedures to obtain smooth surfaces. The discs were then for the two target elements of each couple (i.e., only Cr and
placed in a molybdenum container and assembled as the Mn show concentration gradient for the Cr–Mn couple).
diffusion couples, as shown in Fig. 1. The molybdenum con- Concentration profiles for other elements remain virtually
tainer was further sealed in an argon-filled quartz tube and horizontal throughout the diffusion zone. This is due to
annealed at 1173, 1223, 1273 and 1323 K for different times. the same initial concentration design of the three non-tar-
After annealing, the quartz tubes were quenched in water. get elements and the use of a near-ideal solution. There-
The diffusion couples were then carefully sliced, ground fore, the quasi-binary approach mentioned in Section 2
and polished perpendicularly to the bond interface. The dif- could be applied in the following calculation.
fusion zone across the bond interface was observed by scan- The locations of Matano planes calculated by Eq. (3)
ning electron microscopy (SEM; JEOL JSM-5410). The coincide well with the Kirkendall interfaces for all couples.
concentration profiles of all elements in the diffusion zone For example, in the Cr–Mn couple annealing at 1273 K for
were obtained using an energy dispersive spectrometry 100 h, the calculated marker movement was only 0.3 lm,
automation program. Each data point had an individual representing a negligible marker speed 1013 m s1 (the
spectrum and was analyzed quantitatively with ZAF correc- marker speed is usually >1011 m s1 in traditional fcc sys-
tion. At least three profiles were determined from every tems [36,40–45]). This indicates that the Kirkendall effect
individual sample, and at least three samples were per- could be neglected in the present study, and thus the
formed for each diffusion couple at each annealing temper-
ature. All the profiles were fitted with a Boltzmann function.

4. Results

4.1. Concentration profiles and diffusion coefficients

Fig. 2 shows a representative SEM micrograph to


clearly reveal the initial welding surface, i.e., the Kirkendall
interface. Most interfaces indeed contain few oxide

Fig. 2. SEM micrograph of the Fe–Ni HEA diffusion couple after Fig. 3. Concentration profiles in the Cr–Mn couple after annealing at
annealing at 1223 K for 90 h. 1273 K for 100 h.
K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897 4891

D ¼ D0 expðQ=RT Þ ð11Þ
and tabulated in Table 2. The results show that Ni has the
highest activation energy, 317.5 kJ mol1, in the CoCrF-
eMn0.5Ni alloy. It is interesting to find this value is very
close to the activation energy (321.7 kJ mol1) for grain
growth in the equimolar CoCrFeMnNi alloy [49]. This
can be explained by the following mechanism. In the grain
growth process, the composition essentially remains the
same, which requires the cooperative diffusion of all com-
Fig. 4. Concentration-dependent diffusion coefficients at 1273 K. The DFe posing elements during the movement of the grain bound-
curves obtained from the Fe–Co and Fe–Ni couples are within experi-
mental error. Here, only the DFe curve from the Fe–Co couples is shown.
ary. Therefore, the slowest diffusing element would
Dashed lines near the terminal compositions show the anomalous determine the grain growth rate. In addition, the composi-
diffusivities, which are artifacts caused by the numerical procedures and tion in the grain-growth study is in fact close to that of the
should be ignored [46]. present diffusion couples. It is thus reasonable to believe
that Ni is also the slowest diffusion element in the CoC-
calculated interdiffusion coefficients could be taken approx- rFeMnNi alloy, and one may obtain similar activation en-
imately as intrinsic diffusion coefficients. As for the small ergy for grain growth in the CoCrFeMnNi alloy. This
Kirkendall effect, which suggests the small net flux of consistency suggests that the slowest diffusion element
vacancy across the interface, the mechanism needs further would also determine the reaction rates of other diffu-
study in the future. sion-controlled process.
The concentration-dependent diffusion coefficients were To compare the diffusion of an element in different fcc
calculated from the concentration profiles by the method matrices, the values of Q and D0 for tracer diffusion in
described in Section 2. As shown in Fig. 4, the change in other conventional fcc metals are also listed. In addition,
diffusion coefficients is small in the composition range the normalized activation energies, Q/Tm (or Q/Ts, where
involved in the diffusion couples. The temperature depen- Ts is the solidus temperature for alloys), are listed to take
dence of the diffusion coefficients at the Kirkendall inter- into account the effect of melting point of the matrix. This
face where the composition approximately equals to is because the activation energies of diffusion in a matrix
CoCrFeMn0.5Ni is illustrated in Fig. 5. It is seen that, if are usually linearly related to the melting points of that
the elements are listed in the order of decreasing diffusion matrix [50]. Moreover, the extrapolation values of the dif-
rate, the sequence is Mn, Cr, Fe, Co and Ni. This sequence fusion coefficients DTm are also listed to compare the diffu-
is similar to that in conventional metals [47,48]. sion rates near the melting temperatures. The temperature
dependences of the diffusion coefficients in different systems
4.2. Activation energies and pre-exponential factors are plotted in Fig. 6. It can be seen that, for all the listed
elements, the diffusion coefficients in the Co–Cr–Fe–Mn–
The activation energies Q and the pre-exponential fac- Ni alloy system are the smallest in all the matrices com-
tors D0 are determined by the Arrhenius equation pared here. Correspondingly, the Q/Tm values in the pres-
ent HEA are the largest (Fig. 7). This is direct evidence for
the sluggish diffusion effect in HEA. It is also notable that,
for the same element, the above tendencies are related to
the number of elements in that matrix. For example, the
Q/Tm values in the present HEA are the highest; those in
Fe–Cr–Ni(–Si) alloys are the second; and those in pure
metals are the lowest. The mechanism behind these results
will be revealed in the next section.

5. Discussion

5.1. Quasi-chemical model and the distribution of seven-bond


interaction energy

The origin of the sluggish diffusion effect is the fluctua-


tion in LPE when an atom migrates from one site to
another, which is a result of the difference in total interac-
tion energy from site to site. Interaction energy is the excess
Fig. 5. Temperature dependence of the diffusion coefficients obtained free energy that arises when two unlike atoms bond with
from the three diffusion couples after annealing at different temperatures. each other. In HEA, each lattice site has different
4892 K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897

Table 2
Diffusion parameters for Cr, Mn, Fe, Co and Ni in different matrices; the compositions of Fe–Cr–Ni(–Si) alloys are in wt.%.
Solute System D0 (104 m2 s1) Q (kJ mol1) Tm (Ts) (K) Q/Tm DTm (1013 m2 s1)
Cr CoCrFeMnNi 5.59 292.9 1607 0.1823 1.69
Fcc Fe [51] 10.8 291.8 1812 0.1610 41.9
Co [52] 0.084 253.7 1768 0.1435 2.69
Ni [53] 5.2 289 1728 0.1673 9.55
Fe–15Cr–20Ni [48] 8.3 309 1731 0.1785 3.94
Fe–15Cr–45Ni [48] 4.0 293 1697 0.1727 3.83
Fe–22Cr–45Ni [48] 4.1 295 1688 0.1748 3.05
Fe–15Cr–20Ni–Si [48] 7.1 303 1705 0.1777 3.70
Mn CoCrFeMnNi 5.01 288.4 1607 0.1794 2.12
Fcc Fe [54] 0.019 240.8 1812 0.1329 2.17
Co [55] 0.093 243.7 1768 0.1378 5.87
Ni [56] 7.5 280.9 1728 0.1626 24.2
Fe CoCrFeMnNi 15.1 309.6 1607 0.1927 1.30
Fcc Fe [57] 0.46 284.1 1812 0.1568 2.97
Co [58] 0.21 262.5 1768 0.1485 3.69
Ni [59] 1 269.4 1728 0.1559 7.19
Fe–15Cr–20Ni [48] 5.3 308 1731 0.1779 2.69
Fe–15Cr–45Ni [48] 2.1 288 1697 0.1697 2.87
Fe–22Cr–45Ni [48] 1.5 286 1688 0.1694 2.12
Fe–15Cr–20Ni–Si [48] 5.1 303 1705 0.1777 2.66
Co CoCrFeMnNi 9.26 306.9 1607 0.1910 0.98
Fcc Fe [47] 1.12 301.9 1812 0.1666 2.22
Co [60] 0.55 288.7 1768 0.1633 1.63
Ni [47] 2.26 283.6 1728 0.1641 6.05
Ni CoCrFeMnNi 19.7 317.5 1607 0.1975 0.95
Fcc Fe [61] 3 314 1812 0.1733 2.66
Co [61] 0.43 282.2 1768 0.1596 1.98
Ni [62] 1.77 285.3 1728 0.1651 4.21
Fe–15Cr–20Ni [48] 1.5 300 1731 0.1733 1.33
Fe–15Cr–45Ni [48] 1.8 293 1697 0.1727 1.73
Fe–22Cr–45Ni [48] 1.1 291 1688 0.1724 1.09
Fe–15Cr–20Ni–Si [48] 4.8 310 1705 0.1818 1.53

Fig. 6. Temperature dependence of the diffusion coefficients for Cr, Mn, Fe, Co and Ni in different matrices.
K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897 4893

are adjacent to A (A1–A7 in Fig. 8), type-2 (T2) atoms


are adjacent to V (V1–V7 in Fig. 8), and type-3 (T3) atoms
are adjacent to both A and V (S1–S4 in Fig. 8). When atom
A exchanges with vacancy V, four T3 atoms remain to be
the neighbors of A, but the seven A–T1 bonds are broken,
and seven new A–T2 bonds are established instead. There-
fore, when an atom migrates, the change in LPE comes
from the interaction energy difference between A–T1 bonds
and A–T2 bonds. This energy is named the seven-bond
interaction energy (SBIE), which can be calculated using
a pair potential or quasi-chemical model. In this model,
the properties of a system are represented by the sum of
interactions between neighboring pairs, and any complica-
tions due to three-body interactions are ignored [64].
The first step is to calculate the interaction energy of a
single bond. Here, the interaction energies are independent
of concentration based on quasi-chemical model [64,65].
The interaction energy of a bond between the ith element
and jth element is defined by
1
Xij ¼ H ij  ðH ii þ H jj Þ ð12Þ
2
where Hij is the mean potential energy (cohesive energy or
binding energy) between the ith element and jth element.
According to Miedema’s model [66], the mixing enthalpy
of an equimolar i–j solid solution alloy DH mix
ij is contributed
by the unlike-atom bonds, which account for half of the to-
tal bonds. Furthermore, there are ZN0/2 bonds in an alloy
containing one mole of atoms, where N0 is Avogadro’s
Fig. 7. Normalized activation energies of diffusion for Cr, Mn, Fe, Co and number, and Z is the coordination number. Z is 12 for
Ni in different matrices.
the fcc structure, and 14 for the body-centered cubic struc-
ture [67]. Thus, the interaction energy contributed by an i–j
bond can be calculated from the binary mixing enthalpy:
Xij ¼ 4DH mix
ij =ZN 0 ð13Þ
Table 3 lists the binary mixing enthalpy and the calculated
single-bond interaction energy between each unlike atom
pair. For crystalline solid solutions, the contribution of
elastic strain to the mixing enthalpy needs to be considered
[68–70]. This is due to the lattice distortion arising from the
difference in the size of the atoms occupying equivalent
lattice sites. To take this effect into account, the mixing
enthalpies used here are the experimental values from solid
solutions [71–75], which already include the lattice strain
energies. Among these data, only the mixing enthalpy for

Fig. 8. Illustration of an atom–vacancy pair (A–V) and their neighboring Table 3


atoms in an fcc lattice: A1–7, atoms adjacent to A only, type-1; V1–7, atoms Binary mixing enthalpies (upper-right half) and the calculated single-bond
adjacent to V only, type-2; and S1–4, atoms adjacent to both A and V, interaction energies (lower-left half) for each atom pair in Co–Cr–Fe–Mn–
type-3. Ni alloys.
1
DH mix
ij ðkJ mol Þ

surrounding atoms. Thus, each site has different total inter- Cr 3.8 5.9 6.8 6.4
11.3 Mn 4.7 2.2 14
action energy. Consider an atom–vacancy pair (A–V) and 17.5 16.2 Fe 1.3 4.6
their neighboring atoms in an fcc lattice, as shown in 23.5 7.6 4.5 Co 0.0
Fig. 8. The nearest neighboring atoms of the A–V pair 22.1 48.4 15.9 0.0 Ni
can be divided into three types [63]. Type-1 (T1) atoms Xij (meV)
4894 K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897

(the approximate composition at the Kirkendall interface


in the annealed diffusion couples) is compared with that
in conventional Fe–Cr–Ni alloys (Fig. 10). The number
of energy states in HEA is apparently larger than that in
ternary alloys. This is because the number of possible states
increases with the number of constituent elements. The
number of combinations of seven atoms composing by n
elements is (7 + n  1)!/7!(n  1)!, which means that there
are 36 combinations in ternary alloys and 330 combina-
tions in quinary alloys. Additionally, the distribution of
SBIE in HEA is apparently wider than that in ternary
alloys. This is due to the large negative interaction energy
between Mn and Ni atoms (DHmix = 14 kJ mol1), lead-
ing to a wider energy distribution in the negative side of
the profile. In contrast, the negative interaction energies
Fig. 9. The mixing enthalpies of Co–Cr alloys as a function of Cr in the Fe–Cr–Ni alloys are smaller in value. Thus, the SBIE
concentration at 1473 K [73].
histogram cannot extend much towards the negative side.
It should be mentioned that, in conventional alloys, large
the Co–Cr system, 6.8 kJ mol1, is obtained by extrapolat- negative values of interaction energies usually lead to the
ing from low Cr-concentration region [73] to equal-molar formation of intermetallic compounds, whereas large posi-
composition through polynomial fitting, as shown in tive values lead to the segregation of elemental phases. For
Fig. 9. example, the formation of different compounds in the
The next step is to generate all the possible configura- Mn–Ni phase diagram are related to the large negative
tions of the seven neighbors (whether T1 or T2) around
atom A in a fully random n-element alloy. One can then
calculate the SBIE and probability of occurrence of each
configuration. For a specific seven-atom configuration k,
assume that atom A belongs to the ith element, and among
the seven neighbors around A, the number of jth element
atoms is Nj. Then the SBIE is given by
X n
Eki ¼ N j Xij ð14Þ
j¼1

where n is the number of constituent elements. And the


probability of this configuration is given by
       
7 7  N1 7  N1  N2 Nn
P ki ¼ X N1 1  X N2 2  X N3 3    X Nn n ð15Þ
N1 N2 N3 Nn

where Xj is the molar fraction of j element in the alloy, and


the superscripts to the right of Xj are exponents. In this
equation, the numbers of possible combinations and the
corresponding probabilities (X N1 1 , the probability to have
N1 1st-element atoms; X N2 2 , the probability to have N2
2nd-element atoms, and so on) for configuration k are mul-
tiplied together. Therefore, by calculating Eki and P ki for all
the possible configurations, the probability distribution
P ðEki Þ, is obtained. Furthermore, one can estimate the mac-
roscopic average of SBIE in a fully random solid solution
for the ith element simply by
X n
Ei ¼ 7X j Xij ð16Þ
j¼1

Thus, another expression P ðEki  Ei Þ, which emphasizes the


deviation of SBIE from the macroscopic average, is also
obtained.
The Ni element is now used as an example of the present
five elements, and P ðEkNi  ENi Þ in CoCrFeMn0.5Ni HEA Fig. 10. Probability distributions of SBIE for Ni in different matrices.
K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897 4895

interaction energy between them (DHmix = 14 kJ mol1). tend to minimize their energies, and those sites with lower
However, the positive interaction energy between Cr–Ni LPE for an element would become atomic traps and
(DHmix = 6.4 kJ mol1) accounts for the segregation of ele- increase the energy barrier and the activation energy for
mental phases at below eutectic temperature in the Cr–Ni diffusion. Consider two adjacent sites L and M, as shown
system. In contrast, high-entropy solid solutions have in Fig. 11. In pure metals, the LPE of sites L and M would
higher tolerance for elements with strong interactions. In be the same, since configurations around any atom are the
the present Co–Cr–Fe–Mn–Ni alloys, no phase separation same. Thus, the energy barrier (Eb) for a migration from L
was observed, even though they were furnace-cooled from to M equals that from M to L. However, for alloys, the
homogenization temperatures. This can be understood LPE of the two sites will differ, and the average difference
based on two reasons. First, in HEA, the high mixing is MD. Thus, the energy barriers for the two opposite
entropy enhances the mutual solubility between elements atomic jumps are different, and the transition frequencies
and the stability of solution phases [76]. This phenomenon between L and M are asymmetric. This means atoms will
has been well evidenced in many different HEA [77,78]. prefer to stay at lattice sites with lower LPE. Even if an
Secondly, owing to the multiple principal elements in atom jumps into a high-LPE site, the atom will tend to
HEA, the atom pairs with large interaction energies will hop back to its previous site. Therefore, the low-LPE sites
be diluted, and thus the driving force to form stoichiome- serve as atomic traps that hinder diffusion. For the diffu-
tric compounds is, in general, smaller. For example, in sion of Ni in CoCrFeMn0.5Ni HEA, MD, the average
CoCrFeMn0.5Ni alloy, the concentration of the strongest LPE difference between two lattice sites, is 60.3 meV. Thus,
Mn–Ni bonds is only 5%, which means that most of the at 1273 K the ratio of transition frequencies is CL!M/
Mn atoms bond with Co, Cr and Fe with lower interaction CM!L = exp(MD/kBT)  0.58, and the occupation time
energies. These two effects allow the HEA to have broader for Ni atom at a lower energy site is 1.73 times that at a
LPE distributions in the matrix. Indeed, the wider P ðEki Þ higher energy site in average. As a result, the larger the
histogram in the present HEA relative to that in the refer- LPE fluctuation, the stronger the trapping effect, and the
ence conventional alloys is not specific to Ni. It is common lower the diffusion rate will be. Indeed, Fultz and Anthony
to all other elements in the present HEA, although not [80] showed that for diffusion in two-dimensional lattices,
shown here. the activation energy is proportional to the LPE fluctua-
tion, which is a function of the binding energy difference
(HiiHij). In CoCrFeMn0.5Ni HEA, the significantly larger
5.2. Change in SBIE during atomic migration and its effect
LPE fluctuation relative to the reference conventional
on diffusion kinetics
alloys thus accounts for the slower diffusion kinetics, as
seen in Section 4.
The SBIE and probability of each seven-neighbor con-
Besides the LPE fluctuation mentioned above, the effect
figuration for Ni in CoCrFeMn0.5Ni and Fe–Cr–Ni alloys
of the saddle-point energies is also a factor. Saddle-point
were calculated in Section 5.1. As mentioned previously,
energies are influenced by both the migrating atom and
the change in LPE during atomic migration is the SBIE dif-
the nearest neighbors around the saddle point (T3 atoms
ference before and after the migration process. In order to
quantitatively compare the LPE change of a species in dif-
ferent matrices, the mean difference (MD) values of SBIE
in different matrices are calculated. MD is the summation
of all possible differences between two independent values,
0
Ek and Ek , weighted by their probabilities [79]:
X
K X K  0  0
MD ¼ P ðEk ÞP Ek Ek  Ek  ð17Þ
k¼1 k 0 ¼1

where K is the total number of possible configurations.


Therefore, MD represents the average LPE change during
atomic migration in a matrix. Again, Ni is used as an exam-
ple. The calculated MD values for Ni in CoCrFeMn0.5Ni,
Fe–15Cr–20Ni, Fe–15Cr–45Ni and Fe–22Cr–45Ni alloys
are 60.3, 41.7, 39.3 and 42.2 meV, respectively. Therefore,
the average LPE change for a Ni atom migrating to an
adjacent site in CoCrFeMn0.5Ni alloy is 1.5 times those
in Fe–Cr–Ni ternary alloys. This indicates that a diffusing
atom experiences significantly greater LPE fluctuation in
HEA than in traditional alloys. Fig. 11. Schematic diagram of the variation of LPE and MD during the
Atoms that experience greater fluctuation in LPE have migration of a Ni atom in different matrices. The MD for pure metals is
greater difficulty in diffusion. This is because atoms always zero, whereas that for HEA is the largest.
4896 K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897

in Fig. 8) [81]. In HEA, the saddle-point energies could also [6] Zhou YJ, Zhang Y, Wang YL, Chen GL. Appl Phys Lett
have a broader distribution than that of conventional 2007;90:181904.
[7] Wen LH, Kou HC, Li JS, Chang H, Xue XY, Zhou L. Intermetallics
alloys. Therefore, locations with high saddle-point energies 2009;17:266.
become roadblocks for diffusion. This factor thus also [8] Chen ST, Tang WY, Kuo YF, Chen SY, Tsau CH, Shun TT, et al.
reduces the diffusion efficiency. Mater Sci Eng A 2010;527:5818.
[9] Zhu JM, Fu HM, Zhang HF, Wang AM, Li H, Hu ZQ. Mater Sci
Eng A 2010;527:7210.
6. Conclusions [10] Chuang MH, Tsai MH, Wang WR, Lin SJ, Yeh JW. Acta Mater
2011;59:6308.
[11] Hsu CY, Juan CC, Wang WR, Sheu TS, Yeh JW, Chen SK. Mater
This paper presented the first study on the diffusion Sci Eng A 2011;528:3581.
kinetics in HEA. The quasi-binary interdiffusion coeffi- [12] Senkov ON, Wilks GB, Scott JM, Miracle DB. Intermetallics
cients in ideal-solution-like fcc Co–Cr–Fe–Mn–Ni alloys 2011;19:698.
were obtained in the temperature range 1173–1373 K. [13] Tsai MH, Wang CW, Tsai CW, Shen WJ, Yeh JW, Gan JY, et al. J
The results show that the interdiffusion coefficients are Electrochem Soc 2011;158:H1161.
[14] Chou YL, Wang YC, Yeh JW, Shih HC. Corros Sci 2010;52:3481.
approximately equal to the values of tracer diffusion coef- [15] Kao YF, Lee TD, Chen SK, Chang YS. Corros Sci 2010;52:1026.
ficients. The diffusion coefficients are not sensitive to the [16] Tsai MH, Yeh JW, Gan JY. Thin Solid Films 2008;516:5527.
composition in the diffusion couples. The sequence of ele- [17] Tong CJ, Chen YL, Chen SK, Yeh JW, Shun TT, Tsau CH, et al.
ments in the order of decreasing diffusion rate is: Mn, Cr, Metall Mater Trans A 2005;36A:881.
Fe, Co and Ni. In addition, the diffusion coefficients are [18] Shun TT, Hung CH, Lee CF. J Alloy Compd 2010;493:105.
[19] Tsai MH, Yuan H, Cheng G, Xu W, Tsai KY, Tsai CW, et al.
smaller than those in the pure fcc metals and Fe–Cr–Ni(– Intermetallics 2013;32:329.
Si) alloys. However, the normalized activation energies [20] Tsai MH, Wang CW, Lai CH, Yeh JW, Gan JY. Appl Phys Lett
Q/Tm in the present HEA are higher than those in the ref- 2008;92:052109.
erence metals. The above tendencies are related to the num- [21] Chang SY, Chen DS. Appl Phys Lett 2009;94:231909.
ber of constituent elements in the matrix. Therefore, the [22] Chang SY, Wang CY, Chen MK, Li CE. J Alloy Compd
2011;509:L85.
sluggish diffusion effect in HEA is observed. [23] Vineyard GH. Phys Rev 1956;102:981.
The higher Q/Tm values in alloys with a larger number [24] Cheng CY, Wynblatt PP, Dorn JE. Acta Metall 1967;15:1035.
of composing elements are attributed to the larger LPE [25] Pruthi D. B Mater Sci 1985;7:43.
fluctuation between the lattice sites. A quasi-chemical [26] Haus JW, Kehr KW. Phys Rep 1987;150:263.
model is used to calculate the average LPE change during [27] Temkin DE. Sov Math Dokl 1972;13:1172.
[28] Temkin DE. Sov Phys Solid State 1972;13:2840.
the atomic migration in each matrix, which is simply the [29] Otto F, Yang Y, Bei H, George EP. Acta Mater 2013;61:2628.
MD value of SBIE in that matrix. The results show that [30] Fujita H, Gosting LJ. J Am Chem Soc 1956;78:1099.
the CoCrFeMn0.5Ni alloy has larger LPE fluctuation than [31] Kirkaldy JS. Can J Phys 1957;35:435.
the reference fcc metals. For the diffusion of Ni in CoCrF- [32] Dayananda MA. Defect Diffus Forum 1992;83:73.
eMn0.5Ni alloy, the MD of SBIE is 60.3 meV, which is 1.5 [33] Darken LS. Trans AIME 1949;180:430.
[34] Sauer F, Freise V. Z Elektrochem 1962;66:353.
times that in Fe–Cr–Ni ternary alloys. Therefore, the abun- [35] den Broeder FJA. Scripta Metall 1969;3:321.
dant low-LPE sites can serve as traps and hinder atomic [36] Ronka KJ, Kodentsov AA, VanLoon PJH, Kivilahti JK, VanLoo
diffusion. As a result, the larger LPE fluctuation in HEA FJJ. Metall Mater Trans A 1996;27:2229.
leads to higher normalized activation energies and a lower [37] Kirkaldy JS, Young DJ. Diffusion in the condensed state. Lon-
diffusion rate, and thus the sluggish diffusion effect. don: Institute of Metals; 1987.
[38] Darken LS. Formal basis of diffusion theory. In: Hollomon JH,
editor. Atom movements. Cleveland, OH: ASM; 1951.
[39] Le Claire AD. Prog Met Phys 1953;4:265.
Acknowledgement [40] van Dal MJH, Pleumeekers MCLP, Kodentsov AA, van Loo FJJ. J
Alloy Compd 2000;309:132.
The authors gratefully acknowledge the financial sup- [41] van Dal MJH, Pleumeekers MCLP, Kodentsov AA, van Loo FJJ.
port for this research from the National Science Council Acta Mater 2000;48:385.
[42] Divya VD, Ramamurty U, Paula A. J Mater Res 2011;26:2384.
of Taiwan under Grant No. 100-2221-E-007-049-MY2.
[43] Iijima Y, Taguchi O, Hirano KI. Metall Trans A 1977;8:991.
[44] Iijima Y, Taguchi O, Hirano KI. Trans Jpn Inst Met 1980;21:366.
References [45] Butrymowicz DB, Manning JR. Metall Trans A 1978;9:947.
[46] Glicksman ME. Diffusion in solids: field theory, solid-state principles,
[1] Huang KH, Yeh JW. Master’s thesis, National Tsing Hua University; and applications. 1st ed. New York: Wiley; 1999.
1996. [47] Neumann G, Tuijn C. Self-diffusion and impurity diffusion in pure
[2] Hsu CY, Yeh JW, Chen SK, Shun TT. Metall Mater Trans A metals: handbook of experimental data. 1st ed. Oxford: Elsevier;
2004;35A:1465. 2009.
[3] Yeh JW, Chen SK, Lin SJ, Gan JY, Chin TS, Shun TT, et al. Adv [48] Rothman SJ, Nowicki LJ, Murch GE. J Phys F: Met Phys
Eng Mater 2004;6:299. 1980;10:383.
[4] Yeh JW. Ann Chim-Sci Mat 2006;31:633. [49] Liu WH, Wu Y, He JY, Nieh TG, Lu ZP. Scripta Mater 2013;68:526.
[5] Tong CJ, Chen MR, Chen SK, Yeh JW, Shun TT, Lin SJ, et al. [50] Brown AM, Ashby MF. Acta Metall 1980;28:1085.
Metall Mater Trans A 2005;36A:1263. [51] Bowen AW, Leak GM. Metall Trans 1970;1:1695.
K.-Y. Tsai et al. / Acta Materialia 61 (2013) 4887–4897 4897

[52] Davin A, Leroy V, Coutsouradis D, Habraken L. Mém Sci Rev [68] Loeff PI, Weeber AW, Miedema AR. J Less Common Met
Métall 1963;20:275. 1988;140:299.
[53] Jung SB, Yamane T, Minamino Y, Hirao K, Araki H, Saji S. J Mater [69] Janotti A, Krcmar M, Fu CL, Reed RC. Phys Rev Lett
Sci Lett 1992;11:1333. 2004;92:085901.
[54] Nohara K, Hirano K. Suppl Trans Iron Steel Inst Jpn 1971;11:1267. [70] Karunaratne MSA, Reed RC. Acta Mater 2003;51:2905.
[55] Iijima Y, Hirano KI, Taguchi O. Philos Mag 1977;35:229. [71] Hultgren R. Selected values of the thermodynamic properties of
[56] Swalin RA, Martin A. Trans AIME 1956;206:567. binary alloys. 1st ed. Cleveland, OH: ASM; 1973.
[57] Heumann T, Imm R. J Phys Chem Solids 1968;29:1613. [72] Jacob KT. Z Metallkd 1985;76:415.
[58] Mead HW, Birchenall CE. Trans AIME 1955;203:994. [73] Allibert C, Bernard C, Valignat N, Dombre M. J Less Common Met
[59] Bakker H, Backus J, Waals F. Phys Status Solidi B 1971;45:633. 1978;59:211.
[60] Bussmann W, Herzig C, Rempp W, Maier K, Mehrer H. Phys Status [74] Eremenko VN, Lukashenko GM, Sidorko VR. Russ J Phys Chem
Solidi A 1979;56:87. 1968;42:343.
[61] Badia M, Vignes A. Acta Metall 1969;17:177. [75] Moser Z, Zakulski W, Spencer P, Hack K. CALPHAD 1985;9:257.
[62] Bakker H. Phys Status Solidi 1968;28:569. [76] Yeh JW, Chen SK, Gan JY, Lin SJ, Chin TS, Shun TT, et al. Metall
[63] Welch DO. Mater Sci Eng 1969;4:9. Mater Trans A 2004;35A:2533.
[64] Swalin RA. Thermodynamics of solids. 2nd ed. New York: John [77] Guo S, Liu CT. Prog Nat Sci 2011;21:433.
Wiley & Sons; 1972. [78] Ng C, Guo S, Luan J, Shi S, Liu CT. Intermetallics 2012;31:165.
[65] Nix FC, Shockley W. Rev Mod Phys 1938;10:0001. [79] Kendall MG, Stuart A. The advanced theory of statistics. 3rd
[66] de Boer FR, Boom R, Mattens WCM, Meiedema AR, Niessen AK. ed. London: Charles Griffin; 1969.
Cohesion in metals: transition metal alloys. 2nd ed. Amster- [80] Fultz B, Anthony L. Defect Diffus Forum 1988;59:253.
dam: North-Holland; 1988. [81] Le Bouar Y, Soisson F. Phys Rev B 2002;65:094103.
[67] Miracle DB, Wilks GB, Dahlman AG, Dahlman JE. Acta Mater
2011;59:7840.

S-ar putea să vă placă și