ARTICLE

Received 8 Apr 2015 | Accepted 25 Aug 2015 | Published 29 Sep 2015 DOI: 10.1038/ncomms9485 OPEN

Entropy-stabilized oxides
Christina M. Rost1, Edward Sachet1, Trent Borman1, Ali Moballegh1, Elizabeth C. Dickey1, Dong Hou1,
Jacob L. Jones1, Stefano Curtarolo2 & Jon-Paul Maria1

Configurational disorder can be compositionally engineered into mixed oxide by populating a

single sublattice with many distinct cations. The formulations promote novel and entropy-
stabilized forms of crystalline matter where metal cations are incorporated in new ways.
Here, through rigorous experiments, a simple thermodynamic model, and a five-component
oxide formulation, we demonstrate beyond reasonable doubt that entropy predominates
the thermodynamic landscape, and drives a reversible solid-state transformation between
a multiphase and single-phase state. In the latter, cation distributions are proven to
be random and homogeneous. The findings validate the hypothesis that deliberate
configurational disorder provides an orthogonal strategy to imagine and discover new phases
of crystalline matter and untapped opportunities for property engineering.

1 Department of Materials Science and Engineering, North Carolina State University, Raleigh, North Carolina 27695, USA. 2 Department of Mechanical

Engineering and Materials Science, Center for Materials Genomics, Duke University, Durham, North Carolina 27708, USA. Correspondence and requests for
materials should be addressed to J.-P.M. (email: jpmaria@ncsu.edu) or to S.C. (email: stefano@duke.edu).

NATURE COMMUNICATIONS | 6:8485 | DOI: 10.1038/ncomms9485 | www.nature.com/naturecommunications 1

& 2015 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9485

A
grand challenge facing materials science is the continuous Inspired by research activities in the metal alloy communities
hunt for advanced materials with properties that satisfy and fundamental principles of thermodynamics we extend the
the demands of rapidly evolving technology needs. The entropy concept to five-component oxides. With unambiguous
materials research community has been addressing this problem experiments we demonstrate the existence of a new class of mixed
since the early 1900s when Goldschmidt reported the ‘the method oxides that not only contains high configurational entropy but
of chemical substitution’1 that combined a tabulation of cationic also is indeed truly entropy stabilized. In addition, we present a
and anionic radii with geometric principles of ion packing and hypothesis suggesting that entropy stabilization is particularly
ion radius ratios. Despite its simplicity, this model enabled a effective in a compound with ionic character.
surprising capability to predict stable phases and structures. As
early as 1926 many of the technologically important materials Results
that remain subjects of contemporary research were identified Choosing an appropriate experimental candidate. The candi-
(though their properties were not known); BaTiO3, AlN, GaP, date system is an equimolar mixture of MgO, CoO, NiO, CuO and
ZnO and GaAs are among that list. ZnO, (which we label as ‘E1’) so chosen to provide the appropriate
These methods are based on overarching natural tendencies for diversity in structures, coordination and cationic radii to test
binary, ternary and quaternary structures to minimize polyhedral directly the entropic ansatz. The rationale for selection is as fol-
distortions, maximize space filling and adopt polyhedral linkages lows: the ensemble of binary oxides should not exhibit uniform
that preserve electroneutrality1–3. The structure-field maps crystal structure, electronegativity or cation coordination, and
compiled by Muller and Roy catalogue the crystallographic there should exist pairs, for example, MgO–ZnO and CuO–NiO,
diversity in the context of these largely geometry-based that do not exhibit extensive solubility. Furthermore, the entire
predictions4. There are, however, limitations to the predictive collection should be isovalent such that relative cation ratios can
power, particularly when factors like partial covalency and be varied continuously with electroneutrality preserved at the net
heterodesmic bonding are considered. cation to anion ration of unity. Tabulated reference data for each
To further expand the library of advanced materials and component, including structure and ionic radius, can be found in
property opportunities, our community explores possibilities Supplementary Table 1.
based on mechanical strain5, artificial layering6, external
fields7, combinatorial screening8, interface engineering9,10 and
structuring at the nanoscale6,11. In many of these efforts, Testing reversibility. In the first experiment, ceramic pellets of
computation and experiment are important companions. E1 are equilibrated in an air furnace and quenched to room
Most recently, high-throughput methods emerged as a power- temperature. The temperature spanned a range from 700 to
ful engine to assess huge sections of composition space12–17 and 1,100 °C, in 50-°C increments. X-ray diffraction patterns showing
identified rapidly new Heusler alloys, extensive ion substitution the phase evolution are depicted in Fig. 1. After 700 °C, two
schemes18,19, new 18-electron ABX compounds20 and new ferroic prominent phases are observed, rocksalt and tenorite. The
semiconductors21. tenorite phase fraction reduces with increasing equilibration
While these methods offer tremendous predictive power and temperature. Full conversion to single-phase rocksalt occurs
an assessment of composition space intractable to experiment, between 850 and 900 °C, after which there are no additional
they often utilize density functional theory calculations conducted peaks, the background is low and flat, and peak widths are narrow
at 0 K. Consequently, the predicted stabilities are based on in two-theta (2y) space.
enthalpies of formation. As such, there remains a potential Reversibility is a requirement of entropy-driven transitions.
section of discovery space at elevated temperatures where entropy Consequently, low-temperature equilibration should transform
predominates the free-energy landscape. homogeneous 1,000 °C-equilibrated E1 back to its multiphase
This landscape was explored recently by incorporating state (and vice versa on heating). Figure 1 also shows a sequence
deliberately five or more elemental species into a single of X-ray diffraction patterns for such a thermal excursion; initial
lattice with random occupancy. In such crystals, entropic equilibration at 1,000 °C, a second anneal at 750 °C, and finally a
contributions to the free energy, rather than the cohesive energy, return to 1,000 °C. The transformation from single phase, to
promote thermodynamic stability at finite temperatures. The multiphase, to single phase is evident by the X-ray patterns and
approach is being explored within the high-entropy-alloy family demonstrates an enantiotropic (that is, reversible with tempera-
of materials (HEAs)22, in which extremely attractive properties ture32) phase transition.
continue to be found23,24. In HEAs, however, discussion
remains regarding the true role of configurational entropy25–28, Testing entropy though composition variation. A composition
as samples often contain second phases, and there are experiment is conducted to further characterize this phase tran-
uncertainties regarding short-range order. In response to these sition to the random solid solution state. If the driving force is
open discussions, HEAs have been referred to recently as entropy, altering the relative cation ratios will influence the
multiple-principle-element alloys29. transition temperature. Any deviation from equimolarity will
It is compelling to consider similar phenomena in non-metallic reduce the number of possible configurations O (Sc ¼ kBlog(O)),
systems, particularly considering existing information from thus increasing the transition temperature. Because Sc(xi) is
entropy studies in mixed oxides. In 1967 Navrotsky and logarithmically linked to mole fraction via Bxilog(xi), the com-
Kleppa showed how configurational entropy regulates the positional dependence is substantial.
normal-to-inverse transformation in spinels, where cations This dependency underpins our gedankenexperiment where the
transition between ordered and disordered site occupancy among role of entropy can be tested by measuring the dependency of
the available sublattices30,31. These fundamental thermodynamic transition temperature as a function of the total number of
studies lead one to hypothesize that in principle, sufficient components present, and of the composition of a single
temperature would promote an additional transition to a component about the equimolar formulation.
structure containing only one sublattice with random cation The calculated entropy trends for an ideal mixture are
occupancy. From experiment we know that before such illustrated in Fig. 2b, which plots configurational entropy for a
transitions, normal materials melt, however, it is conceivable set of mixtures having N species where the composition of an
that synthetic formulations exist, which exhibit them. individual species is changed and the others (N  1) are kept

2 NATURE COMMUNICATIONS | 6:8485 | DOI: 10.1038/ncomms9485 | www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9485 ARTICLE

200
111
900 °C
1,000 °C
850 °C
Log intensity

1,000 °C

800 °C
1 2

T(200)
T(002)

T(200)
T(002)
T(110)
T(110)
750 °C

750 °C

2 (°)

Figure 1 | X-ray diffraction patterns for entropy-stabilized oxide formulation E1. E1 consists of an equimolar mixture of MgO, NiO, ZnO, CuO and CoO.
The patterns were collected from a single pellet. The pellet was equilibrated for 2 h at each temperature in air, then air quenched to room temperature by
direct extraction from the furnace. X-ray intensity is plotted on a logarthimic scale and arrows indicate peaks associated with non-rocksalt phases, peaks
indexed with (T) and with (RS) correspond to tenorite and rocksalt phases, respectively. The two X-ray patterns for 1,000 °C annealed samples are offset in
2y for clarity.
200
111

1.8
220

N=5 1,100 1,100

1.6
311
222

1.4

Temperature (°C)

Temperature (°C)
1,050 1,050
J14 1.2 N=4
* 1.0 1,000 1,000
S/kB

* N=3
0.8 950 950
No CuO 0.6
N=2
* 0.4 900 900
*
0.2 850 850
0.0
Intensity

0.0 0.5 1.0 0.1 0.2 0.3 0.1 0.2 0.3

No CoO XN XMgo XCoO
*
*
1,100 1,100 1,100
No NiO
*
Temperature (°C)

Temperature (°C)

1,050 1,050 Temperature (°C) 1,050

*
1,000 1,000 1,000
No MgO
* 950 950 950
*
900 900 900
No ZnO
850 850 850
20 30 40 50 60 70 80 0.1 0.2 0.3 0.1 0.2 0.3 0.1 0.2 0.3
2 (°) XNiO XCuO XZnO

Figure 2 | Compositional analysis. (a) X-ray diffraction analysis for a composition series where individual components are removed from the parent
composition E1 and heat treated to the conditions that would otherwise produce full solid solution. Asterisks identify peaks from rocksalt while carrots
identify peaks from other crystal structures. (b) Calculated configurational entropy in an N-component solid solutions as a function of mol% of the Nth
component, and (c–g) partial phase diagrams showing the transition temperature to single phase as a function of composition (solvus) in the vicinity of the
equimolar composition where maximum configurational entropy is expected. Error bars account for uncertainty between temperature intervals. Each phase
diagram varies systematically the concentration of one element.

equimolar. Two dependencies become apparent: the entropy patterns in Fig. 2a show that removing any component oxide results
increases as new species are added and the maximum entropy is in material with multiple phases. A four-species set equilibrated
achieved when all the species have the same fraction. Both under these conditions never yields a single-phase material.
dependencies assume ideal random mixing. Two series of The second experiment uses five individual phase diagrams to
composition-varying experiments investigate the existence of explore the configurational entropy versus composition trend. In
these trends in formulation E1. each, the composition of a single component is varied by ±2,
The first experiment monitors phase evolution in five ±6 and ±10% increments about the equimolar composition
compounds, each related to the parent E1 by the extraction of a while the others are kept even. Since any departure from
single component. The sets are equilibrated at 875 °C (the threshold equimolarity reduces the configurational entropy, it should
temperature for complete solubility) for 12 h. The diffraction increase transition temperatures to single phase, if that transition

NATURE COMMUNICATIONS | 6:8485 | DOI: 10.1038/ncomms9485 | www.nature.com/naturecommunications 3

& 2015 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9485

is in fact entropy driven. The specific formulations used are given cations and release of oxygen to maintain stoichiometry. To
in Supplementary Table 2. address concerns regarding CuO reduction, Supplementary Fig. 2
Figure 2c–g are phase diagrams of composition versus shows a differential scanning calorimetry and thermal gravimetric
transformation temperature for the five sample sets that varied analysis curve for pure CuO collected under the same conditions.
mole fraction of a single component. The diagrams were There is no oxygen loss in the vicinity of 875 °C.
produced by equilibrating and quenching individual samples in
25 °C intervals between 825 and 1,125 °C to obtain the Ttrans- Testing homogeneity. All experimental results shown so far
composition solvus. In all cases equimolarity always leads to the support the entropic stabilization hypothesis. However, all
lowest transformation temperatures. This is in agreement with assume that homogeneous cation mixing occurs above the tran-
entropic promotion, and consistent with the ideal model shown sition temperature. It is conceivable that local composition fluc-
in Fig. 2b. One set of raw X-ray patterns used to identify Ttrans for tuations produce coherent clustering or phase separation events
10% MgO is given as an example in Supplementary Fig. 1. that are difficult to discern by diffraction using a laboratory sealed
tube diffractometer. The solvus lines of Fig. 2c–g support random
Testing endothermicity. Reversibility and compositionally mixing, as the most stable composition is equimolar (a condition
dependent solvus lines indicate an entropy-driven process. As only expected for ideal/regular solutions), but it is appropriate to
such, the excursion from polyphase to single phase should be ensure self-consistency with direct measurements. To characterize
endothermic. An entropy-driven solid–solid transformation is the cation distributions, extended X-ray absorption fine structure
similar to melting, thus requires heat from an external source33. (EXAFS) and scanning transmission electron microscopy with
To test this possibility, the phase transformation in formulation energy dispersive X-ray spectroscopy (STEM EDS) is used to
E1 can be co-analysed with differential scanning calorimetry and analyse structure and chemistry on the local scale.
in situ temperature-dependent X-ray diffraction using identical EXAFS data were collected for Zn, Ni, Cu and Co at the
heating rates. The data for both measurements are shown in Advanced Photon Source 12-BM-B34,35. The fitted data are
Fig. 3. Figure 3a is a map of diffracted intensity versus diffraction shown in Fig. 4, the raw data are given in Supplementary Fig. 3.
angle (abscissa) as a function of temperature. It covers B4° of 2y The fitted data for each element provide two conclusions: the
space centred about the 111 reflection for E1. At a temperature cation-to-anion first-near-neighbour distances are identical
interval between 825 and 875 °C, there is a distinct transition to (within experimental error of ±0.01 Å) and the local structures
single-phase rocksalt structure—all diffraction events in that for each element to approximately seven near-neighbour
range collapse into an intense o1114 rocksalt peak. distances are similar. Both observations are only consistent with
Figure 3b contains the companion calorimetric result where a random cation distribution.
one finds a pronounced endotherm in the identical temperature As a corroborating measure of local homogeneity, chemical
window. The endothermic response only occurs when the system analysis was conducted using a probe-corrected FEI Titan STEM
adds heat to the sample, uniquely consistent with an entropy- with EDS detection. Thin film samples of E1, prepared by pulsed
driven transformation33. We note the small mass loss (B1.5%) at laser deposition, are the most suitable samples to make the
the endothermic transition. This mass loss results from the assessment. Details of preparation are given in the methods, and
conversion of some spinel (an intermediate phase seen by X-ray X-ray and electron diffraction analysis for the film are provided in
diffraction) to rocksalt, which requires reduction of 3 þ to 2 þ Supplementary Figs 4 and 5. The sample was thinned by
mechanical polishing and ion milling. Figure 5 shows a collection
of images including Fig. 5a, the high-angle annular dark-field
Mass change (%) signal (HAADF).
0 5 10 15 In Fig. 5b–f, the EDS signals for the Ka emission energies of
Mg, Co, Ni, Cu and Zn are shown (additional lower magnification
R 111

1,100
images are included in Supplementary Fig. 6). All magnifications
1,000 reveal chemically and structurally homogeneous material.
S 311

900

800
Temperature (°C)

Temperature (°C)

700
Zn
W 101

600
 (k )×k 2 (Å–2)

500 Ni

400
R 111

Cu
300
R 111

R 111

200
DSC Co
Mass 100

35.5 36.5 37.5 –30 –20 –10 0 10 0 2 4 6 8 10 12

2θ (°) Endo DSC (mW) Exo k (Å–1)

Figure 3 | Demonstrating endothermicity. (a) In situ X-ray diffraction Figure 4 | Extended X-ray absorption fine structure. EXAFS measured at
intensity map as a function of 2y and temperature; and (b) differential Advanced Photon Source beamlime 12-BM after energy normalization and
scanning calorimetry trace for formulation ‘E1’. Note that the conversion to fitting. Note that the oscillations for each element occur with similar relative
single phase is accompanied by an endotherm. Both experiments were intensity and at similar reciprocal spacing. This suggests a similar local
conducted at a heating rate of 5 °C min  1. structural and chemical environment for each.

4 NATURE COMMUNICATIONS | 6:8485 | DOI: 10.1038/ncomms9485 | www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9485 ARTICLE

a HAADF of the mol fraction of each multiplied by the reference transition

Mg Co
enthalpy.
This assumption is consistent with the report of Davies et al.
who showed that the chemical potential of a particular cation in a
particular structure is associated with the molar volume of that
structure36. Since the rocksalt phases of ZnO and CuO have
5Å molar volumes comparable to E1, their reference transition
enthalpy values are considered suitable proxies.
Ni Cu Zn
In comparison, the maximum theoretically expected config-
urational entropy difference at 875 °C (the temperature were we
observe the transition experimentally) between the single species
and the random five-species solid solution is B15 kJ mol  1,
5 kJ mol  1 larger than the calculated enthalpy of transition. It is
possible that the origins of this difference are related to mixing
energy as the reference energy values for structural transitions to
rocksalt do not capture that aspect.
Figure 5 | STEM–EDS analysis of E1. (a) HAADF image. Panels labelled as While the present phase diagrams that monitor Ttrans as a
Zn, Ni, Cu, Mg and Co are intensity maps for the respective characteristic function of composition demonstrate rather symmetric behaviour
X-rays. The individual EDS maps show uniform spatial distributions for each about the temperature minima, it is unlikely that mixing
element and are atomically resolved. enthalpies are zero for all constituents. Indeed, literature reports
show that enthalpies of mixing between the constituent oxides in
E1 are finite and of mixed sign, and their magnitudes are on the
X-ray diffraction, EXAFS and STEM–EDS probes are sensitive same order as the 5 kJ mol  1 difference between our calculated
to 10 s of nm, 10 s of Å and 1 Å length scales, respectively. While predictions36. This energy difference may be accounted for by
any single technique could be misinterpreted to conclude finite and positive mixing enthalpies.
homogenous mixing, the combination of X-ray diffraction, Following this argument, we can achieve a self-consistent
EXAFS and STEM–EDS provide very strong evidence. We note, appreciation for the entropic driving force and the enthalpic
in particular, the similarity in EXAFS oscillations (both in penalties for solution formation in E1 by considering enthalpies
amplitude and position) out to 12 inverse angstroms. This of the associated structural transitions and expected entropy
similarly would be lost if local ordering or clustering were present. values for ideal cation mixing.
Consequently, we conclude with certainty that the cations are As a final test, these predictions can be compared with
uniformly dispersed. experiment, specifically by calculating the magnitude of the
endotherm observed by DSC at the transition from multiple-
Discussion phase to single-phase states. Doing so we find a value
The set of experimental outcomes show that the transition from B12 kJ mol  1 (with an uncertainty of ±2 kJ mol  1). While
multiple-phase to single phase in E1 is driven by configurational we acknowledge the challenge of quantitative calorimetry, we
entropy. To complete our thermodynamic understanding of this note that this experimental result is intermediate to and in close
system, it is important to understand and appreciate the enthalpic agreement with the predicted values.
penalties that establish the transition temperature. In so doing, Compared with metallic alloys, the pronounced impact of
the data set can be tested for self-consistency, and the present entropy in oxides may be surprising given that on a per-atom
data are brought into the context of prior research on oxide basis the total disorder per volume of an oxide seems be lower
solubility. than in a high-entropy alloy, as the anion sublattice is ordered
First, we consider an equation relating the initial and final (apart from point defects). The chemically uniform sublattice is
states of the proposed phase transition: perhaps the key factor that retains cation configurational entropy.
As an illustration, consider a comparison between random metal
alloys and random metal oxide alloys.
MgOðRSÞ þ NiOðRSÞ þ CoOðRSÞ þ CuOðTÞ þ ZnOðWÞ ¼ðMg; Ni; Co; Cu; ZnÞOðRSÞ
Begin by reviewing the case of a two-component metallic
mixture A–B. If the mixture is ideal, the energy of interaction
For MgO, NiO and CoO, the crystal structures of the initial EA–B ¼ (EA–A þ EB–B)/2, there is no enthalpic preference for
and final states are identical. If we assume that solution of each bonding, and entropy regulates solution formation. In this
into the E1 rocksalt phase is ideal, the enthalpy for mixing is zero. scenario, all lattice sites are equivalent and configurational
For CuO and ZnO, there must be a structural transition to entropy is maximized. This situation, however, never occurs as
rocksalt on dissolution from tenorite and wurtzite, respectively. If no two elements have identical electronegativity and radii values.
we again assume (for simplicity) that the solution is ideal, the Figure 6a illustrates a two-component alloy scenario A–B where
mixing energy is zero, but there is an enthalpic penalty associated species B is more electronegative than A. Consequently, the
with the structure transition. From Davies et al. and Bularzik interaction energies EA–A, EB–B and EA–B will be different. A
et al., we know the reference chemical potential changes for the random mixture of A–B will produce lattice sites with a
wurtzite-to-rocksalt and the tenorite-to-rocksalt transitions of distribution of first near neighbours, that is, species A
ZnO and CuO; they are 25 and 22 kJ mol  1, respectively36,37. If coordinated to 4-B atoms, 2-A and 2-B atoms, etcy Different
we make the assumption that the transition enthalpies of coordinations will have different energy values and the sites are
ZnO(wurtzite) to ZnO(rocksalt E1) and CuO(tenorite) to no longer indistinguishable. Reducing the number of equivalent
CuO(rocksalt E1) are comparable, then the enthalpic penalty sites reduces the number of possible configurations and S.
for solution into E1 can be estimated. For ZnO and CuO, the Now consider the same two metallic ions co-populating a
transition to solid solution in a rocksalt structure involves an cation sublattice, as in Fig. 6b. In this case, there is always an
enthalpy change of (0.2)  (25 kJ mol  1) þ (0.2)  (22 kJ mol  1), a intermediate anion separating neighbouring cation lattice sites.
total of þ 10 kJ mol  1. This calculation is based on the product Again, in the limiting case where only first near neighbours are

NATURE COMMUNICATIONS | 6:8485 | DOI: 10.1038/ncomms9485 | www.nature.com/naturecommunications 5

& 2015 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9485

anion (or vice versa), which includes broad classes of chalcogenides,

A B O A B nitrides and halides; particularly when covalent character is modest.
The entropic driving force—engineered by cation composition—
provides a departure from traditional crystal-chemical principles
that elegantly predict structural trends in the major ternary and
quaternary systems. A companion set of structure–property
relationships that predict new entropy-stabilized structures with
novel cation incorporation await discovery and exploitation.

Methods
Solid-state synthesis of bulk materials. MgO (Alfa Aesar, 99.99%), NiO (Sigma
Aldrich, 99%), CuO (Alfa Aesar, 99.9%), CoO (Alfa Aesar, 99%) and ZnO (Alfa
Aesar 99.9%) are massed and combined using a shaker mill and 3-mm diameter
yttrium-stabilized zirconia milling media. To ensure adequate mixing, all batches
Figure 6 | Binary metallic compared with a ternary oxide. A schematic
are milled for at least 2 h. Mixed powders are then separated into 0.500-g samples
representation of two lattices illustrating how the first-near-neighbour and pressed into 1.27-cm diameter pellets using a uniaxial hydraulic press at
environments between species having different electronegativity (the 31,000 N. The pellets are fired in air using a Protherm PC442 tube furnace.
darker the more negative charge localized) for (a) a random binary metal
alloy and (b) a random pseudo-binary mixed oxide. In the latter, near- Temperature evolution of phases. Ceramic pellets of E1 are equilibrated in an air
neighbour cations are interrupted by intermediate common anions. furnace and quenched to room temperature by direct extraction from the hot zone.
Phase analysis is monitored by X-ray diffraction using a PANalytical Empyrean
X-ray diffractometer with Bragg-Brentano optics including programmable diver-
considered, every cation lattice site is ‘identical’ because each has gence and receiving slits to ensure constant illumination area, a Ni filter, and a 1-D
the same immediate surroundings: the interior of an oxygen 128 element strip detector. The equivalent counting time for a conventional point
octahedron. Differentiation between sites is only apparent when detector would be 30 s per point at 0.01° 2y increments. Note that all X-ray are
the second near neighbours are considered. From the configura- collected using substantial counting times and are plotted on a logarithmic scale.
To the extent knowable using a laboratory diffractometer, the high-temperature
tional disorder perspective, if each cation lattice site is identical, samples are homogeneous and single phase: there are no additional minor peaks,
and thus energetically similar to all others, the number of the background is low and flat, and peak widths are sharp in two-theta (2y) space.
microstates possible within the macrostate will approach the Temperature-dependent diffraction data are collected with PANalytical
maximum value. Empyrean X-ray diffractometer with Bragg-Brentano optics including
programmable divergence and receiving slits to ensure constant illumination area,
This crystallographic argument is based on the limiting a Ni filter, and a 1-D 256 element strip detector. The samples are placed in a
case where first-near-neighbour interactions predominate the resistively heated HTK-1200N hot stage in air. The samples are ramped at a
energy landscape, which is an imperfect approximation. constant rate of 5 °C min  1 with a theta–two theta pattern captured every 1.5 min.
Second and third near neighbours will influence the distribution Calorimetry data are collected using a Netzsch STA 449 F1 Jupiter system in a Pt
crucible at 5 °C min  1 in flowing air.
of lattice site energies and the number of equivalent microstates—
but the impact will be the same in both scenarios. A larger
number of equivalent sites in a crystal with an intermediate Determining solvus lines. Five series of powders are mixed where the amount of
one constituent oxide is varied from the parent mixture E1. Supplementary Table 2
sublattice will increase S and expand the elemental diversity lists the full set of samples synthesized for this experiment. Each individual sample
containable in a single solid solution and to lower the is cycled through a heat-soak-quench sequence at 25 °C increments from 850 °C up
temperature at which the transition to entropic stabilization to 1,150 °C. The soak time for each cycle is 2 h, and samples are then quenched to
occurs. We acknowledge the hypothesis nature of this model at room temperature in o1 min.
After the quenching step for each cycle, samples are immediately analysed for
this time, and the need for a rigorous theoretical exploration. It is phase identification using a PANalytical Empyrean X-ray diffractometer using the
presented currently as a possibility and suggestion for future conditions identified above. If more than one phase is present, the sample would be
consideration and testing. put through the next temperature cycle. The temperature at which the structure is
We demonstrate that configurational disorder can promote determined to be pure rocksalt, with no discernable evidence of peak splitting or
secondary phases, is deemed the transition temperature as a function of
reversible transformations between a poly-phase mixture and a composition. Supplementary Fig. 1 shows an example of the collected X-ray
homogeneous solid solution of five binary oxides, which do not patterns after each cycle using the E1L series with þ 10% MgO. Once single phase
form solid solutions when any of the constituents are removed is achieved, the sample is removed from the sequence.
provided the same thermal budget. The outcome is representative Note that this entire experiment is conducted two times. Initially in 50 °C
of a new class of materials called ‘entropy-stabilized oxides’. increments and longer anneals, and to ensure accuracy of temperature values and
reproducibility, a second time using shorter increments and 25 °C anneals.
While entropic effects are known for oxide systems, for example, Findings in both sets are identical to within experimental error bar values. In the
random cation occupancy in spinels30, order–disorder transfor- latter case, error bars correspond to the annealing interval value of 25 °C.
mations in feldspar38, and oxygen nonstoichiometry in layered In the main text relating to Fig. 2a we note that in addition to small peaks from
perovskites39, the capacity to actively engineer configurational second phases, X-ray spectra for N ¼ 4 samples with either NiO or MgO removed
show anisotropic peak broadening in 2y and skewed relative intensities where
entropy by composition, to stabilize a quinternary oxide with a I(200)/I(111) is less than unity. This ratio is not possible for the rocksalt structure.
single cation sublattice, and to stabilize unusual cation Supplementary Table 3 shows the result of calculations of structure factors for a
coordination values is new. Furthermore, these systems provide random equimolar rocksalt oxide with composition E1. Calculations show that the
a unique opportunity to explore the thermodynamics and 200 reflection is the strongest, and that the experimentally measured relative
intensities of 111/200 are consistent with calculations. We use this information as a
structure–property relationships in systems with extreme means too best assess when the transition to single phase occurs since the most
configurational disorder. likely reason for the skewed relative intensity is an incomplete conversion to the
Experimental efforts exploring this composition space are single-phase state. This dependency is highlighted in Supplementary Fig. 1.
important considering that such compounds will be challenging
to characterize with computational approaches minimizing X-ray absorption fine structure. X-ray absorption fine structure (XAFS) is made
formation energy (for example, genetic algorithms) or with possible through the general user programme at the Advanced Photon Source in
adhoc thermodynamic models (for example, CALPHAD, cluster Lemont, IL (GUP-38672). This technique provides a unique way to probe the local
environment of a specific element based on the interference between an emitted
expansion)6. core electron and the backscattering from surrounding species. XAFS makes no
We expect entropic stabilization in systems where near- assumption of structure symmetry or elemental periodicity, making it an ideal
neighbour cations are interrupted by a common intermediate means to study disordered materials. During the absorption process, core electrons

6 NATURE COMMUNICATIONS | 6:8485 | DOI: 10.1038/ncomms9485 | www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9485 ARTICLE

will absorb incident X-ray energies equal to or greater than their respective binding We note that STEM analysis is also performed on cryogenically fractured E1
energies. The emitted photoelectron wave interacts with neighbouring species, and powder samples, and epitaxial thin films along [001] and [110] zone axes. In all
the resulting absorption spectrum, displayed as absorption intensity versus incident cases STEM EDS analysis revealed no second phases and homogeneous and
energy, shows characteristic modulations unique to the target atom and its random elemental distributions within the E1 crystals. The STEM data featured in
environment. Fig. 5 of the main text was chosen since the thin film configuration coated with a
Equimolar amounts of the constituent oxides (MgO, NiO, CuO, CoO and ZnO) capping layer of ITO mitigated charging most effectively and allowed access to
are mixed and pre-reacted at 1,000 °C in air for 12 h with intermittent stirring near-atomic resolution with channelling conditions.
during calcination. The product is mixed into an isopropyl alcohol slurry and ball Supplementary Fig. 5 is a selected area diffraction pattern for E1 taken along
milled using yttrium-stabilized media for 24 h. The powder is then dried in a fume o0014, the pattern contains no diffraction events that are attributable to second
hood at room temperature then re-fired at 1,000 °C for 12 h, then checked via phases or to cation ordering. As such, we conclude single phase on the local scale.
X-ray diffraction to ensure phase purity and that peaks remain narrow and intense. Supplementary Fig. 6 is a lower magnification STEM image showing a wider area
Milled grain size is measured using scanning electron microscopy and determined view as compared with STEM EDS data in the main text. Two observations are of
to average B10 mm. particular note: (1) the HAADF-STEM image on the left suggests high crystallinity;
A 2:1 powder to 10% PVA/H2O suspension is mixed continuously to disperse and (2) the STEM–EDS analysis shows no evidence for chemical segregation or
particles within the solution as well as aid in breaking up any agglomerates. Using a phase separation over a lager range.
Cookson Electronics P-6000 spin coater, thin layers are spun onto 2  2 cm square
pieces of 25-mm thick Kapton. By trial and observation, it is determined that Configurational entropy in the ideal model. The following derivation describes
spinning at 2,000 r.p.m. for 1 min makes a homogenous thin film with the the method to determine the composition dependence of configurational entropy
appropriate quantity of particles for XAFS analysis. shown in Fig. 2b of the main text. An N-species system having composition {xi} has
Advanced Photon Source beamline 12-BM is utilized for its energy range of ideal entropy equal to:
4.5–23 keV, which can probe all cation species except magnesium. Absorption
spectra are recorded as a function of energy using a fluorescence set-up40 with a X
N

Canberra 13-element Ge detector. The energies per measurement range from S ¼  kB xi logðxi Þ:
i¼1
150 eV before the known K absorption edge of the target element to B1,000 eV
past the edge onset. Supplementary Table 3 lists the cation species of interest and The maximum S is reached at equicomposition xi ¼ 1/N for each i, so:
their respective K edges. Simultaneously to the sample fluorescence, reference foils Smax ¼  kB logðN Þ
are measured in transmission mode. This enables the energy calibration of the data If only one species is varied, composition x1 ¼ x for instance, while leaving the
relative to the theoretical edge of the metal, since compounds tend to have a slight other N  1 species at equicomposition:
variation in their absorption-edge energies. Each measurement is repeated three
times to check for systematic error and to improve signal to noise ratio. 1x
xi 6¼ 1 ¼
The raw data shown in Supplementary Fig. 2a plots the absorption edge and N 1
modulations on the post edge background. In order to isolate the EXAFS from the ideal entropy becomes:
these spectra, a background function is fit and subtracted. Energy space is   
1x 1x
transformed into k-space via the equation25: S ¼  kB xlogðxÞ þ ðN  1Þ log
N 1 N 1
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   
2mðE  E0 Þ 1x
k¼ ¼  kB xlogðxÞ þ ð1  xÞlog :
‘2 N 1
where m is electron mass, : is the reduced Planck’s constant, and E0 is the An expanded plot of entropy versus N for the entire series is shown in the
absorption-edge energy. Supplementary Fig. 2b shows the isolated EXAFS from the Supplementary Fig. 7.
measurement. With this data, qualitative conclusions can be made pertaining to the
degree of randomness of the cation species. If there were ordering within the
system, these spectra would not demonstrate such consistent oscillatory structure
References
1. Goldschmidt, V. M. The laws of crystal chemistry. Naturwissenschaften 14,
and the scattering pathways for individual species would be unique. We limit our
477–485 (1926).
current conclusions at this somewhat conservative level as it provides the evidence
2. Hume-Rothery, W. & Powell, H. M. On the theory of super-lattice structures in
needed to support a random solid solution.
alloys. Z. Kristallogr. 91, 23–47 (1935).
3. Pauling, L. The sizes of ions and the structure of ionic crystals. J. Am. Chem.
Scanning transmission electron microscopy. To best facilitate sample prepara- Soc. 49, 765–790 (1927).
tion and atomic-resolution analysis in STEM, a single crystal E1 thin film is grown 4. Muller, O. & Roy, R. The Major Ternary Structural Families (Springer, 1974).
on a {100} MgO substrate using pulsed laser deposition and thinned to electron 5. Choi, K. et al. Enhancement of ferroelectricity in strained BaTiO3 thin films.
transparency. The deposition process used a KrF 248 nm excimer laser; with an Science 306, 1005–1009 (2004).
energy density of 3 J cm  2, substrate temperature of 600 °C, an oxygen pressure of 6. Gudiksen, M., Lauhon, L., Wang, J., Smith, D. & Lieber, C. Growth of nanowire
50 mtorr and target to substrate distance of 4 cm. A deposition rate of 6 Hz and superlattice structures for nanoscale photonics and electronics. Nature 415,
40,000 pulses resulted in an B400-nm-thick film. The thin film sample was used 617–620 (2002).
for two reasons: (1) an edge-oriented substrate facilitates imaging along a low index 7. Park, S. & Shrout, T. Ultrahigh strain and piezoelectric behavior in relaxor
zone axis perpendicular to the thinnest portion of the sample (this can be chal- based ferroelectric single crystals. J. Appl. Phys. 82, 1804 (1997).
lenging for random powder specimens); and (2) by capping the thin film with a 8. Xiang, X. et al. A combinatorial approach to materials discovery. Science 268,
conductor, one can provide a conductive pathway to mitigate the sample charging 1738–1740 (1995).
that ultimately manifests in image drift. To do so, E1 films were coated with 50 nm 9. Rijnders, G. & Blank, D. Materials science: build your own superlattice. Nature
of indium tin oxide (ITO) at room temperature using radio frequency-magnetron 433, 369–370 (2005).
sputtering. Indium tin oxide is the preferred conductor as it is mechanically similar 10. Paisley, E. et al. Surfactant assisted growth of MgO films on GaN. Appl. Phys.
to a halide oxide and thus responds comparably to mechanical polishing.
Lett. 101, 092904 (2012).
Laser ablated samples were examined by four-circle diffraction to assess
11. Gao, P. et al. Conversion of zinc oxide nanobelts into superlattice-structured
crystallinity and epitaxy. Supplementary Fig. 4 shows a theta–two theta and an
nanohelices. Science 309, 1700–1704 (2005).
omega scan for E1 prepared at 600 °C. The films are epitaxial to the MgO substrate
(expected since the lattice mismatch is below 1%), and the mosaicity observed in 12. Wang, Y. et al. Ab initio lattice stability in comparison with CALPHAD lattice
the omega circle is consistent with that present in the MgO substrate. MgO stability. Calphad 28, 79–90 (2004).
substrates are known to have limited crystal quality (B0.02° in omega) due to the 13. Curtarolo, S. et al. AFLOW: An automatic framework for high-throughput
flame-fusion technique used to grow them. materials discovery. Comput. Mater. Sci. 58, 218–226 (2012).
An Allied Multiprep polishing system is utilized to prepare a cross-sectional 14. Curtarolo, S. et al. AFLOWLIB.ORG: a distributed materials properties
electron microscopy sample by wedge polishing technique41. To achieve electron repository from high-throughput ab initio calculations. Comput. Mater. Sci. 58,
transparency, the polished sample is ion milled with a Fischione Model 1050 Ion 227–235 (2012).
Mill while cooling with liquid nitrogen. 15. Potyrailo, R. et al. Combinatorial and high-throughput screening of materials
A JEOL 2000 S/TEM is used to collect selected area diffraction patterns from the libraries: review of state of the art. ACS Comb. Sci. 13, 579–633 (2011).
E1 thin film. An aberration corrected FEI Titan G260–300 kV S/TEM equipped with 16. Curtarolo, S., Morgan, D., Persson, K., Rodgers, J. & Ceder, G. Predicting crystal
an X-FEG source and an advanced Super-XTM EDS detector system is used to analyse structures with data mining of quantum calculations. Phys. Rev. Lett. 91 (2003).
the structure and chemistry of E1. The Titan is operated at 200 kV for HAADF STEM 17. Curtarolo, S. et al. The high-throughput highway to computational materials
imaging and EDS mapping with the convergence semi-angle set to 15 mrad. The design. Nat. Mater. 12, 191–201 (2013).
atomic-resolution EDS map indicating the position and the arrangement of the ions in 18. Hautier, G., Fischer, C., Ehrlacher, V., Jain, A. & Ceder, G. Data mined ionic
the unit cell can be explained by corresponding HAADF–STEM images in which the substitutions for the discovery of new compounds. Inorg. Chem. 50, 656–663
atomic columns containing heavier elements are observed brighter. (2011).

NATURE COMMUNICATIONS | 6:8485 | DOI: 10.1038/ncomms9485 | www.nature.com/naturecommunications 7

& 2015 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9485

19. Carrete, J., Mingo, N. & Wang, S. et al. Nanograined half-Heusler 40. Calvin, S. XAFS for Everyone (CRC, 2013).
semiconductors as advanced thermoelectrics: an ab-initio high-throughput 41. Voyles, P., Grazul, J. & Muller, D. Imaging individual atoms inside crystals with
statistical study. Adv. Funct. Mater. 24, 7427–7432 (2014). ADF-STEM. Ultramicroscopy 96, 251–273 (2003).
20. Gautier, R. et al. Prediction and accelerated laboratory discovery of previously
unknown 18-electron ABX compounds. Nat. Chem. 7, 308–316 (2015).
21. Bennet, J., Garrity, K., Rabe, K. & Vanderbilt, D. Orthorhombic A BC
Acknowledgements
J-P.M., E.C.D. and C.M.R. acknowledge support from ARO under contract W911NF-14-
semiconductors as antiferroelectrics. Phys. Rev. Lett. 110, 017603 (2013).
0285. J-P.M. and C.M.R. acknowledge the Advanced Photon Source (supported by
22. Cantor, B., Chang, I., Knight, P. & Vincent, A. Microstructural development in
proposal 38672) for access to synchrotron experiments. S.C. acknowledges partial sup-
equiatomic multicomponent alloys. Mater. Sci. Eng. A 375, 213–218 (2004).
port by DOD (ONR-MURI- N000141310635), DOE (DE-AC02-05CH11231, BES
23. Gludovatz, B. et al. A fracture-resistant high-entropy alloy for cryogenic
#EDCBEE) and the Duke Center for Materials Genomics and the aflowlib.org con-
applications. Science 345, 1153–1158 (2014).
sortium. J-P.M. and S.C. acknowledge support from DOD (ONR-MURI-N00014-15-1-
24. Gali, A. & George, E. Tensile properties of high- and medium-entropy alloys.
2863). The authors acknowledge the use of the Analytical Instrumentation facility at
Intermetallics 39, 74–78 (2013).
North Carolina State University who provided access to X-ray diffraction and electron
25. Jones, N., Aveson, J., Bhowmik, A., Conduit, B. & Stone, H. On the entropic
microscopy facilities. AIF is supported by the State of North Carolina and the National
stabilisation of an Al0.5CrFeCoNiCu high entropy alloy. Intermetallics 54,
Science Foundation. J-P.M. and C.M.R. acknowledge useful discussions with Dr Sungsik
148–153 (2014).
Lee at the Advanced Photon Source regarding collection and interpretation of XAFS data.
26. Otto, F., Yang, Y., Bei, H. & George, E. Relative effects of enthalpy and entropy
on the phase stability of equiatomic high-entropy alloys. Acta Mater. 61,
2628–2638 (2013). Author contributions
27. Wu, Z., Bei, H., Otto, F., Pharr, G. & George, E. Recovery, recrystallization, C.M.R., E.S., T.B. and J-P.M. designed the experimental plan, performed sample
grain growth and phase stability of a family of FCC-structured multi- synthesis and ex situ sample characterization; J-P.M. and S.C. envisioned and imple-
component equiatomic solid solution alloys. Intermetallics 46, 131–140 (2014). mented the experiments to test the entropy hypothesis; S.C. performed thermodynamic
28. Zhang, F. et al. An understanding of high entropy alloys from phase diagram calculations for composition dependence of entropy; D.H. and J.L.J. performed tem-
calculations. Calphad 45, 1–10 (2014). perature dependent X-ray diffraction experiments, and E.C.D. and A.M. conducted TEM
29. Santodonato, L. et al. Deviation from high-entropy configurations in the atomic investigations.
distributions of a multi-principal-element alloy. Nat. Commun. 6, 5964 (2015).
30. Navrotsky, A. & Kleppa, O. The thermodynamics of cation distributions in
simple spinels. J. Inorg. Nucl. Chem. 29, 2701–2714 (1967). Additional information
31. Navrotsky, A. & Kleppa, O. Thermodynamics of formation of simple spinels. Supplementary Information accompanies this paper at http://www.nature.com/
J. Inorg. Nucl. Chem. 30, 479–498 (1968). naturecommunications
32. Jones, S., Fenerty, J. & Pearce, J. The enantiotropic phase transition of
antimony(III) oxide. Thermochim. Acta 114, 61–66 (1987). Competing financial interests: The authors declare no competing financial interests.
33. Bragg, W. & Williams, E. The effect of thermal agitation on atomic
arrangement in alloys. II. Proc. R. Soc. Lond. A 151, 540–566 (1935). Reprints and permission information is available online at http://npg.nature.com/
34. Knapp, G., Nevitt, M., Aldred, A. & Klippert, T. An EXAFS study of interionic reprintsandpermissions/
distances in complex lanthanide oxides. J. Phys. Chem. Solids 46, 1321–1325 (1985).
35. Antonioli, G., Lottici, P., Parisini, A. & Razzetti, C. EXAFS study of mixed crystals How to cite this article: Rost, C. M. et al. Entropy-stabilized oxides. Nat. Commun.
of the AIIBIII2XVI4 family. Prog. Cryst. Growth Charact. 10, 9–18 (1984). 6:8485 doi: 10.1038/ncomms9485 (2015).
36. Davies, P. & Navrotsky, A. Thermodynamics of solid solution formation in
NiO-MgO and NiO-ZnO. J. Solid State Chem. 38, 264–276 (1981).
37. Bularzik, J., Davies, P. & Navrotsky, A. Thermodynamics of solid-solution This work is licensed under a Creative Commons Attribution 4.0
formation in NiO-CuO. J. Am. Ceram. Soc. 69, 453–457 (1986). International License. The images or other third party material in this
38. Megaw, H. Crystal Structures: A Working Approach (Saunders, 1973). article are included in the article’s Creative Commons license, unless indicated otherwise
39. Navrotsky, A. Thermochemistry of perovskite-related oxides with high in the credit line; if the material is not included under the Creative Commons license,
oxidation states: superconductors, sensors, fuel cell materials. Pure Appl. Chem. users will need to obtain permission from the license holder to reproduce the material.
66, 1759–1764 (1994). To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

8 NATURE COMMUNICATIONS | 6:8485 | DOI: 10.1038/ncomms9485 | www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.