Академический Документы
Профессиональный Документы
Культура Документы
LONDON www.nature.com/naturematerials
The Macmillan Building, 4 Crinan Street, London N1 9XW
T: +44 207 833 4000 F: +44 207 843 4563
Editor Vincent Dusastre
Senior Editors Victoria Cleave, Joerg Heber, Fabio Pulizzi
Associate Editors Hilary Crichton, Alison Stoddart
Consultant Editor Philip Ball
Senior Production Editor Derna Simpson
Assistant Production Editor Jenny Marsden
Art Editor David Shand
Editorial Assistant Theresa Tuson
ManagEMEnt officEs
nPg london nature@nature.com
The Macmillan Building, 4 Crinan Street, London N1 9XW
T: +44 207 833 4000 F: +44 207 843 4563
Managing Director Steven Inchcoombe
Publishing Director David Swinbanks
Publisher Jason Wilde
Associate Publisher Emma Green
Editor-in-Chief, Nature Publications Philip Campbell
Marketing Director Della Sar
Operations Director John Carroll
Director Of Web Publishing Timo Hannay
Associate Director, UK Production Jenny Henderson
Head Of Marketing, Physical Sciences Jane Macmillan
Marketing Manager, Physical Sciences Gurpreet Gill-Bains
Editorial Production Director James McQuat
Managing Production Editor Donald McDonald
Senior Copy Editor Jane Morris
Web Production Manager, UK Deborah Anthony
Production Director Yvonne Strong
Senior Production Controller Kelly Hopkins
Production Controller Emilia Orviss
nPg nEw York nature@natureny.com
75 Varick Street, 9th Floor, New York, NY 10013-1917
T: +1 212 726 9200 F: +1 212 696 9006
Chief Technology Officer Howard Ratner
Head Of Web Services Anthony Barrera
Executive Editor Linda Miller
nPg asia-Pacific nature@natureasia.com
Chiyoda Building 2-37 Ichigayatamachi, Shinjuku-Ku, Tokyo 162-0843 Japan
T: +81 3 3267 8751 F: +81 3 3267 8746
Associate Director Asia-Pacific Antoine E. Bocquet
Manager Koichi Nakamura
Operations Director Hiroshi Minemura
Asia-Pacific Sales Director Kate Yoneyama
Marketing Manager Masahiro Yamashita
Production Manager Takesh Murakami
Asia-Pacific Sales Manager Ken Mikami
nPg india npgindia@nature.com
3a, 4th Floor, Dlf Corporate Park, GurGaon 122002, India
T: +91 12 4288 1054/55 F: +91 12 4288 1052
Head Of Business Development Debashish Brahmachari
Sales And Marketing Manager Harpal Singh Gill
disPlaY advErtising physicalsciences@nature.com
Global Head of Display Advertising Andrew Douglas T: +44 207 843 4975 F: +44 207 843 4996
Asia-Pacific Sales Director Kate Yoneyama T: +81 3 3267 8765 F: +81 3 3267 8746
Advertising Director George Lui T: +44 207 843 4966 F: +44 207 843 4749
Advertising Manager, Physical Sciences Simon Allardice T: +1 415 403 9034 F: +1 415 781 3805
Asia-Pacific Display Advertising Manager Ken Mikami T: +81 3 3267 8751 F: +81 3 3267 8746
naturEjobs naturejobs@nature.com
European Sales Manager Andrew Douglas T: +44 207 843 4975 F: +44 207 843 4996
US Sales Manager Kenneth Finnegan T: +44 207 843 4975 F: +44 207 843 4996
Asia-Pacific Sales Manager Ayako Watanabe T: +81 3 3267 8765 F: +81 3 3267 8746
rEPrints reprint@nature.com
For commercial reprint orders of 600 or more, please contact:
US/Canada: reprints@natureny.com
Northern Europe/UK/ROW: reprints@nature.com
Southern Europe/Latin America: v.jurado@macmillanmedical.com
Asia-Pacific: m.kurosaki@natureasia.com
India: d.brahmachari@nature.com
sitE licEnsE businEss unit
Americas T: +1 888 331 6288 institutions@natureny.com
Asia/Pacific T: +81 3 3267 8751 institutions@natureasia.com
Australia/New Zealand T: +61 3 9825 1160 nature@macmillan.com.au
Europe/Row T: +44 207 843 4759 institutions@nature.com
India T: +91 124 2881054/55 npgindia@nature.com
custoMEr sErvicE
For all print and online assistance, please visit www.nature.com/help
Senior Global Customer Service Manager Gerald Coppin
ORIGINAL RESEARCH TyPESET By River Valley Technologies, www.river-valley.com
PRINTED By Wyndeham Grange, www.wyndeham.co.uk
editorial
161 Putting evolution to good use
research highlights
163 our choice from the recent literature
reView article
175 Deformation and failure of protein materials in physiologically extreme
conditions and disease
Markus J. Buehler and Yu Ching Yung
letters
189 Supramolecular control of the magnetic anisotropy in two-dimensional
high-spin Fe arrays at a metal interface
Pietro Gambardella, Sebastian Stepanow, Alexandre Dmitriev, Jan Honolka,
Frank M. F. de Groot, Magalí Lingenfelder, Subhra Sen Gupta, D. D. Sarma,
Peter Bencok, Stefan Stanescu, Sylvain Clair, Stéphane Pons, Nian Lin,
Ari P. Seitsonen, Harald Brune, Johannes V. Barth and Klaus Kern
→N&V p165
oN THe Cover 194 magnetic memory of a single-molecule quantum magnet wired to a
Biomaterials
gold surface
Model behaviour Matteo Mannini, Francesco Pineider, Philippe Sainctavit, Chiara Danieli,
Review Article p175 Edwige Otero, Corrado Sciancalepore, Anna Maria Talarico, Marie-Anne Arrio,
Andrea Cornia, Dante Gatteschi and Roberta Sessoli
Nanocatalysis
→N&V p165
Catalysis by clusters
Letter p213; News & Views p167 198 room-temperature defect-engineered spin filter based on a
epitaxial graphene non-magnetic semiconductor
The right atmosphere X. J. Wang, I. A. Buyanova, F. Zhao, D. Lagarde, A. Balocchi, X. Marie, C.W. Tu,
Letter p203; News & Views p171 J. C. Harmand and W. M. Chen
Nature Materials (ISSN 1476-1122) is published monthly by Nature Publishing Group (Porters South, 4 Crinan Street, London N1 9XW, UK). Editorial Office: Porters South, 4 Crinan Street, London N1 9XW, UK. Telephone:
+44 (0)20 7833 4000. Fax: +44 (0)20 7843 4563. Email: materials@nature.com. North American Advertising: Nature Materials, 75 Varick Street, 9th Floor, New York, NY 10013-1917, USA. Telephone: +1 212 726 9200.
Fax: +1 212 696 9006. European Advertising: Nature Materials, Porters South, 4 Crinan Street, London N1 9XW, UK. Telephone: +44 (0)20 7833 4000. Fax: +44 (0)20 7843 4749. New subscriptions/renewals/changes of
address/back issues and all other customer service questions should be addressed to: North America: Nature Publishing Group, Customer Services Department, 75 Varick Street, 9th Floor, New York, NY 10013-1917, USA.
Telephone: +1 (866) 363 7860. Fax: +1 (212) 334 0879. Outside North America: Nature Publishing Group, Subscriptions Department, Brunel Road, Houndmills, Basingstoke, Hants, RG21 6XS, UK. Telephone: +44 (0)1256
329242. Fax: +44 (0)1256 812358. Annual Subscription Rates: US/Canada US$3060 (Canada add 5% GST) (institutional/corporate), US$199 (Canada add 5% GST) (personal 1 year); UK/RoW (excluding Europe and Japan)
£1570 (institutional/corporate), £106 (personal 1 year); Europe ¤2430 (institutional/corporate), ¤164 (personal 1 year). Japan: contact NPG Nature Asia-Pacific, Chiyoda Bldg. 5F, 2-37 Ichigayatamachi, Shinjuku-Ku, Tokyo,
162-0843, JAPAN. Telephone: +81 3 3267 8751. For single back issue prices contact the publisher. Subscription information is available at the Nature Materials homepage at http://www.nature.com/naturematerials. Application
to mail Periodicals postage is paid at Rahway NJ. Postmaster: send address changes to Nature Materials Subscriptions Department, Brunel Road, Houndmills, Basingstoke, Hants, RG21 6XS, UK or Nature Materials Subscriptions
Department, PO Box 5054, Brentwood, TN 37024-5054, USA. Reprints: Nature Materials Reprints Department, Porters South, 4 Crinan Street, London N1 9XW, UK. © 2009 Macmillan Publishers Limited. All rights reserved.
K. S. NovoSelov
an advantage when looking for variety, but enough apart to allow the characterization
not if exact structures are required. Martin of individual wires still attached to their
and Mijangos have investigated the range native substrate. Not only were they able to
of polymer nanostructures available via uncover lasing properties such as the Fabry–
template synthesis and produced guidelines Pérot lasing modes, they also produced
for achieving different results. The authors way that conducting electrons are removed the first height-selective 3D maps of the
used the widely studied porous anodic and an energy gap is opened. The pristine photoluminescence emission of nanowires
aluminium oxide templates. They found material investigated shows a resistivity by using UV-laser scanning confocal
that the structures obtained depended on independent of temperature, which is typical microscopy. Such detailed characterization
infiltration method — wetting-based, vacuum of graphene. After exposure to hydrogen of heterogeneous nanostructure optical
or spin — and the diameter of the template: plasma, however, the resistivity increases properties should aid research aimed at using
narrower pores resulted in nanofibres rather considerably and shows a temperature them in optoelectronic and solar applications.
than nanotubes being created. Composite dependence typical of insulators. Most
tubes containing magnetic nanoparticles importantly, the modification can be perfectly
were created by melting a bulk polystyrene/ reversed by annealing the hydrogenated
Crossbar memories
Nano Lett. doi:10.1021/nl8037689 (2009)
iron–platinum composite on the surface of sample. Transmission electron microscopy
the template. The small size of the particles images indicate that the structure of the
(3–4 nm) compared with the thickness of the hydrogenated samples is comparable to that of
nanotube walls (70 nm) enabled the particles graphane, a material that had previously only
to infiltrate the template with polymer. The been studied theoretically. Aside from the
rules should prove valuable in simplifying the clear potential for electronic applications, the
creation of arrays of polymer nanostructures result demonstrates the possibility of creating
of various aspect ratios. a range of two-dimensional molecules by
chemical modification of graphene.
© 2009 ACS
From graphene to graphane
Science 323, 610–631 (2009) Vertical nanocavities
J. Am. Chem. Soc. doi:10.1021/ja8092339 (2009)
Among the various routes for controlling
the electronic properties of graphene, Single-crystalline ZnO nanowires behave With the end of Moore’s law looming in
chemical modification seems to be the least as lasing nanocavities — affording a gain the not-too-distant future, researchers
investigated. Kostya Novoselov and colleagues medium and a resonant cavity owing to are intensively researching alternatives
have now demonstrated that hydrogenation the reflectivity of the planar end-facets. to silicon electronics. One of the more
enables the electronic structure of this single Usually the wires are produced in closely promising approaches is to use self-assembled
layer graphite to be modified, transforming packed arrays, so characterizing the optical nanostructures to complement existing top-
it into an insulator. The secret is that on properties of a single wire requires the down technology. In particular, crossbar
attaching hydrogen atoms to graphene, transfer of one wire from its initial growth arrays promise very high integration densities
the carbon bond is transformed in such a substrate onto another substrate. This process that can easily exceed those achieved with
the present technology. Researchers from the
University of Michigan have now realized
transparent paper Adv. Mater. doi:10.1002/adma.200803174 (2009) a non-volatile crossbar memory array
capable of storing up to 1 kbit of information.
As well as being optically transparent, glass has a low thermal expansion coefficient The array is based on silver nanowires
making it suitable for use in electronic devices. Conversely, it is fragile; therefore a stronger, crossing sandwiched amorphous/p-type
flexible material that retains the thermal expansion coefficient of glass is desirable. Now silicon nanowires. At their intersections,
Masaya Nogi and colleagues have produced a material satisfying these requirements Ag/a-Si/p-Si structures form. The amorphous
from the main constituent of paper — cellulose. A suspension of wood nanofibres is silicon acts as a memristive switch, so that a
filtered, producing a wet sheet of uniformly positioned fibres. Slowly drying this while it is voltage applied between Ag nanowire and p-Si
compressed between meshwire/filter paper layers gives a highly dense sheet of nanofibres layer is able to switch the material between a
bound by hydrogen bonds with few or no cavities. The lack of interstices between fibres high and low resistance state. This crossbar
suppresses light scattering, and once the surface has been smoothed by polishing, the array has write speeds faster than 10 ns, good
sheet becomes up to 71% transparent at a wavelength of 600 nm. The ‘cellulose nanofibre endurance and long retention rates. With
paper’ can be folded and has good mechanical properties. Alternative ways of smoothing a yield of 92% and the possibility of scaling
the surface can prevent cellulose deterioration, making it a prime candidate for roll-to-roll the size of a single element below 30 nm, the
processing for the production of flexible electronic devices such as solar cells. devices establish the potential of these arrays
for non-volatile memory applications.
Heiko Wende
W
olfgang Pauli once claimed that due to an exchange interaction within issue: in contrast to the work on the single
God created the volume but the molecule. molecular magnets by Mannini et al.,
the surface was invented by Gambardella and colleagues follow Gambardella et al. observe no coupling
the devil. Maybe for this reason surfaces a different route to control magnetism2. of the Fe spins in their supramolecular
have a sinister attraction for scientists At first they assemble a two-dimensional network through, for example,
fascinated by the possibilities of completely supramolecular network in situ on a copper intramolecular exchange interactions. In
new effects and hence functionality single crystal. Then they demonstrate that their case, the Fe spins are ordered only by
at surfaces. But the interactions of the preferred magnetic orientation of the the applied magnetic field. However, for
molecular adsorbates with surfaces can be system, its magnetic anisotropy, can be the single molecular magnets investigated
complicated and the experimental control controlled by the adsorption of oxygen by Mannini et al., the spins within one
of these systems can easily turn into a molecules. Without the adsorbed oxygen molecule show ordering phenomena at
nightmare. Two works published in this molecules, the supramolecular assembly a temperature of 0.5 K. Still, owing to
issue by Matteo Mannini and colleagues1 consisting of Fe and benzenedicarboxylic resonant tunnelling effects in zero field,
(page 194) and Pietro Gambardella and acid can be magnetized easily parallel to the remanent magnetization of their single
colleagues2 (page 189) go through this the surface plane. After oxygen adsorption, molecular magnets remains pretty small.
nightmare of surface physics in order this easy direction turns out of the plane. As both studies are carried out at very
to come closer to realizing the vision of The use of oxygen as an adsorbate is low temperatures in the kelvin and even
molecular spintronics. particularly interesting, as in another recent sub-kelvin regime, and the important
In addition to conventional electronics, work it is demonstrated that the magnetic magnetic effects are determined at sizeable
where electron charge is used to carry coupling of paramagnetic porphyrin applied magnetic fields ranging from
information, the basic idea in spintronics molecules to ferromagnetic surfaces can be 0.25 T up to 6 T, an immediate realization
is to exploit the spin, where the use of tailored through oxygen atoms3. of a molecular spintronic device is not
magnetic systems allows control through, It is helpful to look at the similarities yet possible. Nevertheless, both works
for example, magnetic fields. Magnetic but also the fundamental differences demonstrate the current possibilities and
molecules are widely studied as suitable between the two studies published in this the new frontiers of surface science. An
candidates for spintronics applications.
Their advantage is their broad flexibility,
shown by the fact that complex organic
molecules are also nature’s preferred
systems for realizing highly complex
processes. However, molecules must
always be seen as part of the environment
they are placed in, and to use molecular
spintronics for real, a fundamental
understanding of the basic interactions
of magnetic molecules with surfaces is
essential. It is in this context that these two
studies move a step closer to the molecular
spintronics goal.
In the work by Mannini et al., the focal
point is the identification of a magnetic
memory effect of a monolayer of single-
molecule magnets consisting of Fe4
complexes wired to a gold surface1 (Fig. 1).
© AndreA CorniA
important conclusion of both studies use X-ray absorption spectroscopy at Fe4 complexes is not destroyed by their
is that the interaction of the magnetic modern third-generation synchrotron connection to the surface, owing to their
molecules with the surface does not radiation facilities to reveal the secrets structural stability and redox robustness.
destroy the magnetic behaviour. In of their molecules. The X-ray magnetic This is a clear advance over well-studied
particular, in the study by Mannini et al. circular dichroism technique provides single molecular systems such as Mn12.
the single molecular magnets were the necessary sensitivity to probe the In conclusion, both studies demonstrate
connected to the surface by thiolate- magnetic properties of the molecules in an that the hell that is surface science is not
terminated aliphatic chains that allow element-specific manner. the worst place to be, and that fascinating
direct control of the magnetic interaction1. With these techniques it is shown science can emerge — as long as the place
This bonding was achieved through a that Fe does indeed connect to the TPA is cooled down sufficiently. ❐
wet-chemical synthesis making use of molecules, and thereby the interaction
self-assembly effects. with the Cu substrate is weakened2. To Heiko Wende is at the Fachbereich Physik and
In contrast, the supramolecular network support these findings on the electronic Center for Nanointegration Duisburg-Essen
studied by Gambardella and colleagues structure, the experimental results of (CeNIDE), Universität Duisburg-Essen, 47048
was constructed using self-assembly by Gambardella et al. are accompanied by Duisburg, Germany.
evaporating a molecular precursor layer solid density functional calculations and e-mail: heiko.wende@uni-due.de
of terephthalic acid (TPA) followed by atomic multiplet calculations to model the references
deposition of Fe atoms from an electron- experimental X-ray absorption spectra2. 1. Mannini, M. et al. Nature Mater. 8, 194–197 (2009).
beam evaporator under ultrahigh vacuum In ref. 1 the experimental spectra reveal 2. Gambardella, P. et al. Nature Mater. 8, 189–193 (2009).
conditions. Interestingly, both studies that the single molecular behaviour of the 3. Bernien, M. et al. Phys. Rev. Lett. 102, 047202 (2009).
Material Witness
Shaping fate
The promise of stem-cell therapies channels in their outer membranes.
for tissue regeneration hinges on the But more surprising is the fact that
fact that, to a first approximation, they seem responsive to texture and
stem cells can become any other order. MSCs grown on nanopatterned
cell type. But therein lies one of polymer surfaces have been found to
the biggest challenges — for how become more osteoblast-like when the
does a stem cell decide its fate? This surfaces are embossed with random
decision is generally made in the arrays of nanopits, compared with
body through complex biochemical regular, ordered arrays2. That raises
pathways involving diffusing signalling the prospect of using nanopatterned
molecules. One approach to stem-cell matrices to define the distributions PHILIP BALL
therapeutics is to manipulate these of cell types in new tissue seeded
routes using either natural signalling from stem cells, for example in protein: an extracellular matrix
factors or synthetic small molecules bone regeneration. deposited by the cells, through which
that serve the same role. How does this work? Shu Chien they can adhere to the surface. But
But there can be another, perhaps and co-workers at the University of these blobs were far less abundant on
more surprising determinant of California at San Diego now think the wider tubes, simply because there
stem-cell fate. It may be influenced they have some clues3. They have is less space to put them. As a result,
by purely mechanical means such as found a new guiding factor for MSC cells seeking to anchor themselves have
stretching or stressing cells, for example differentiation that seems to be purely to stretch further in the latter case,
by altering the stiffness of the matrix geometric. They grew human MSCs and the researchers think that this
in which they grow. Mesenchymal on substrates of aligned arrays of deformation triggers differentiation to
stem cells (MSCs), which are potential titanium dioxide nanotubes with a bone-forming lineage.
progenitors of many cell types varying diameter, from 30 to 100 nm, That not only suggests a way to
including bone-growing osteoblasts, made electrochemically from thin guide bone growth by controlling
muscle-making myoblasts and films of titanium. The behaviour of nanostructure; because titanium
tissue-making fibroblasts, will guide the cells was strongly dependent on nanotubes are themselves good
differentiation towards myoblasts in a the nanotube size: for 30-nm tubes, candidates for a biocompatible bone-
soft matrix that resembles brain tissue, they adhered well but didn’t really fostering implant material, they can
but towards osteoblasts in a hard, differentiate at all, whereas for 100-nm do two jobs at once, providing both
bone-mimicking matrix 1. This suggests tubes they became long, thin and support and guidance. ❐
a role for materials engineering in osteoblast-like.
references
stem-cell therapy. Elongation is the key. Chien 1. Engler, A. J. et al. Cell 126, 677–689 (2006).
It’s been long known that cells can and colleagues saw that the smaller 2. Dalby, M. J. et al. Nature Mater. 6, 997–1003 (2007).
sense and respond to deformation, nanotubes became quickly decorated 3. Oh, S. et al. Proc. Natl Acad. Sci. USA
for example via switchable ion around their open ends with blobs of doi:10.1073/pnas.0813200106 (2009).
nanocatalysis
staying put
Preparation of supported subnanometre platinum clusters that are stable provides a new design strategy for
industrial nanocatalysts.
gianfranco Pacchioni
i
t has been suggested that catalysis Biased substrate holder
by supported particles is one of the Cluster source Collimator Quadrupole
first examples of nanotechnology: mass selector
heterogeneous catalysts are often made of Metal rod
Skimmer Focusing Focusing Focusing
small particles of precious metals, typically lens lenses lenses
a few nanometres in size, deposited on an
‘inert’ oxide support (silica, alumina) or He
other porous materials such as zeolites.
However, methods used to fabricate
industrial catalysts produce metal particles Continuous Focusing Detector
that cover a wide range of sizes, with a cluster beam lenses Quadrupole
Vaporization deflector
classical statistical distribution. In technical laser Ion Clusters of single size /
focusing lens narrow size distribution
terms, the particles are not monodispersed, 4.5 kHz
which opens the question: do they all have
the same chemical activity? To answer this,
methods to construct collections of well- Figure 2 | The apparatus used to generate, mass-select and soft-land platinum clusters on an oxide
defined, nearly monodispersed ensembles of support to perform catalytic tests on well-defined nanoclusters.
nanoparticles are essential1–3.
On page 213 of this issue, Vajda and
colleagues report on the preparation of size- practical interest. It was only towards the The kinetic energy of the clusters landing
selected Pt8–10 clusters stabilized on a high- end of the previous century that Ueli Heiz on the support is controlled by biasing
surface-area aluminium oxide film prepared and co-workers were able to produce and the substrate. In this way low deposition
by atomic layer deposition (Fig. 1)4. The gently deposit mass-selected metal clusters energies are obtained, thus avoiding
catalytic activity of the clusters for the on a thin magnesium oxide film grown on a fragmentation of the clusters on their impact
oxidative dehydrogenation of propane is metal support 6,7. These experiments opened with the surface, an approach known as
40–100 times higher than that of more the way to the preparation of monodispersed soft landing. The kinetic energy per atom
conventional platinum catalysts. ensembles of nanoparticles8,9. These have is similar to that used in the deposition
Studies performed on gas-phase mass- surprising properties; some cluster sizes of Ag19 clusters on Pt(111) in an earlier
selected metal clusters reveal that below are chemically very active, whereas others experiment by Schaub and colleagues who
a given size, typically a few tens of atoms, are inert. The part that surface defects in showed, using atomic-resolution scanning
one enters a non-scalable regime where the the supporting oxide play in stabilizing tunnelling microscope images, that in
properties of the particle change markedly the clusters and in modifying their charge these conditions the clusters remain intact
with size5. For instance, properties may state was also discovered: a Au8 cluster after deposition10. This is vital to guarantee
vary substantially on going from an even grown on an oxygen vacancy in a MgO that one has the same cluster sizes on the
to an odd number of atoms in the cluster. film converts CO to CO2 in the presence of surface as were produced in the gas phase.
The study of such phenomena is even more oxygen at low temperature, unlike the same Further characterization of the samples is
complex when one deals with supported cluster deposited on a defect-free MgO provided by synchrotron X-ray scattering
clusters, thus approaching materials of film6. Despite these considerable advances, measurements that show the stability and
the possibility of using tiny subnanometre the three-dimensional shape of the platinum
particles in catalytic processes of industrial clusters. Critically, because of the porosity of
a b
interest has been hampered by the inability the support, the clusters remain stable over a
to stabilize the clusters on a substrate and temperature range of 20–400 °C, an essential
prevent them from sintering — that is, aspect in performing catalytic reactions on
forming large particles at the expense of the these well-defined nanocatalysts.
smaller ones. Despite the great advances in the
Al2O3 Pt Vajda and colleagues have been able characterization of supported nanoclusters,
to overcome this problem. They have there are still aspects that are difficult to
produced gas-phase neutral and positively evaluate experimentally. For instance, it is
Figure 1 | The catalytic system. a, Schematic charged platinum clusters using a laser not easy to determine the cluster charge.
view of a Pt8–10 cluster deposited on a thin vaporization source; the positively charged This is often an important parameter
alumina film covering a porous anodized clusters are guided into a quadrupole mass because charged gas-phase clusters can be
aluminium oxide support. b, Structure of a Pt8 spectrometer, and selected according to highly reactive11. To overcome this problem,
cluster on the support. mass (Fig. 2). the group performed density functional
theory calculations, and these show that likely to be the rate-limiting step of the subnanometre clusters stabilized on specific
the Pt8 cluster on alumina withdraws reaction because the rest of the pathway to supports could result in new catalysts for a
electrons from the surface and becomes propylene formation is thermodynamically variety of industrially relevant processes. ❐
negatively charged. downhill. Furthermore, there are no
Once the nanocatalysts had been important differences between the Gianfranco Pacchioni is in the Dipartimento di Scienza
characterized, catalysis tests were carried computed reaction profiles on a gas-phase dei Materiali, Università degli Studi Milano-Bicocca,
out under atmospheric pressure, and the or supported Pt8 nanocatalyst, providing via R. Cozzi 53, 20125 Milano, Italy.
turnover frequencies were measured at a evidence that the support has little effect e-mail: gianfranco.pacchioni@unimib.it
temperature around 500 °C. The surprising on the reaction barrier in this case. The
result is that the activity of the Pt8–10 catalyst conclusion is that the undercoordinated References
is far higher than that of any other reported sites in small platinum clusters are much 1. Bell, A. T. Science 299, 1688–1691 (2003).
platinum-based catalyst for oxidative more active than a platinum surface for 2. Heiz, U. & Landman, U. (eds) Nanocatalysis (Springer, 2006).
3. Freund, H. J. & Pacchioni, G. Chem. Soc. Rev.
dehydrogenation of propane. Clearly, propane dehydrogenation. 37, 2224–2242 (2008).
the special activity is associated with the The work of Vajda and collaborators 4. Vajda, S. et al. Nature Mater. 8, 213–216 (2009).
tiny dimension of the catalyst particles. represents an important example of catalytic 5. Socaciu, L. D. et al. J. Chem. Phys. 120, 2078–2081 (2004).
To rationalize these findings Vajda and activity of mass-selected metal clusters 6. Sanchez, A. et al. J. Phys. Chem. A 103, 9573–9578 (1999).
colleagues performed density functional under realistic conditions of temperature 7. Abbet, S. et al. J. Am. Chem. Soc. 122, 3453–3457 (2001).
theory calculations on the reaction and pressure. The fact that the clusters 8. Benz, L. et al. J. Chem. Phys. 122, 081102 (2005).
9. Fan, C. Y., Wu, T. P., Kaden, W. E. & Anderson, S. L.
mechanism using a Pt8 cluster on an Al2O3 are both active and stable is a sizeable Surf. Sci. 600, 461–467 (2006).
model. The calculations show the existence step towards the design of nanocatalysts 10. Schaub, R. et al. Phys. Rev. Lett. 86, 3590–3593 (2001).
of a relatively small energy barrier for the of practical use. Although challenging, 11. Fallace, W. T. & Whetten, R. L. J. Am. Chem. Soc.
breaking of the first C–H bond, which is the development of nanometre and 124, 7499–7505 (2002).
Multiferroics
C
orrelated oxides are an
expanding field of study, with a
richness of fundamental physics
and phenomenal possibilities for
multifunctional applications1. Despite
[001]
their relatively simple structure, these
systems present a great complexity
of different properties, including [100]
superconducting as well as ferroic orders [010]
Nature Materials 8, 168–169 (2009); published online: 20 February 2009; corrected after print: 20 February 2009.
In the print version of this News & Views article the reference list should have included the following:
3. Spaldin, N. A. & Fiebig, M. Science 309, 391–392 (2005).
This reference is included in the HTML and PDF versions.
transition Metals
Daniel errandonea
i
n materials science and geophysics, it is transition metals are among the elements complete, and controversies exist regarding
essential to understand the transition most extensively studied at high pressures. the interpretation of different results.
between solid and liquid phases of Laboratory experiments conducted using In their study, Wu and colleagues
elements at high pressures. An accurate diamond-anvil cells (DACs) — devices that now analyse the role that shear stresses
picture of materials performance under consist of two opposing diamonds with a have on tantalum at extreme conditions
extreme conditions and detailed models sample compressed between their culets2 — before melting. Under shear stress, the
of planetary interiors both depend on have covered pressures up to 200 GPa shape of a material tends to change
this knowledge. On page 223 of this issue, and temperatures up to 5,000 K, thereby without volume change. In particular,
Christine Wu and colleagues1 address the mimicking Earth’s outer core3–5. Higher their molecular dynamics simulations —
long-standing issue of tantalum’s melting pressures and temperatures have also where atoms are allowed to interact for a
curve and show how, before melting, the been explored by taking advantage of the period of time by known physics, giving
studied materials might become a one- propagation of shock-wave fronts through a view of the atomic motion — suggest
dimensional, liquid-like glass. the studied metal and measurement of the that on plastic deformation beyond
Interest in the melting at high pressure sound speed6. In theoretical calculations, 3,250 K, solid tantalum becomes a one-
of transition metals such as iron and transition metals have been studied up to dimensional, viscous glass (1D glass),
tantalum stems from their importance to 400 GPa and 10,000 K (refs 7, 8). Despite in analogy to liquid glasses before they
understanding the inner properties of the such dramatic advances, knowledge about transform into an ideal liquid. Apparently,
Earth and other planets, because iron is the transition metals at extreme pressure the proposed picture is consistent with
most abundant element in their cores. Thus, (P) and temperature (T) is still far from DAC experiments4 and might provide an
explanation of the as yet unexplained low
melting slopes of transition metals.
a 298 K bcc b 3,250 K bcc c 1D glass d Glass To study the structural behaviour of
tantalum, Wu et al. applied a technique
widely used to identify amorphization
{111} plane
by monitoring changes in the radial
distribution function — a function that
describes how the density of surrounding
matter varies with distance from a
particular point. Their results (Fig. 1)
illustrate how solid tantalum, with a
–
{112} plane body-centred cubic (b.c.c.) structure,
becomes a 1D glass at high pressures and
temperatures. The main characteristic
of this glass is the merging of peaks in
the radial distribution function, which
indicates that the third and fourth nearest-
–
{110} plane neighbour distances in the {110} planes
become indistinguishable. Therefore,
tantalum is proposed to become a partial-
disordered structure with only long-range
order along the [111] direction, resembling
3 3,250 K 3,250 K 5,000 K
a smectic liquid (made of well-defined
layers that can slide along one another like
g(r)
2
soap). The calculations by Wu et al. also
1
suggest that on further heating, the 1D glass
0 transforms into an ideal liquid with only
2 4 6 2 4 6 2 4 6 2 4 6
r (Å) r (Å) r (Å) r (Å) short-range order. Basically, the molecular
dynamics simulations indicate that the
Figure 1 | Computer simulations comparing the distribution of atoms in different phases of tantalum at b.c.c. lattice of tantalum breaks down
30 GPa. Planar projections of tantalum atomic coordinates are shown as well as the corresponding radial before the actual melting. Shear strain
distribution functions g(r). a, Un-sheared b.c.c. crystal. b, B.c.c. crystal heated to 3,250 K without shear. provides the mechanical driving force for
c, 1D glass induced by shear and heating. d, 3D glass. the solid–1D-glass transition. This required
stress reflects the existence of an energy beneficial to understanding the structural hold generally for other transition metals,
barrier needed to overcome the resistance properties of tantalum beyond 3,250 K. probably even for iron. Such conclusions
of materials to plastic deformation, which Furthermore, recording of second-scale therefore could have important geophysical
is material-specific and depends strongly time-resolved X-ray diffraction patterns implications. It is possible that a new era
on pressure. This conclusion is supported during temperature cycling would in the study of high-pressure melting of
by the fact that in soft metals with isotropic provide valuable information9. These metals has begun. An explanation for
bonding, such as aluminium and copper, experiments will not only be appropriate different experimental observations and
the experimental melting temperatures to test Wu and colleagues’ findings, but theoretical predictions may finally be
agree with those calculated for the solid– also to check other hypotheses on the within our grasp. ❐
ideal-liquid transition. high-pressure behaviour of transition
Daniel Errandonea is with the MALTA Consolider
The proposed behaviour for tantalum metals. For example, the creation of local
Team, Fundación General Universidad de Valencia,
will not resolve all of the questions about structures in the liquid, owing to Peierls
46100 Burjassot, Spain.
its behaviour at extreme P–T conditions, and Jahn–Teller distortions10, are expected
e-mail: daniel.errandonea@uv.es
but could imply significant progress. In to lower the melting point of transition
particular, it does not necessarily explain metals on compression. In particular, X-ray references
why, in experiments performed under diffraction experiments can give valuable 1. Wu, C. J., Söderlind, P., Glosli, J. N. & Kepleis, J. E. Nature Mater.
different stress conditions (with the sample information on the radial distribution 8, 223–228 (2009).
immersed under soft or stiff pressure function of tantalum and similar metals. 2. Walker, J. Nature 265, 498–499 (1977).
3. Boehler, R. Nature 363, 534–536 (1993).
media), the same melting temperature is These experiments will be fundamental for 4. Errandonea, D. et al. Phys. Rev. B 63, 132104 (2001).
detected in DAC experiments4. However, distinguishing between the plastic flow of a 5. Errandonea, D. J. Phys. Chem. Solids 67, 2017–2026 (2006).
it clearly provides a new framework for liquid-like 1D glass and the free motion of 6. Hixson, R. S., Boness, D. A., Shaner, J. W. & Moriarty, J. A.
studying the P–T region constrained a ‘true’ or a ‘clustered’ liquid. Phys. Rev. Lett. 62, 637–640 (1989).
by 200 GPa and 5,000 K. The work of The most important implication of the 7. Belonoshko, A. B. et al. Phys. Rev. Lett. 100, 135701 (2008).
Wu et al. will, without doubt, stimulate proposed shear-induced melting is that 8. Taioli, S., Cazorla, C., Gillan, M. J. & Alfe, D. Phys. Rev. B
75, 214103 (2007).
further laboratory experiments and similar 1D-glass formation features are 9. Dewaele, A., Mezouar, M., Guignot, N. & Loubeyre, P.
computer calculations. In particular, new also found in other transition metals such Phys. Rev. B 76, 144106 (2007).
shock-wave measurements of the sound as molybdenum and vanadium, implying 10. Ross, M., Errandonea, D. & Boehler, R. Phys. Rev. B
speed between 100 and 200 GPa might be that this melting mechanism is expected to 76, 184118 (2007).
ePitaxial graPHene
Peter sutter
s
ince it was first isolated in 2004, perfect, macroscopically large graphene on a reconstructed interfacial layer on
graphene, a sheet of pure carbon sheets with uniform thickness, into which Si-face SiC, graphitization in vacuum has
just one atom thick, has generated active device structures can be carved. so far invariably transformed the ordered
a flurry of research activities. Although On page 203 of this issue, Emtsev et al. substrate surface (Fig. 1a) into a rough,
much of the initial ‘gold rush’ has focused demonstrate an important step towards highly corrugated landscape covered by
on the fascinating properties of this two- this goal for a specific fabrication strategy: small graphene domains with pronounced
dimensional crystal1 — which have as graphene epitaxy on silicon carbide (SiC)2. thickness variations (Fig. 1c). Such disorder
much to do with fundamental quantum Epitaxial growth on SiC (Fig. 1), a not only hinders further processing, it can
electrodynamics and particle physics as highly resistive material which would not also limit device performance by scattering
with solid state physics and materials shunt the current flow in graphene and charge carriers flowing in the active
science — researchers have recently begun which is already available in the form of graphene layer.
addressing the more mundane question of large-diameter wafers, is often seen as one The solution is quite obvious, at least
how some of these characteristics might of the most likely avenues to graphene- in principle. Roughness, excess surface
be harnessed in applications ranging based electronics. The graphitization steps as well as small-grained graphene
from post-Moore’s law electronics over of SiC by Si sublimation during high- are generated if a system cannot attain its
ultra-responsive sensors and actuators to temperature vacuum annealing was most favourable morphology because the
transparent solar cell contacts. Assuming demonstrated as early as the 1960s3, necessary microscopic processes — the
a continuation of ‘top-down’ processing and has recently been refined4. Even so, detachment, diffusion and reattachment of
similar to today’s silicon microelectronics, considerable challenges have remained. surface atoms — are suppressed. Increasing
the bottleneck clearly lies in synthesizing Although graphene grows with a well- the growth temperature should lift such
the required starting material: structurally defined orientation and its lattice fits well constraints. However, for graphitization in
vacuum, higher annealing temperatures substrate steps (Fig. 1d). The carbon sheets6, whereas both epitaxial graphene
also drive faster sublimation and higher layer grown on this surface now consists and supported exfoliated graphene
graphene growth rates, which in turn primarily of monolayer graphene, with show far lower values, raises questions
promote roughening and the nucleation of narrow bilayer (and occasional trilayer) about fundamental limitations to carrier
small graphene flakes. Progress can only stripes at the terrace edges. transport due to long-range scatterers in a
be made if mass transport on the surface is The key question, of course, is whether substrate or support. The nature of these
effectively decoupled from Si evaporation the improved sample morphology scattering centres is generally unknown,
away from the surface. Emtsev et al. translates to a reduced scattering of but the work by Emtsev et al. sets the
accomplished this feat by resurrecting charge carriers in the epitaxial graphene. stage for their identification for epitaxial
an 80-year-old trick originally developed Emtsev et al. begin to address this question graphene on SiC by reducing roughness
to extend the lifespan of incandescent by comparing the electron mobility — a and increasing the graphene domain size.
lightbulb filaments5. Instead of processing measure of the charge carrier drift velocity A careful comparison of transport parallel
their samples in vacuum, they placed in an applied electric field — for samples and perpendicular to substrate steps, for
them in a furnace with an unreactive prepared in vacuum and in ambient- instance, should reveal whether the steps
argon atmosphere at nearly ambient pressure argon, patterned into simple test themselves, or the two- and three-layer
pressure. The dense cloud of gas molecules structures. Their results are encouraging. domains near the steps, cause electron
hinders the transport of Si atoms away Samples produced by the new process show scattering. The complex buffer layer
from the SiC surface, reducing the overall a nearly twofold improvement over the sandwiched between the graphene and the
sublimation rate and allowing an increase previous record mobility in Si-face epitaxial SiC substrate needs to be considered as
in graphitization temperature by several graphene. But the findings also raise a another possible scatterer that would be
hundred degrees. Samples produced by number of issues. Particularly puzzling is difficult to eliminate.
this process have smooth surfaces made up the enormous difference — about a factor For some time now, epitaxial graphene
of large, flat substrate terraces that reach of 100 — in electron mobility between on SiC has been considered one of
tens of micrometres in length and several epitaxial graphene on SiC and exfoliated the leading contenders in the race to
micrometres in width, and are separated by graphene. The fact that the highest carrier find ways to mass-produce graphene,
steps several times the height of the original mobilities were achieved in suspended particularly for electronics. The improved
properties demonstrated by Emtsev et al.
should provide an additional boost
a b for this technology. But the race is still
Si-face
far from over. Competing approaches
have also seen rapid progress recently.
Transition-metal-catalysed graphene
growth, in particular, has been shown
to yield macroscopic single-crystalline
<1 µm SiC graphene domains with very low defect
C
density and outstanding thickness control7.
Si Still needed are methods to isolate the
carbon layers from a metal substrate. A
recent successful demonstration of the
c
transfer of large graphene flakes, using
growth on thin polycrystalline nickel films
that can be etched away to free the carbon
layers, underscores the viability of this
approach8. A true leap forward, however,
SiC
C-face would be the development of an atomically
precise ‘bottom-up’ synthesis of graphene
nanostructures. If the tools of materials
science, chemistry and nanotechnology
e
d G3 G2
are combined, such a vision might just
become a reality. ❐
G1
Peter Sutter is at the Center for Functional
~3 µm Nanomaterials, Brookhaven National Laboratory,
SiC Upton, New York 11973, USA.
e-mail: psutter@bnl.gov
references
1. Geim, A. K. & Novoselov, K. S. Nature Mater. 6, 183–191 (2007).
2. Emtsev, K. V. et al. Nature Mater. 8, 203–207 (2009).
3. Badami, D. V. Nature 193, 570–571 (1962).
4. Berger, C. et al. J. Phys. Chem. B 108, 19912–19916 (2004).
Figure 1 | Evolution of the surface during graphitization of Si-face SiC. a, SiC starting surface with a
5. Fonda, G. R. Phys. Rev. 31, 260–266 (1928).
staircase of flat terraces and atomic steps. b, Unit cell of 6H-SiC. c, Schematic morphology of vacuum-
6. Bolotin, K. I. et al. Phys. Rev. Lett. 101, 096802 (2008).
graphitized SiC. d, Morphology obtained in high-pressure argon. The surface termination is predominantly 7. Sutter, P. W., Flege, J.-I. & Sutter, E. A. Nature Mater.
monolayer graphene, G1 (e), with narrow stripes of bilayer (G2) and trilayer (G3) graphene near the upper 7, 406–411 (2008).
edge of the substrate steps. 8. Reina, A. et al. Nano Lett. 9, 30–35 (2009).
MeMBranes
t
he space occupied by living a system in which a thin membrane cylinder
organisms is determined by (tether) was pulled from a giant vesicle
membranes. Each cell is delimited (Fig. 1a). The pulling force, controlled
by a membrane, and additional experimentally, was compared with those
F F
membranes form the functional hierarchy obtained by membrane shape analysis. For
of compartments packed within it. The that, the membrane was parameterized
intracellular compartments, for example by an arbitrary set of control vertices,
vesicles and organelles, define the spatial excluding the areas of the external force
organization of metabolic pathways, thus b application (Fig. 1b). Then the force
providing the topological framework Boundary balance at each vertex was obtained using
Boundary
for cellular life. Recent advances in a proximal equilibrium approximation6,
structural techniques and super-resolution F F F that is, by making the virtual work
microscopy now allow imaging of F required to displace the vertices equal
intracellular membrane structures with to the corresponding energy of elastic
unprecedented resolution. Paul Wiggins deformations. The minimization was
and co-workers1 are, perhaps, first to conducted by conventional procedures5,7,
notice that the wealth of membrane Figure 1 | Calculation of forces acting on a taking into account elastic deformations
structures obtained awaits detailed membrane boundary from the membrane and the work of hydrostatic pressure
elastic modelling. shape. a, Illustration of a spherical membrane and lateral tension. Finally, they
The core of each membrane is the lipid vesicle with a cylindrical tether pulled from it obtained the external force acting at
bilayer, the bimolecular film consisting of by application of an external force (F). b, The the membrane boundary to cancel the
two monolayers of amphiphilic molecules, membrane shape obtained from images is internal ones exactly (Fig. 1b). This force
the lipids. This bilayer is primarily parameterized by control vertices (red dots); an correctly estimated the pulling force, as
responsible for the barrier function of effective boundary on which the external forces expected given the decades of studies
membranes. However, the lipids are also act is introduced (blue); small perturbations of on membrane shapes that have used the
deeply involved in regulation of membrane the vertices from their stationary positions (red giant vesicle prototype (see, for example,
shape and dynamics: both depend on the arrows) yield local internal forces due to elastic refs 9, 10). However, Wiggins and
material properties of the lipid bilayer 2,3. response of the membrane. The total internal colleagues emphasize that the proximal
One property is the elasticity, the resistance force (F, blue) projected to the tether axis is equilibrium approximation1 is tractable
to stretching and compression that equal to the external force (F, red) acting on for shapes of greater complexity because,
accompanies most of the deformations of the boundary. as is common for algorithms based on
lipid bilayers imposed by the specialized local balancing of forces and moments5,7, it
proteins that determine the shape of does not explicitly rely on assumptions of
cellular membranes. The lipid bilayer can averaged elastic parameters of the lipid shape symmetries generally used to analyse
be approximated as a continuous elastic bilayer. Different algorithms linking simple membrane vesicles.
surface whose behaviour is characterized by protein embedment into a membrane to This notion is yet to be tested.
a limited set of bulk material parameters4. elastic deformations have recently been Conceivably, it is even more important
Despite its relative simplicity, the elastic developed2,7,8, yielding phenomenological that the work of Wiggins’s group
approximation has been successfully used descriptions of membrane morphogenesis reiterates the importance of membrane
to explain a wide variety of membrane in cells. boundary in supporting a bulk elastic
phenomena, from the multitude of shapes The Wiggins group take a different stress11. This conceptualization could be
of red blood cells5 to the structure of approach. They assume that proteins act incisive when studying how particular
metastable intermediates of membrane externally, by applying forces required protein complexes control the shape of
fission and fusion6. to support a membrane shape from cellular membranes.
How can continuum elastic modelling its boundary. The force distribution Such proteins often have to restrain
help us to understand the action of can be calculated by analysing the large membrane areas and/or create a
specialized proteins creating cellular shape membrane shape within this boundary, substantial membrane stress, examples
(and thus how the shape is encoded) by the established algorithms of shape being the protein rings supporting the
at small, almost molecular scales? parameterization and localized force large structure of the tubular endoplasmic
Proteins can be simply considered as a balancing 5,7,9. The researchers tested reticulum. This membrane system makes
new bulk component effectively altering this approach on a simple experimental up the bulk of the internal membranes
Figure 2 | Hypothetical algorithm of calculation of forces shaping an organelle. a, Cartoon illustrating references
the endoplasmic reticulum shape controlled by rings of specialized proteins (blue). b, The shape is 1. Lee, H. J., Peterson, E. L., Phillips, R., Klug, W. S. &
parameterized by a set of vertices labelling ‘free’ membrane pieces. c, From the shape analysis, the forces Wiggins, P. A. Proc. Natl Acad. Sci. USA
(F) and torques (T) exerted by the rings are calculated1,9. d, Protein rings can also enforce membrane 105, 19253–19257 (2008).
narrowing, leading to shape instability and, ultimately, membrane fission13. 2. Zimmerberg, J. & Kozlov, M. M. Nature Rev. Mol. Cell. Biol.
7, 9–19 (2006).
3. McMahon, H. T. & Gallop, J. L. Nature 438, 590–596 (2005).
4. Seifert, U. Adv. Phys. 46, 13–137 (1997).
of a cell, and a substantial part of it of the processes, and because of the large 5. Jenkins, J. T. J. Math. Biol. 4, 149–169 (1977).
needs to remain freely accessible for the size of the proteins involved, the proteins’ 6. Chernomordik, L. V. & Kozlov, M. M. Annu. Rev. Biochem.
72, 175–207 (2003).
endoplasmic reticulum to communicate assembly can be seen as controlling 7. Agrawal, A. & Steigmann, D. J. Biomech. Model Mechanobiol.
with other organelles12. The forces applied membrane remodelling through a ring-like doi:10.1007/s10237-008-0143–0 (2009).
by these boundary rings can be revealed boundary 6,13 (Fig. 2d). 8. Brown, F. L. Annu. Rev. Phys. Chem. 59, 685–712 (2008).
by analysing the shape of the ‘free’ pieces The recent reconstitution of endoplasmic 9. Hess, S. T., Gudheti, M. V., Mlodzianoski, M. & Baumgart, T.
of the membrane (Fig. 2a–c). Protein reticulum morphogenesis and membrane Methods Mol. Biol. 400, 367–387 (2007).
assembly into ring-like structures is also remodelling in simple systems containing 10. Heinrich, V., Bozic, B., Svetina, S. & Zeks, B. Biophys. J.
76, 2056–2071 (1999).
responsible for shaping the intermediates a few proteins and lipid matrix 12,13 is 11. Capovilla, R., Guven, J. & Santiago, J. A. Phys. Rev. E
of membrane fusion and fission6. In these expected to make the membrane shape 66, 021607 (2002).
cases the need for multiple proteins is analysis proposed by Wiggins and 12. Hu, J. et al. Science 319, 1247–1250 (2008).
stipulated by the high energy requirements co-workers more straightforward. If 13. Bashkirov, P. V. et al. Cell doi:10.1016/j.cell.2008.11.028 (2008).
P
roteins are essential building blocks of life, forming a diverse genetic defects, or the inability of a tissue to provide its function
group of biological materials that ranges from spider silk because of the interaction with an ectopic or foreign material.
to bone, and from tendons to the skin, all of which play an Within the biological sciences, the field of genomics has
important part in providing key biological functions1–7. These advanced our knowledge base through the successful sequencing of
materials are distinct from the conventional categories of structure entire genomes. Extensive efforts have been made to move beyond
and material, as they represent the merger of these two concepts genomics, where fields such as systems biology provide explanations
through hierarchical formation of structural elements that range of mechanisms of how genes affect phenotypes and biological
from the nanoscale to the macroscale (Fig. 1a, b). function10,11. More recently, a movement to understand the materials
Protein materials are abundant in biology and play an essential science of proteins, aimed at developing relationships between
part in the biological function of all cells and tissues within structure (of biological protein materials), process (for example self
organisms. Many such materials with mechanical function form assembly) and property (for example strength or elasticity), shows
structural filaments, trusses or fibres, whereas others retain the great promise in contributing to a deeper understanding of biological
globular structure of their protein constituents. Biological protein systems. This study of material properties of hierarchical protein
materials are commonly characterized by where they reside with structures and their effect on molecular and microscopic properties,
respect to their associated tissue, to serve as either extracellular by using mechanistic insight based on structure–process–property
or intracellular protein materials1. Intracellular protein materials relations in the biological context (Fig. 1d), provides a basis for
provide the architecture of a cell and include vimentin, microtubules, understanding disease processes. This could help in developing new
actin (proteins from a cell’s cytoskeleton) and lamins (forming the approaches to treating genetic and infectious diseases, injury and
cell’s nuclear envelope)1–3. Extracellular protein materials, secreted trauma, as well as in improving engineered materials by translating
by cells into the surrounding microenvironment, include elastin material concepts from biology. Associated materials science
and collagen1,4,5. Other protein materials such as fibrin6–9 appear in challenges and opportunities are summarized in Table 1.
biological processes such as the clotting of blood. Distinct protein Materials in biology often involve non-protein components
materials in organisms cannot be considered in isolation. For such as crystal platelets (in bone, dentin and nacre, for example),
instance, lamins are connected to extracellular protein materials which have important effects on their mechanical properties.
(such as collagen) through cytoskeletal proteins (for example actin They also contain larger-scale structural features, which influence
and nesprin). the overall mechanical properties (for example porosities and
Here we focus on defining structural components of biological interaction with fluids). Although proteins are an important
protein materials from a materials science perspective, and on their element that affects most materials in biology, they alone do not
role in mechanical materials phenomena, specifically materials provide the information necessary to characterize all relevant
failure, in diseased or altered physiological conditions. Advances in material properties. The discussion of these effects is beyond the
experimental, theoretical and computational materials science have scope of this review.
led to a deeper understanding through the linking of structure to The cascaded arrangements of building blocks at defined
process and property (Fig. 1c). An overview of failure in biological length scales form hierarchical structures (for example molecules,
systems, broadly defined as the loss of its ability to provide an filaments, mesoscale structures) which control a material’s
intended physiological function, will be specifically presented in the properties. Figure 1a, b shows the structure of two example
context of material breakdown due to a range of causes, including materials, showing lamin intermediate filaments (in the following
altered chemical or physical boundary conditions (such as extreme referred to as ‘lamins’)12–15 and collagen (for example tendon and
forces), weakening of tissues due to structural flaws because of cornea)4,5, and illustrates how structure and material converge.
1
Laboratory for Atomistic and Molecular Mechanics, Department of Civil and Environmental Engineering; 2Center for Computational Engineering; 3Center
for Materials Science and Engineering, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, USA.
4
Laboratory for Cell and Tissue Engineering, Harvard University, School of Engineering and Applied Sciences, 40 Oxford Street, 415 ESL, Cambridge,
Massachusetts 02138, USA. *e-mail: mbuehler@MIT.EDU
O(10 µm)
a O(1 nm) O(1 µm)
α-Helix
Lamina mesh Cell nucleus
network (blue colour)
O(100 µm)
O(10 µm)
b O(1 nm)
O(300 nm) O(1 µm)
O(1 cm)
c Process d Process HN
Structure Process H0
(transcription, Property H1
Requirement translation)
Requirement RN
Property H0
Structure H0 Requirement R1
Structure H1 Requirement R0
Structure HN
Figure 1 | Examples of hierarchical multiscale structures of biological protein materials. Here we show lamin intermediate filaments (lamins),
collagenous tissues (tendon) and the materials science paradigm for the analysis of biological protein materials. a, Structure of lamins. α-Helical protein
domains assemble into dimers, which form filaments that define a lattice-like lamin network of the cell’s nucleus. Lamina mesh network snapshot adapted
and reprinted with permission from ref. 16. © 1986 NPG. Cell nucleus image ©TenOfAllTrades. b, Structure of collagenous tissues, from nanoscale
to macroscale. Tropocollagen (TC) molecules assemble to form fibrils, which form fibres that provide the structural basis of collagen tissues such as
tendon. Collagen fibril image reprinted with permission from ref. 76. © 2002 Elsevier. Image of Achilles tendon from SPL. c, Materials science triangle
that links structure, process and property. d, Materials science paradigm applied to the hierarchical structure of protein materials (Hi refers to hierarchy
levels i = 0 to N, Ri refers to material property requirements at hierarchy levels i = 0 to N). The cycle initiates at H0 (process H0, the only level at which
protein expression occurs) to form protein constituents (structure H0). Their properties (property H0) control the association at the next hierarchical level
(process H1 leading to structure H1). This cycle continues through all hierarchical levels i = 0 to N, where process and structure H1 and beyond denote
protein assembly stages. At each stage, the properties of the structure control the assembly at the next level. Overall, properties at different hierarchical
levels (properties Hi) are regulated by corresponding physiological demands (requirements Ri), which are sensed and transduced intracellularly to activate
genetic regulation, resulting in changes to process H0.
Lamins provide structural support to the cell’s nucleus and of strain-regulated mechanisms of cell response falls broadly into
form an interface between the cell’s cytoskeleton and chromatin. the field of mechanotransduction3,23. Lamins represent a model
Figure 1a shows the structure of lamins12–15, characterized by a material that illustrates the intimate connection between structural
hierarchical assembly of α-helix-based protein domains. The basic material and biochemical properties. Recent studies have provided
building block of lamins is an α-helical coiled-coil dimeric molecule. substantial evidence that underlines the role of lamins in cancer and
Lamin dimers form filaments that develop into a network with a genetic diseases17,21,22,24.
lattice-like structure, found primarily at a cell’s nuclear membrane16. Collagen, the most abundant protein on the Earth, is a fibrous
Because the cell’s cytoskeleton itself is coupled to the surrounding structural protein with superior mechanical properties. It is found in
extracellular matrix (for example collagen) by adhesion proteins, a number of tissues exposed to severe tensile or compressive loading,
macroscopic deformation of tissues typically leads to deformation including tendon, bone, teeth, cartilage and the cornea4,25–29. The
of the cell nucleus17–20. The lamin network thus aids the coupling hierarchical structure of collagen is summarized in Fig. 1b. Collagen
of tissue-level mechanical signals to complex biochemical processes consists of triple helical tropocollagen molecules that appear as
(such as gene regulation) within the cell nucleus16,17,21,22. The study ‘nano-ropes’ with lengths of around 300 nm (ref. 30). Staggered
arrays of tropocollagen molecules form fibrils, which combine to intrinsic ability to repair itself 36. The underlying mechanism is that
form collagen fibres and then to form tissues such as tendon, which small cracks, formed because of physiological mechanical load, are
functions as glue between muscle and bone. Tropocollagen molecules detected by ‘bone remodelling units’ and removed from the tissue
show extreme extensibility, withstanding up to 50% tensile strain before they reach a critical size at which catastrophic failure would
before catastrophic failure. This, along with their ability to sustain occur. Such mechanisms show the intricate connections between
stresses of several gigapascals, is crucial for their physiological tissue formation and overall failure properties. In vivo, the structures
mechanical role in tissues31. The staggered arrangement of molecules of protein materials self-maintain or adapt by means of feedback
into fibrils provides the basis for its ability to dissipate mechanical loops by translating spontaneous demands in the microenvironment
energy through molecular sliding rather than leading to catastrophic (through intracellular signalling) to regulate gene activation or
failure32–34. This staggered architecture plays a key part in increasing deactivation. Ultimately, this alters the material’s structural makeup
the toughness of various collagen materials such as tendon or bone. to better suit the local physiological needs (Fig. 1d). Multifunctionality
The overall hierarchical arrangement is important because, through refers to the material’s ability to provide multiple functions such as
structures formed at characteristic geometric length scales, it enables mechanical strength and the ability to control biochemical processes
superior molecular properties to be visible at larger, biologically (as observed in lamins that have structural as well as biochemical
important, intermediate mesoscales. A possible concept to explain roles). Many protein materials show graceful degradation, reflecting
the observed length scales at each hierarchical level is that they are the material’s ability to induce a controlled breakdown of a system’s
a result of structural adaptation towards maximizing target material function after damage without leading to catastrophic failure and
properties (such as strength and dissipation) through geometric without affecting materials in the environment. The hierarchical
size effects, as discussed in the context of bone35. Several genetic structure may be the basis for these unique properties, distinguishing
diseases are related to defects in the collagen structure, leading to them from many engineered materials.
mechanically compromised tissues (an example is the brittle bone The formation of protein materials in biological systems occurs
disease osteogenesis imperfecta). through self assembly of protein molecules whose structure is
The biological materials introduced above feature properties encoded in an organism’s DNA. Protein molecules are composed
that are characteristic for this entire class of materials, including of long polymer chains, constructed of amino acid monomers,
robustness, adaptability and multifunctionality 4,11. Robustness is and synthesized by a translation process that converts the genetic
defined as the degree of separation between stability and failure, information transcribed from DNA into RNA into a polypeptide
and is crucial to understanding the role of protein materials in sequence of amino acids. Protein synthesis is initiated at precise
biology, for both physiological and pathological conditions (such sites during mRNA translation and terminated when signalled by
as varying pH, forces or structural changes). Adaptability refers the presence of encoded stop codons (messages). Once a protein
to the ability of a material to cope with environmental changes by molecule is constituted, the polypeptide chain folds into its unique
changing its structural arrangement to cope better with changed three-dimensional (3D) conformation with the aid of chaperone
conditions. For example, tissue remodelling in bone plays a crucial proteins1. Functional activation of the protein structure often
part in improving the material’s damage tolerance through its involves binding with other molecules to form the protein’s intended
DNA nucleotides
define type of AA
Example: Proline -
DNA
nucleotides
b Radial growth c
Lamin
Axial growth
dimers Banded structure
I II III
d e
PO42–, K+,H+
Figure 2 | Formation and assembly of biological protein materials, including cellular expression of proteins and assembly into larger-scale hierarchical
structures. a, Generation of protein molecules through translation from the DNA genetic code, leading to the expression of individual 3D protein
molecules. b, Schematic biological assembly of lamins14 and experimental visualization. The self assembly of protein molecules into large-scale structures
proceeds through a complex process that involves several intermediate structural steps. For lamins, assembly occurs first by association of two dimers
into tetramers, then by a period of growth in the axial direction (I), followed by growth in the radial direction (II and III). Experimental visualizations of
the assembly process of lamins are obtained from electron microscopy studies (rat liver lamin A/C, scale bar 100 nm). Adapted and reprinted with
permission from ref. 14. © 2004 Annual Reviews. c, Experimental visualization of the assembly of tropocollagen into collagen microfibrils based on
contact-mode atomic force microscopy height images. The characteristic regular-spaced banding structure of collagen microfibrils is visible in the marked
region where multiple tropocollagen molecules begin to interact (black, 0 nm height, white, 35 nm height). Adapted and reprinted with permission from
ref. 76. © 2002 Elsevier. d, Schematic diagram of the biological process of formation of spider silk protein filaments in a spider’s spinning duct, involving a
controlled change of pH, solvent chemicals and mechanical shear, leading to the formation of the β-sheet-rich spider silk filaments39. e, Synthetic process
of spider silk formation, realized in a microfluidic device (left), together with a micrograph (right) of the resulting synthetic fibres (scale bar 10 μm).
Adapted and reprinted with permission from ref. 39. © 2008 National Academy of Sciences USA.
functional structure1 (Fig. 2a). In vivo, many protein materials obtain posing a challenge in the development of biomimicking and
their specific functionality by post-translational modifications, biocompatible materials.
which include hydroxylations (for example during formation of The resultant individual protein components self-assemble at
tropocollagen’s hydroxyproline residues), phosphorylations and different timescales to form a protein structure with hierarchical
glycosylations (for example during formation of cartilage tissues), as geometrical entities (Fig. 1a, b). Assembly mechanisms have been
well as enzymatic crosslinking. These modifications are particularly investigated based on the combination of imaging tools with
important for the material properties of tissues, as they control the controlled assembly conditions (pH, temperature, solvent). This
interaction between proteins and with other material components approach has elucidated the mechanism of lamin assembly 14,37
(such as inorganic materials or sugar-based components) as well (Fig. 2b). In this process, lamin dimer association and axial growth
as their bioactive properties. These modifications are particularly are followed by radial growth to form larger fibrils, resulting
difficult to mimic ex vivo or through synthetic approaches, in formation of a network. Figure 2b also shows the direct
Length
science
AH Protein, enzymes,
of protein
BH filaments (e.g. dimer,
materials
Universality: protein motifs (AH, BS, RC) tetramer)
Single-
O(µm, µs)
molecule
Protein motif
Universality
ACGT DNA
20 AAs methods
chemistry Double helix AH, BS, RC
Figure 3 | Universality and diversity of biological protein materials. a, Degree of structural diversity and universality in protein materials, plotted against
biological length scales (equivalent to hierarchical levels). Inset, a visualization of the topoisomerase protein (biological role is to cut strands of the DNA
double helix; visualization using visual molecular dynamics140). This inset illustrates how universal motifs define the overall functional properties of this
protein, whereas the entire protein structure represents diversity. See also Table 2. AH, α-helix; BS, β-sheet; RC, random coil; AA, amino acid. b, Placement of
mesoscale science of protein materials, between single-molecule studies and micromechanical and continuum-type biomechanics applications.
Table 2 | Occurrence of protocols in four sample materials, illustrating that some protocols are universal to all protein materials
whereas others are specific.
Protocol Biological protein material
Collagen (tendon) Lamin Spider silk Amyloids
DNA nucleotides* (ACGT) ✓ ✓ ✓ ✓
Vimentin DNA sequence† ✓
Collagen DNA sequence† ✓
Spider silk DNA sequence† ✓
Amyloid DNA sequence† ✓
DNA double-helical structure* ✓ ✓ ✓ ✓
Protein building blocks* (20 AAs) ✓ ✓ ✓ ✓
Motifs* AH, BS, RC ✓ ✓ ✓ ✓
Collagen fibril† ✓
Lamin filaments† ✓
β–Nanocrystal composite† ✓
Cross-β fibril structure† ✓
*Universal; diverse. AH, α-helix; BS, β-sheet; RC, random coil; AA, amino acid.
†
experimental visualization of these assembly processes. A snapshot ex vivo, as has been demonstrated for spider silk fibres in microfluidic
of the assembly process for collagen fibrils is shown in Fig. 2c, devices39 (Fig. 2e).
revealing the characteristic banding structure of collagen fibrils The ability to control self assembly synthetically and genetically
when multiple molecules assemble into microfibrils. The biological provides a powerful framework for the study of the links between
process of in vivo assembly of many structural protein materials structure, property and process. The approach of adding and
involves a dynamic change of physical and chemical conditions, as deleting molecular domains with distinct chemical functionality
has been shown for the synthesis of spider silk38,39 (Fig. 2d). has made possible the detailed study of the mechanisms of self-
In addition to studies of naturally occurring biological protein assembly processes (for example identification of essential protein
materials, recent research led to the development of techniques domains)47–50. This allows one to determine how changes to the
that enable one to change their structural makeup, and to design protein structure alter biological function and disease properties.
and synthesize synthetic analogues40–44 through recombinant DNA The evolution of protein materials through genetic selection
techniques, RNA interference knockdowns or sequence insertions. and structural alterations has resulted in a specific set of protein
Two primary routes of development pursued include ex vivo building blocks that define their structure. Protein materials exist
assembly (for example self-assembling peptide systems40,41,44) and in abundant variety, and the need exists to formulate a widely
in vivo expression of protein materials (for example through bacterial applicable model to categorize all such materials systematically, in
hosts42,45). The ability to control the DNA sequence information at order to establish a fundamental understanding, and to make use of
a fundamental level provides us with the ability to engineer the hierarchical structural building blocks to develop a new generation
structure of protein materials at the molecular (amino acid) scale. of advanced nanomaterials51–53. A protocol is defined here as a
EAK16, for example, belongs to an interesting class of self-assembled term that encompasses a general analysis of protein materials that
peptides that constitutes a material platform for a variety of biological, describes the use of structural building blocks (for example α-helices,
biomimetic and nanotechnology applications46. Knowledge of the β-sheets, random coils) during their formation and function, and the
details of the in vivo processes enables one to mimic these processes process or mechanism of use of this material (for example synthesis,
Continuum
models
s
O(Å, ns)
Meso-
scale
models Optical/magnetic tweezers
Timescale
µs
Non-
reactive Atomic force microscopy b
MD 10
ns Reactive
450 nm
Width (nm)
MD
280 nm 5
Tomography
500 nm
QM/DFT LIF
ps
Transmission elctron
0
microscopy (e.g. cryo)
Figure 4 | Tools for characterization and modelling of deformation and failure of biological protein materials. Respective timescale and length-scale
domains of applicability are shown. Experimental methods include X‑ray diffraction, transmission electron microscopy, atomic force microscopy (AFM),
optical/magnetic tweezers; microelectromechanical systems (MEMS) testing and nanoindentation. Theoretical and simulation tools include quantum
mechanics/density functional theory (QM/DFT), molecular dynamics (MD), coarse-grained models and mesoscale atomistically informed continuum
theories, as well as continuum models. The lower part of the figure indicates classes or scales of protein materials that can be studied with these types of
technique. a, Coarse-graining process in the development of mesoscale models, here illustrating the representation of the full-atomistic description of a
tropocollagen molecule by a collection of ‘super-atoms’ or beads, reaching length scales and timescales several orders larger. b, Deformation of a lamin
intermediate filament (LIF) using AFM. Adapted and reprinted with permission from ref. 141. © 2007 NPG. c, Tensile stretching experiment of a collagen
fibril with a MEMS device (scale bar 2 μm). Reprinted with permission from ref. 82. © 2006 Royal Society.
Mass spectroscopy provides information about the chemical quantitative link between cellular-scale deformation and underlying
content of protein materials. Fourier-transform infrared (FTIR) molecular-scale rupture mechanisms. The basic concept of this
spectroscopy is a related approach that elucidates information about approach is that protein unfolding due to rupture leads to the
chemical bonding and is an established tool used for the structural exposure of binding sites, which can be used to bind fluorescent
characterization of proteins62,63. Developments in single-molecule markers for imaging. This technique has shown that protein rupture
force spectroscopy have made it possible to study chemical, occurs as an underlying mechanism in cells under mechanical
mechanical and structural properties using manipulation techniques load78. Nanoindentation is another method used to probe the
such as atomic force microscopy (AFM) and optical/magnetic mechanical response of protein material composition and structure
tweezers20,64–72. The force-measuring capability of AFM also makes it at multiple length scales79, and has allowed the characterization of
possible to elucidate molecular determinants of mechanical stability multiscale mechanical behaviour, as demonstrated for bone80,81. The
and the role of force-induced conformational changes of proteins development of devices based on microelectromechanical systems
in the regulation of physiological function73,74, and it has been used (MEMS) provides the ability to carry out tensile tests of micrometre-
to image the surface structure and assembly patterns (for example sized samples, as demonstrated for individual collagen fibrils82. The
in bone and collagen72,75–77). AFM has also been effectively used in available experimental methods cover a broad range of timescales
combination with fluorescent techniques for imaging, as reported and length scales. Remaining challenges include specificity with
in studies of fibrin stretching 6. In a recent study, in situ fluorescent respect to the handling of individual molecules and filaments,
labelling was used to visualize unfolding of cytoskeletal proteins in their control at very short timescales, and the ability to observe
cells under different mechanical loading conditions78, providing a mechanisms directly at molecular scales.
Intramolecular H-bonds
40 as shown in Fig. 5a (inset)55. Similar hydrogen‑bond failure
mechanisms have been observed in other proteins, such as β-sheet
Energy
H-bond
30
I I–II II
rupture crystals55. Studies analysing the deformation and failure mechanisms
r 20 Molecular of tropocollagen molecules found that initial molecular rotation is
rotation followed by hydrogen‑bond breaking and subsequent stretching
II 10 Backbone
I I–II r stretching of the protein’s polypeptide backbone, followed by covalent bond
0
3D path 0 20 40 rupture, explaining collagen’s characteristic stiffening 31,86 (Fig. 5b).
Tensile strain (%) Molecular dynamics simulations have also been useful in defining
experimental protocols for in situ fluorescent labelling by identifying
c d
2,500 key rupture mechanisms and exposing domains78.
Experimental (titin 127) Larger length scales and timescales can be reached through
F F 2,000 MD simulation (127)
the use of a hierarchy of simulation techniques, integrated
Rupture force (pN)
Number of events
30 132
25
j 20 3,132
Fij 15
i ∆xapp= 0.125 nm
10
5
FIm
I 3
15
6,221
m 10
∆xapp= 0.09 nm
5
221
j 0
Rupture strength (pN)
6
b 14 c Computational
d
Experiment (type II TC) WLC
12 Experimental
Experiment (type I TC) 600
10 Titin filament
Young’s modulus (GPa)
MD 8 500 Theory
Force (pN)
Figure 6 | Integration of theory, simulation and experimental techniques. a, Results of an elastic network model calculation, used to describe the
anisotropic failure properties of a larger protein structure. Reprinted with permission from ref. 98. © 2008 APS. b, Entropic elastic behaviour of stretching a
single tropocollagen (TC) molecule, comparing experiment (optical tweezers), coarse-grained mesoscale simulation (molecular dynamics, MD) and theory
(worm-like-chain model, WLC). Data replotted from refs 31 and 103; x/L is the ratio of extension x to contour length L. c, Young’s modulus prediction of a
single tropocollagen molecule. The graph shows a comparison between experimental and full-atomistic molecular dynamics simulation results (based on
averages from literature sources summarized in ref. 86; error bars correspond to standard deviation). d, Comparison between experimental and theoretical
modelling of stretching a long titin filament, obtained through the application of a theoretical approach that involves the Bell model and a worm-like-chain
description for protein elasticity. Reprinted with permission from ref. 104. © 1998 APS.
include a description of the transition between multiple transition A multiscale analysis that explicitly calculates or measures material
states, effectively describing the occurrence of several straight lines properties at multiple scales is compulsory.
in the F–ln(v) domain91–96 (similar to the overall behaviour shown Despite these challenges, several studies have led to direct
in Fig. 5d). comparison between experiment and theory. Protein strength
Challenges in linking simulation and experiment include models have been extended to more complex protein geometries,
differences in accessible timescales and length scales. For example, a vital step to arrive at structure–process–property relationships of
the computational prediction of the 3D folded structure of proteins strength properties98–100. Figure 6a shows an elastic bond network
directly from the amino acid sequence is, despite advances in recent model for protein unfolding mechanics, which combines an elastic
years, still at a relatively early stage and can typically only be applied model of a network of bonds with a bond fracture model based
reliably to simpler protein structures, as the timescales for protein on Bell’s concept 98. The model, here applied to green fluorescent
folding are beyond current computational capabilities97. protein, treats bonds as springs and provides a direct structure–
Because biological protein materials feature properties that property relationship between the geometry and bond rupture
are strongly length-scale and timescale dependent, the explicit (for example the protein’s mechanical anisotropy). It is capable of
consideration of particular scales for the quantitative comparison accurately reproducing experimental results reported earlier for the
of measurements and theory is essential. Measurements at different same protein structure101 (for comparison see Fig. 6a).
timescales and length scales lead to significant variations in material A major success is the use of the worm-like-chain model to
properties. For timescales, this can be observed in the results of describe entropic elasticity 69,102. The use of Bell-type strength
strength properties reviewed in Fig. 5d55,67,70. This plot explains why models in conjunction with worm-like-chain models resulted in
many molecular dynamics simulations predict a vastly different quantitative agreement between theory and experiment. Figure 6b
behaviour from experimental studies, because they are carried out depicts the comparison of an optical tweezers experiment of
at different timescales. An implication of these strong size effects stretching a single tropocollagen molecule with results from
is that one cannot directly use single-molecule measurements to molecular dynamics simulation (based on the mesoscale
infer larger-scale properties, and that homogenization approaches coarse-grained model shown in Fig. 4a) and the theoretical
commonly used for crystalline materials cannot be applied directly. worm-like-chain model31,102,103. All three approaches provide a
Tensile load
Fractures
Localized failure
d Mutation site e
Tumour cells Healthy cells
30 15
14 100 20 10
Max. stress (MPa)
Max. strain (%)
12
Counts
80
10
8 60
10 5
6 40
4
2 20
0 0 0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
e
I
O
nc
re
re
fe
fe
Re
Re
Figure 7 | Role of changes in biological protein material properties in diseases. a, Structure of amyloids, spanning from the chemical structure
of individual hydrogen bonds to the scale of amyloid plaques. Included is a visualization of molecular models proposed for amyloid fibrils (‘cross-
β-structure’ and ‘β-helix’). Protofilament structure reprinted with permission from ref. 143. © NPG. b, Effect of lamin mutations in progeria on the
mechanical properties of the cell nuclear membrane (scale bar 5 μm, fluorescent imaging). The cell nuclear membrane forms localized fractures under
large mechanical loading. c, Change in the lamin microstructure due to genetic mutation, and effect on deformation mechanism (b and c reprinted
with permission from ref. 22. © 2006 National Academy of Sciences USA). d, Geometry of point mutations in osteogenesis imperfecta (OI) in the
tropocollagen molecule (upper part), and effect on mechanical properties of tendon, where the maximum strain and maximum stress of tendon are
severely reduced under mutations (lower part, data plotted from ref. 132). e, Mechanical signature of cancer versus normal cells, revealing that cancer
cells are softer than normal cells. Adapted and reprinted with permission from ref. 136. © 1997 NPG.
Magnetic atoms at surfaces are a rich model system for (2D) arrays of regularly spaced Fe spins, the magnitude and
solid-state magnetic bits exhibiting either classical1,2 or magnetic anisotropy of which are manipulated by lateral and
quantum3,4 behaviour. Individual atoms, however, are difficult axial molecular ligands. Although Fe atoms adsorbed on Cu(100)
to arrange in regular patterns1–5 . Moreover, their magnetic constitute a metallic system owing to the substantial overlap
properties are dominated by interaction with the substrate, of the Fe 3d and 4s states with the Cu bands, and exhibit a
which, as in the case of Kondo systems, often leads to Kondo temperature of about 55 K (ref. 7), X-ray absorption
a decrease or quench of their local magnetic moment6,7 . spectroscopy (XAS) and multiplet calculations shows that Fe–Cu
Here, we show that the supramolecular assembly of Fe and hybridization is drastically weakened by square-planar coordina-
1,4-benzenedicarboxylic acid molecules on a Cu surface results tion to 1,4-benzenedicarboxylate molecules (hereafter indicated as
in ordered arrays of high-spin mononuclear Fe centres on a terephthalate acid or TPA), leading to a high-spin configuration
1.5 nm square grid. Lateral coordination with the molecular with strongly localized Fe2+ character. X-ray magnetic circular
ligands yields unsaturated yet stable coordination bonds, dichroism (XMCD) measurements confirm the presence of sizeable
which enable chemical modification of the electronic and spin moments at the Fe sites, together with a large orbital moment
magnetic properties of the Fe atoms independently from the and in-plane magnetic anisotropy that originate from spin–orbit-
substrate. The easy magnetization direction of the Fe centres induced mixing of ground and first excited molecular states. The
can be switched by oxygen adsorption, thus opening a way to Fe centres in the supramolecular arrays act as selective adsorption
control the magnetic anisotropy in supramolecular layers akin sites for O2 , which produces an out-of-plane reorientation of
to that used in metallic thin films8–11 . the easy magnetization axis at the Fe sites. Furthermore, the
The control of magnetic anisotropy is a key issue in the saturation of the orbital magnetic moments of TPA-coordinated
development of molecule–metal interfaces for magnetic applica- Fe is found to be strongly anisotropic, revealing an unusual feature
tions, both at the single-molecule12 and extended-film level13 . In of surface-supported magnetic systems.
metallic multilayers used as storage media or spin-valve devices at The co-deposition of transition-metal ions and organic ligands
present, tuning of the magnetic anisotropy is achieved either by a on crystalline surfaces offers the potential to design supramolecular
careful choice of the overlayer/substrate composition and thickness grids with programmable structural and chemical features19,20 ,
or by oxidation of the magnetic elements11,14,15 . Recent studies where the interaction with the substrate is used to stabilize a
showed that the magnetization direction of surface-supported para- planar geometry21,22 . Fex (TPA)y complexes on Cu(100) constitute
magnetic molecules can be controlled through exchange coupling a prototypical 2D hetero-assembled system forming a variety of
with a magnetic film, which provides robust ferromagnetic proper- mono- and bi-nuclear network structures, in which the morphology
ties but does not enable each molecule to be switched independently is determined by the Fe:TPA stoichiometry, substrate symmetry
from the substrate or its neighbours16,17 . Alternatively, theoret- and annealing temperature22–24 . Figure 1a shows a scanning
ical work suggested that the sign of magnetic anisotropy could tunnelling microscopy image of the square-symmetry array of self-
be reversed in metal–organic complexes by exploiting oxidation assembled mononuclear Fe(TPA)4 complexes investigated here.
processes that affect the hybridization of molecular orbitals with Each Fe atom is coordinated to four TPA molecules through
metal states carrying non-zero orbital magnetization18 . Fe–carboxylate bonds, with the supramolecular Fe(TPA)4 units
Here, we investigate supramolecular self-assembly on a organized in a (6 × 6) unit cell with respect to the underlying
non-magnetic Cu surface as a means to produce two-dimensional Cu lattice (Fig. 1c). Weak hydrogen-bonding interactions between
1 Centre d’Investigacions en Nanociència i Nanotecnologia (ICN-CSIC), UAB Campus, E-08193 Barcelona, Spain, 2 Institució Catalana de Recerca i Estudis
Avançats (ICREA), E-08010 Barcelona, Spain, 3 Institut de Physique des Nanostructures, Ecole Polytechnique Fédérale de Lausanne (EPFL), CH-1015
Lausanne, Switzerland, 4 Max-Planck-Institut für Festkörperforschung, D-70569 Stuttgart, Germany, 5 Department of Applied Physics, Chalmers University
of Technology, 41296 Göteborg, Sweden, 6 Department of Chemistry, Utrecht University, 3584 CA Utrecht, The Netherlands, 7 Solid State and Structural
Chemistry Unit, Indian Institute of Science, Bangalore 560 012, India, 8 European Synchrotron Radiation Facility, BP 200, F-38043 Grenoble, France,
9 Institut de Minéralogie et de Physique des Milieux Condensé, Université Pierre et Marie Curie, F-75252 Paris, France, 10 Physik-Department E20,
θ = 0° θ = 0° θ = 0°
XAS intensity (arb. units)
I+ 1
1 I¬ 1
705 710 715 720 725 730 705 710 715 720 725 730 705 710 715 720 725 730
Photon energy (eV) Photon energy (eV) Photon energy (eV)
d e
Experiment
Calculation
I+ + I¬0°
I+ + I¬0°
XAS intensity (arb. units)
I+ + I¬70°
I+ + I¬70°
I+ ¬ I¬0°
I+ ¬ I¬0°
705 710 715 720 725 730 705 710 715 720 725 730
Photon energy (eV) Photon energy (eV)
Figure 2 | Circularly polarized X-ray absorption spectra of Fe(TPA)4 , O2 –Fe(TPA)4 and Fe/Cu(100) measured at the Fe L2,3 -edge and calculated
multiplet structure. a–c, XAS intensity of 0.025 monolayer Fe/Cu(100) (a), Fe(TPA)4 (b) and O2 –Fe(TPA)4 (c) for parallel (I+ ) and antiparallel (I− )
orientation of the photon helicity with field-induced magnetization. Spectra were recorded in the electron-yield mode at T = 8 K and B = 6 T at normal
(θ = 0◦ ) and grazing incidence (θ = 70◦ ). The XMCD signal (I+ –I− ) is shown for each orientation. d,e, Calculated XAS and XMCD spectra of Fe(TPA)4 (d)
and O2 –Fe(TPA)4 (e). See text for details.
agreement with the ligand-field simulations. There are 0.8 excess axis out-of-plane. The field dependence of the XMCD signal in
electrons localized on the O2 ligand, the largest fraction of which Fig. 3a,b underlines the strong change in magnetic anisotropy
is supplied by the Cu substrate acting as a charge reservoir, a induced by O2 . This effect is comparable to that reported for
behaviour that is specific to surface-supported metal–organic sys- ultrathin metal films9–11 , but opposite in sign with respect to
tems. Notably, the formation of the supramolecular complexes oxygen-dosed Fe layers on Cu(100) (ref. 11). With respect to bulk
is accompanied by a substantial increase of the Fe–substrate dis- molecular crystals, the planar and open coordination structure
tance, calculated as 2.32 Å for individual Fe atoms on Cu(100), of the self-assembled Fe array makes such a system extremely
2.71 Å in Fe(TPA)4 and 3.32 Å in O2 –Fe(TPA)4 , showing together sensitive to chemisorption, providing straightforward control of
with XAS how the Fe–Cu metal interactions progressively weaken the preferred Fe spin orientation. Through the analysis of the
with increasing number of ligands. A more detailed discussion of XAS and XMCD spectra, we can further identify the cause for
the crystal-field and DFT electronic-structure results is reported in the easy-axis switch in the metal–organic complexes, revealing
Supplementary Information. a different mechanism from the one proposed by Atodiresei
Confirmation of the strong magnetic anisotropy behaviour of Fe and co-workers18 based on the control of the metal oxidation
in the supramolecular complexes was obtained by angle-dependent state. Our calculations show that only the ligand field is affected
XMCD measurements, exploiting the XMCD proportionality to by O2 adsorption, whereas the formal Fe oxidation does not
the projection of the Fe magnetic moment on the X-ray incidence change. As demonstrated by the crystal-field level schemes in
direction. The relative magnitude of the XMCD spectra at θ = 0◦ Fig. 3c,d, the axial O2 ligand induces a change of the Fe ground
and 70◦ in Fig. 2 indicates that, whereas Fe/Cu(100) has weak state from A1g to Eg . This effect can be understood also in
out-of-plane anisotropy, the Fe(TPA)4 easy axis lies in-plane. O2 an intuitive way, as the on-top O2 molecule binds axially to
adsorption at the Fe sites, however, further drives an abrupt the dz 2 Fe orbital, thus effectively pushing up the antibonding
magnetic anisotropy reorientation transition, rotating the Fe easy A1g state (that is, the term with dz 2 -like symmetry) relative to
a b
0.5
In-plane 0.8 Out-of-plane
0.4
0.4
0.2 Out-of-plane
In-plane
0.2
0.1
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
B (T) B (T)
c d
0.5 eV B1g(dx2 ¬ y2) B1g(dx2 ¬ y2)
B2g(dxy)
A1g(dz2)
5D
5D 2
2
B2g(dxy)
Eg(dxz, dyz)
Eg(dxz, dyz)
A1g(dz2)
O3 Oh C4v O3 Oh C4v
Figure 3 | Element-selective magnetization curves of the Fe centres and crystal-field diagrams. a,b, Fe(TPA)4 (a) and O2 –Fe(TPA)4 (b) magnetization
measured at T = 8 K with the field applied out-of-plane (θ = 0◦ , filled symbols) and close to the in-plane direction (θ = 70◦ , open symbols). The data
points represent the Fe XMCD intensity integrated over the L3 -edge and normalized by the corresponding L3 -integrated XAS intensity. c,d, Crystal-field
diagram of Fe(TPA)4 (c) and O2 –Fe(TPA)4 (d) derived from the XAS multiplet simulations.
the Eg term in O2 –Fe(TPA)4 . The change of ground state has strong anisotropy of the saturation magnetic moments of Fe. This
important consequences for the magnetic properties: before taking effect is borne out both by the XMCD curves in Fig. 3 and the
the spin–orbit perturbation on the crystal-field levels into account, ligand-field calculations, which reveal a large and highly anisotropic
the orbital angular momentum is quenched for an A1g term. In orbital magnetization due to the predominant influence of the
C4v symmetry, on the other hand, the Eg term is an orbitally low-symmetry ligand field over the Fe–metal interaction, and
degenerate doublet with non-zero orbital moment pointing along confirmed by the different XMCD multiplet structure observed at
the principal symmetry direction. The tendency of O2 –Fe(TPA)4 θ = 70◦ relative to 0◦ (see Supplementary Information, Fig. S2).
to magnetize out-of-plane, together with its stronger magnetic Finally, we note that no sign of magnetic coupling was detected
anisotropy compared with Fe(TPA)4 , can be traced back to its between the Fe centres. Future investigations may address this issue
unperturbed crystal-field configuration and non-zero first-order using different kinds of molecular ligand to adjust the Fe–Fe spacing
orbital moment perpendicular to the surface plane. In Fe(TPA)4 , and spin-dependent electron correlation.
the orbital moment arises as a second-order perturbation effect The spin behaviour of individual atoms in purely metallic and
as spin–orbit coupling admixes the nearly degenerate A1g and molecular environments has been intensively studied in recent
Eg states28 . XMCD sum-rule measurements reflect the difference years, leading to the discovery of unusual magnetic1–4,12,17 and
between the two compounds, showing that O2 –Fe(TPA)4 has electron transport properties29,30 . The capability to fabricate 2D
30% higher orbital magnetic moment compared with Fe(TPA)4 , arrays of monodisperse spin centres with nanometre spacing,
namely 0.55 ± 0.07 and 0.42 ± 0.06µB at T = 8 K, when B = 6 T and to understand and control their magnetic properties at the
is applied out-of-plane and in-plane, that is, along the respective interface with a metal substrate, constitutes a basic step towards
easy magnetization direction. Note that the orbital magnetic the exploitation of single-spin phenomena in heterogeneous devices
moment measured for Fe/Cu(100) is much smaller, 0.18 ± 0.03µB . based on either a mono- or multilayer geometry.
A complete discussion of the sum-rule analysis is reported in Methods
Supplementary Information. Experiments were carried out at the ID08 beamline of the European Synchrotron
The ligand-field calculations yield zero-field splitting of about Radiation Facility in Grenoble. Fe(TPA)4 and O2 –Fe(TPA)4 layers were prepared by
−3 and +5 meV for Fe(TPA)4 and O2 –Fe(TPA)4 , respectively. sequential deposition of TPA and Fe on a clean Cu(100) surface and characterized
However, we caution that these parameters may be overestimated in situ by STM and low-energy electron diffraction before the XAS measurements.
by as much as a factor of two, as are the ligand-field predictions Ultrahigh-vacuum conditions were maintained throughout the structural and XAS
characterization measurements. For further details about the sample preparation,
for the orbital moments, owing to covalency effects that are multiplet calculations and XMCD analysis, see Supplementary Information. The
not treated in the ligand-field approach28 . Our data disclose an DFT calculations were carried out using the generalized gradient approximation
extra unusual effect for surface magnetic systems, namely the as the exchange–correlation functional of the Kohn–Sham equations using the
In the field of molecular spintronics1 , the use of magnetic on surfaces17–20 , evidence that SMMs retain molecular magnetic
molecules for information technology is a main target and the hysteresis when wired to metallic surfaces is still lacking. Without
observation of magnetic hysteresis on individual molecules such evidence, one of the pillars of this emerging field would
organized on surfaces is a necessary step to develop molecular break down.
memory arrays. Although simple paramagnetic molecules can Demonstration that SMM behaviour is not destroyed on
show surface-induced magnetic ordering and hysteresis when interaction with the surface can be achieved by using synchrotron
deposited on ferromagnetic surfaces2 , information storage radiation at very low temperature. In fact, X-ray magnetic circular
at the molecular level requires molecules exhibiting an dichroism21 (XMCD) is a unique tool to provide a detailed
intrinsic remnant magnetization, like the so-called single- magnetic characterization down to sub-monolayers of magnetic
molecule magnets3 (SMMs). These have been intensively atoms22 or molecules2,5 .
investigated for their rich quantum behaviour4 but no magnetic Previous investigations have mainly focused on the archetypal
hysteresis has been so far reported for monolayers of SMMs dodecamanganese family of SMMs (Mn12 ; refs 3,4), but adsorbates
on various non-magnetic substrates, most probably owing of Mn12 complexes on gold undergo systematic reduction, with
to the chemical instability of clusters on surfaces5 . Using concomitant disappearance of magnetic hysteresis5 . Moreover,
X-ray absorption spectroscopy and X-ray magnetic circular even the surface layers of bulk Mn12 samples show no magnetic
dichroism synchrotron-based techniques, pushed to the limits hysteresis, probably owing to structural deformations20 . In contrast,
in sensitivity and operated at sub-kelvin temperatures, we we have found that a different class of SMMs comprising four
have now found that robust, tailor-made Fe4 complexes retain high-spin iron(iii) ions, Fe4 , exhibits the sought redox stability and
magnetic hysteresis at gold surfaces. Our results demonstrate structural robustness. Studies on bulk samples have revealed that
that isolated SMMs can be used for storing information. The Fe4 clusters in the layers closer to the surface–vacuum interface
road is now open to address individual molecules wired to retain both their static and dynamic magnetic features20 and are
a conducting surface6,7 in their blocked magnetization state, therefore good candidates to investigate the existence of magnetic
thereby enabling investigation of the elementary interactions hysteresis in molecules grafted onto conducting surfaces.
between electron transport and magnetism degrees of freedom In the propeller-like molecular structure of Fe4 (see Fig. 1a), the
at the molecular scale8,9 . outer individual s = 5/2 spins interact antiferromagnetically with
The observation of a magnetic memory effect3 in the inner one, to give a ground S = 5 spin state characterized by an
single-molecule magnets (SMMs) has represented a breakthrough anisotropy barrier of about 16 K to be overcome for magnetization
in nanomagnetism10 . It has opened the exciting perspective reversal23–25 . The ligands lying above and below the plane of
of exploiting the information storage capability of individual the iron ions are derivatives of 2-hydroxymethyl-1,3-propanediol,
SMMs, made even more appealing owing to their rich quantum which can be functionalized to promote interactions with different
behaviour4,11 . In fact, the combination of a large spin of the substrates as well as to enhance magnetic anisotropy23,24 . For
molecule with an easy-axis magnetic anisotropy results in an deposition on gold surfaces, we have designed the sulphur-
energy barrier that hampers the reversal of the magnetization, functionalized derivative [Fe4 (L)2 (dpm)6 ] with H3 L representing
although the barrier can be cross-cut by a tunnelling mechanism 11-(acetylthio)-2,2-bis(hydroxymethyl)undecan-1-ol and Hdpm
for some particular values of the magnetic field4 . Anchoring SMMs representing dipivaloylmethane25 . The compound can be isolated
on conducting surfaces is required to make them individually in pure, crystalline form and is completely stable in solution (see
addressable by scanning probes8 , which enable both high-resolution Supplementary Information).
morphological studies and local spectroscopic investigations12 . A monolayer of Fe4 molecules on a Au(111) surface was obtained
Moreover, the proposed scheme provides access to electron by self-assembly from solution. Its scanning tunnelling microscopy
transport through the magnetic molecules, the target of a field (STM) characterization (Fig. 1b) carried out at room temperature
that is referred to as molecular spintronics1,9 . Despite the successful in air confirms the surface coverage by a homogeneous layer of
efforts to organize SMMs on various substrates13 and the reported objects with size 2.5 ± 0.4 nm (see Fig. 1c and Supplementary
possibility to detect magnetic features of atoms14–16 or molecules Information), in good agreement with the molecular size
1 Department of Chemistry and INSTM research unit, University of Florence, 50019, Sesto Fiorentino, Italy, 2 ISTM-CNR, URT Firenze, 50019, Sesto
Fiorentino, Italy, 3 Institut de Minéralogie et de Physique des Milieux Condensés, CNRS UMR7590, Université Pierre et Marie Curie, 75252 Paris Cedex 5,
France, 4 Department of Chemistry and INSTM research unit, University of Modena and Reggio Emilia, 41100, Modena, Italy, 5 Synchrotron Soleil, Saint
Aubin BP48, 91192 Gif sur Yvette Cedex, France. *e-mail: roberta.sessoli@unifi.it.
8.6 nm
c 120
90
Counts
60
30
0
1.0 1.5 2.0 2.5 3.0 3.5 4.0
Diameter (nm)
Figure 1 | Monolayer of Fe4 on gold. a, Schematic diagram of the anchoring on a gold surface of the Fe4 derivative studied in this work through its
thiolate-terminated aliphatic chains. A view along the plane of the four metal atoms with the chains fully stretched has been chosen for the sake of clarity.
Iron atoms are drawn as large green spheres, oxygen in red, carbon in grey and sulphur in light yellow. Inset: A view of the magnetic core structure of Fe4 ,
with the ground-state spin arrangement (white arrows). b, Room-temperature constant-current STM image of the Fe4 monolayer obtained on a scan area
of 55 × 55 nm2 (bias voltage Vb = 350 mV, tunnelling current It = 9 pA). c, Statistical distribution of molecular diameters extracted from five images
obtained in the same experimental conditions as in b. The diameters follow a log-normal distribution centred at 2.50 nm with a standard deviation of
0.14 nm (solid curve).
determined by X-ray diffraction. The gold–sulphur chemical coordination environment of the central and peripheral iron
affinity enables the selective formation of a single-layer-thick ions, an antiferromagnetic interaction is expected to afford an
deposit, but no bidimensional order comparable to that observed approximately halved XMCD/XAS ratio as compared with full
in self-assembled monolayers of aliphatic thiols26 can be detected in polarization induced by the magnetic field (see Supplementary
the STM topographic images. In fact, the shape of the molecule and Information). Fragmentation into simple paramagnetic units
the strong steric hindrance of the equatorial ligands are expected or marked alterations of the core structure would then be
to prevent significant interactions between the aliphatic chains accompanied by significant changes in the XMCD/XAS ratio as
of adjacent molecules, a key factor in the formation of organic compared with the bulk, not observed here.
self-assembled monolayers. As the presence of hysteresis is directly related to the magnetic
The electronic structure of the layer has been investigated anisotropy, we have first investigated the field dependence of
through X-ray absorption spectroscopy (XAS) with an endstation the XMCD signal at 709.2 eV at two different temperatures
capable of reaching sub-kelvin temperatures27,28 . In Fig. 2a, we (T = 1.5 ± 0.1 and 4.5 ± 0.1 K). Interestingly the magnetization
show the XAS spectra recorded at the Fe L2,3 -edge (2p → 3d) isotherms so obtained do not rescale on a unique curve when
on the Fe4 monolayer with left- (σ+ ) and right- (σ− ) hand plotted versus H /T (Fig. 2c), in agreement with what is expected
circular polarization. The average spectra, calculated as (σ− +σ+ )/2, for an anisotropic paramagnet. The field dependence of the
coincide over the entire energy range with those recorded on XMCD signal compares well with the magnetization curve
a thick film of the same compound prepared by drop-casting calculated using the spin Hamiltonian that describes the magnetic
(see Supplementary Information)20 , showing that the electronic behaviour of bulk samples.
structure of the molecule is fully preserved on grafting. However, The encouraging results prompted us to explore the
the signal-to-background intensity ratio is about 30 times smaller field-dependent magnetic response of the sample down to the
in the monolayer as compared with the thick deposit and has been very low temperatures required to observe magnetic memory
found reproducible over five different preparations. As the total- effects in SMM materials. We have therefore exploited the assets
electron-yield mode used to detect X-ray absorption probes only of the X11MA-SIM beamline (beam stability and fast switching
the first few layers in bulk samples29 , the observed intensity fully of X-ray polarization) and further optimized the endstation so
confirms that we are actually monitoring a monolayer deposit. as to achieve the lowest possible level of iron contamination in
Element- and surface-selective magnetometry has then been the thermal shields required to reach sub-kelvin temperatures (see
carried out by recording the dichroic component, that is, the Supplementary Information for details). The recorded hysteresis
difference between the XAS signals obtained using right-hand curves are shown in Fig. 3a–c.
and left-hand circularly polarized X-rays, in an applied field Whereas simple paramagnetic behaviour is found at 1.0 K, a
of 3 T at 0.50 ± 0.05 K. The XMCD features (see Fig. 2b) and gradual opening of the hysteresis loop is detected at sub-kelvin
the XMCD/XAS intensity ratio (0.25–0.30) are identical to those temperatures. At T = 0.50±0.05 K, the loop has the typical butterfly
observed in bulk samples. The XMCD/XAS ratio has to be shape resulting from fast magnetization tunnelling in zero field23,24 ,
regarded as a reliable fingerprint of the spin alignment within showing unambiguously that SMM behaviour is observable on
the tetra iron(iii) core. In fact, owing to the very similar gold-wired molecules. Even though XMCD has recently been
a c
L3
1.50 σ+
L2
XAS (a.u.)
1.45 σ¬
X-ray 0.06
1.40
H
XMCD (a.u.)
1.35
0.04
b
XMCD (a.u.)
0
0.02 Exp. (T = 4.5 K)
Figure 2 | XAS/XMCD of Fe4 monolayer. a, Iron L2,3 -edge XAS spectra recorded on the Fe4 monolayer at T = 0.50 ± 0.05 K using left- (σ+ ) and right-
(σ− ) hand circularly polarized light in a field induction of 3.0 T. b, XMCD spectrum (open circles) obtained as (σ− − σ+ ) and compared to the bulk
reference spectrum (solid line) recorded in similar experimental conditions on a thick film on the same substrate. c, Field and temperature dependence of
the XMCD signal at 709.2 eV (multiplied by −1) recorded on the Fe4 monolayer. The data have been taken by scanning the field up to 5.5 T at two different
temperatures, and the standard deviation (bars) has been evaluated by averaging four data sets. The lines correspond to the calculated magnetization by
using the spin Hamiltonian H = DS2z + gµB µ0 H ·S, where D is the axial anisotropy parameter, g is the Landé factor, µB is the Bohr magneton and H is the
applied magnetic field. Taking into account a random orientation of the molecular z axes with respect to the magnetic field, an acceptable agreement is
obtained assuming D/kB = −0.62 K and g = 2.00, as found in bulk samples25 .
a b
T = 1.0 K T = 0.70 K
0.02 0.02
0.01 0.01
XMCD (a.u.)
XMCD (a.u.)
0 0
¬0.01 ¬0.01
¬0.02 ¬0.02
¬1.5 ¬1.0 ¬0.5 0 0.5 1.0 1.5 ¬1.5 ¬1.0 ¬0.5 0 0.5 1.0 1.5
µ0H (T) µ0H (T)
c d
T = 0.50 K
Experimental data
0.02 ¬0.008 Fit
0.01
XMCD (a.u.)
XMCD (a.u.)
¬0.010
0
¬0.01 ¬0.012
¬0.02
¬0.014 T = 0.50 K
Figure 3 | Hysteresis and magnetization dynamics of Fe4 monolayer. a–c, Temperature dependence of the magnetic hysteresis for the Fe4 monolayer,
monitored through the XMCD intensity (multiplied by −1) at 709.2 eV obtained from XAS signals normalized to the pre-edge intensity (704.0 eV). The
error bars represent the standard deviations estimated by averaging over three field cycles. A sweeping rate of 2 mT s−1 has been used. d, Time dependence
of the dichroic signal (multiplied by -1) at 709.2 eV for the Fe4 monolayer at T = 0.50 ± 0.05 K. The sample was first magnetized in a strong positive
magnetic field (+2.0 T), then the field was rapidly ramped to a moderate negative value (−0.25 T) and the time dependence of the dichroic signal was
measured. The procedure was repeated four times and the grey error bars represent the standard deviation. The black line provides the best-fit to
experimental data using a mono-exponential decay.
used to detect the substrate-driven magnetism of paramagnetic Further evidence that the deposited Fe4 complexes retain slow
(octaethylporphyrinato)iron(ii) complexes deposited on a ferro- magnetic relaxation was gained by measuring the time dependence
magnetic surface2 , this is the first observation of a magnetic hystere- of the XMCD signal after magnetizing the sample in a strong
sis of purely molecular origin in monolayers. magnetic field and then rapidly reducing the field to a small
Generating, manipulating and detecting electron spin Spin polarization of the electrons at Ga2+ i was provided by
polarization and coherence at room temperature is at the heart optical orientation15 using circularly polarized light (σ ± ) via
of future spintronics and spin-based quantum information spin-dependent recombination16–21 , as described in the Methods
technology1–4 . Spin filtering, which is a key issue for spintronic section. The spin-filtering effect induced by the spin-polarized
applications, has been demonstrated by using ferromagnetic defects drives the spins of the conduction and localized elec-
metals5–8 , diluted magnetic semiconductors9,10 , quantum point trons at Ga2+ i towards complete alignment when no further elec-
contacts11 , quantum dots12 , carbon nanotubes13 , multiferroics14 tron capture and recombination can occur via the defects as
and so on. This filtering effect was so far restricted to a shown in Fig. 1b. Consequently, a high degree of conduction
limited efficiency and primarily at low temperatures or under electron spin polarization Pe can be achieved. Pe is defined as
a magnetic field. Here, we provide direct and unambiguous Pe = ((n+ − n− )/(n+ + n− )), where n+ (n− ) denotes the spin-up
experimental proof that an electron-spin-polarized defect, (spin-down) conduction electron concentration. For comparison,
such as a Gai self-interstitial in dilute nitride GaNAs, can linearly polarized optical excitation (σ x ) was also used. Such exci-
effectively deplete conduction electrons with an opposite spin tation equally generates electrons of both spins, and will thus lead
orientation and can thus turn the non-magnetic semiconductor to zero electron spin polarization of the defects at zero magnetic
into an efficient spin filter operating at room temperature field (B = 0). Now, the defect can capture conduction electrons of
and zero magnetic field. This work shows the potential of both spins with an equal probability (Fig. 1c); that is, spin filtering
such defect-engineered, switchable spin filters as an attractive ceases to function. This leads to zero Pe and shorter lifetimes
alternative to generate, amplify and detect electron spin for the photo-generated carriers as compared with that under σ ±
polarization at room temperature without a magnetic material excitation, which can be monitored by photoluminescence of the
or external magnetic fields. band-to-band transition.
In Fig. 1a we illustrate the physical principle of the approach. If In Fig. 2a we show the band-to-band photoluminescence
an electron localized at a defect is spin polarized, the defect will spectrum from GaNAs obtained at room temperature and B = 0.
capture only a conduction electron with an opposite spin owing It consists of the strongly overlapping electron–heavy hole (e–hh)
to the Pauli exclusion principle and can subsequently annihilate it and electron–light hole (e–lh) transitions, which split under the
with a hole. If the corresponding process is efficient, spin-polarized tensile strain leaving the e–hh emission at the shorter wave-
defects will effectively deplete the conduction electrons with the length (for a review on GaNAs, see, for example, ref. 22). The
opposite spin and will leave behind only those with the same photoluminescence polarization of these transitions, defined by
spin as the defect electrons. This process will turn a pool of spin- (I (σ + ) − I (σ − ))/(I (σ + ) + I (σ − )), where I (σ + ) and I (σ − ) denote
unpolarized electrons entering the material into spin-polarized ones photoluminescence intensities with the corresponding polariza-
when they exit, making the material a spin filter. The incoming tions, directly reflects Pe . (Owing to strong spectral overlap with
electrons to be spin filtered as well as the spin polarization of the the more intense e–hh emission with opposite photolumines-
electrons at the defects can be provided either by electrical injection cence polarization, the photoluminescence polarization of the
or by optical excitation, making this versatile spin filter relevant to a e–lh emission is severely undermined in the measured spectra.)
variety of applications in future spin electronics or spin photonics. Pe = −((I (σ + ) − I (σ − ))/(I (σ + ) + I (σ − ))) when only the e–hh
For a proof-of-concept purpose, we used optical excitation transition is considered, which is used below to obtain Pe . It
and detection as illustrated in Fig. 1b,c. We show below that Gai should be pointed out that the actual Pe values should be higher
self-interstitials in GaNAs are identified as suitable defects for than that deduced in this way when hh–lh mixing and spectral
efficient spin filtering. This is because each Gai is occupied by a overlap are taken into account. Under σ x excitation, zero Pe was
single electron in the doubly positively charged state Ga2+ i . When observed as expected. Under σ ± excitation, on the other hand,
this electron is spin polarized, the defect can efficiently capture a a sizable Pe can be detected. Its sign can be changed following
conduction electron with an opposite spin and subsequently rapidly a change in the helicity of the excitation light. Pe was found to
annihilate it with a free hole. The complete carrier recombination critically depend on N composition and growth conditions, which
path is identified as: determine the concentration of Gai . The observed values (∼32%)
photo-excitation e capture h capture
are much higher than that observed in N -free GaAs (<6%), being
Ga2+ 2+
−−−→ Ga1+
i −−−−−−−→ Gai + e + h − i +h −−−−→ Ga2+
i a direct consequence of the strong spin-filtering effect due to
1 Departmentof Physics, Chemistry and Biology, Linköping University, 58183 Linköping, Sweden, 2 Université de Toulouse, LPCNO: INSA, UPS, CNRS, 135
avenue de Rangueil, 31077 Toulouse cedex, France, 3 Department of Electrical and Computer Engineering, University of California, La Jolla, California
92093, USA, 4 LPN, route de Noazay, 91460 Marcoussis, France. *e-mail: irb@ifm.liu.se; wmc@ifm.liu.se.
PL polarization (%)
Exc:σσ + 40 σσ+ σσ+ σσ¬ σσ¬
¬Pe lh
Gai2+ ¬1/2 +1/2
GaNAs hh
h Exc:σσ x ¬3/2 +3/2
0
950 1,000
Wavelength (nm)
c b 1,050
b CB CB Exc:σσ x
(1)
Exc: Gai Exc: Gai
PL PL
Wavelength (nm)
σσ+ (1+/2+) σσx (1+/2+) 950
(2) 50
c 1,050
VB VB
Exc:σσ +
0
Spin filtering No spin filtering
PL (arb. units)
self-interstitials in GaNAs. b, In our approach, spin polarization of the first Exc:σσ +
electrons at the defects is achieved by circularly polarized photo-excitation
(σ + as an example), leading to the spin-filtering effect and spin blockade of Exc:σσ x
carrier recombination via the defects when both conduction and defect B=0
electrons are completely polarized to the same spin orientation. 0 100 200
Consequently, higher free carrier concentrations and thus stronger Time (ps)
intensity of the associated band-to-band photoluminescence (PL)
transition are expected in case b as compared with case c. c, As a reference Figure 2 | Photoluminescence intensity and polarization (corresponding
in our approach, linearly polarized photo-excitation (σ x ) corresponds to the to the conduction electron spin polarization Pe ) with or without the
case when no spin polarization of the first electrons at the defects can be spin-filtering effect. a, A typical band-to-band photoluminescence
created and thus no spin-filtering effect is expected. CB and VB denote the spectrum under σ x excitation, and photoluminescence polarization
conduction band and valence band, respectively. The charge transition obtained under σ x and σ + excitation. A schematic diagram of the
(1+/2+) of the spin-filtering Gai defects is also indicated, corresponding to conduction band and valence band states and the associated e–hh and e–lh
a transition between Ga1+ 2+
i and Gai that have two and one bound electron, transitions is also shown. The notations ±1/2 and ±3/2 refer to |mcs i and
respectively. The numbers in the parentheses associated with the electron |mvJ i of the conduction and valence band electrons. b,c, Temporal evolution
capture and annihilation via the defects shown in c indicate the sequence of of the band-to-band photoluminescence spectra (by detecting the total
events. The thickness of the photoluminescence arrows symbolizes the photoluminescence intensity) after σ x - and σ + -polarized pulsed laser
relative intensity of the band-to-band transition. excitation, respectively. Time-integrated photoluminescence spectra, under
σ x (black curve) and σ + (red curve) excitation, are shown in the inset.
Gai preferably introduced in GaNAs. The observed strong Pe is d, Decay curves of the total intensity of the band-to-band
accompanied by a higher photoluminescence intensity under σ ± photoluminescence under σ x and σ + excitation, normalized to their peak
excitation than that under σ x excitation (Fig. 2b,c). This arises from intensity for easy comparison. All results were obtained from the as-grown
spin blockade of the Gai -mediated carrier recombination under GaN0.0076 As0.9924 epilayer grown at 420 ◦ C. The wavelength of the
σ ± excitation, consistent with the slower photoluminescence decay excitation photons is 840 nm. All results were obtained at room
shown in Fig. 2b–d. temperature and B = 0.
To obtain quantitative information on the concentration of
spin-filtering defects and their effect on Pe , we carried out a Here, G± is the photo-generation rate of free carriers and n±
detailed study of Pe dependence on excitation light intensity (N± ) is the density of conduction electrons (the density of the
(that is, on the number of photo-generated free carriers). defects occupied by a single electron), where the ‘±’ signs refer
Representative results obtained at room temperature and B = 0 to the electron spin orientations Sz = ±1/2. N↑↓ corresponds to
are shown in Fig. 3, and were analysed by the following coupled the concentration of the defects having two spin-paired electrons
nonlinear rate equations19,21 : and Nc is the total defect concentration. The density of free
holes is denoted by p. τs (τsc ) is the spin relaxation times of
dn±
= −γe n± N∓ −
n± − n∓
+ G± −
n± conduction (localized) electrons. τd denotes the free carrier decay
dt 2τs τd time, including all radiative and non-radiative recombination
channels except that via the spin-filtering defects. γe (γh ) is the
dN± N± − N∓ 1 trapping coefficient of free electrons (free holes) by the defects. In
= −γe n∓ N± − + γh pN↑↓
dt 2τsc 2 the analysis, we use τd = 10 ns, τsc = 1.5 ns and τs = 150 ps (refs 19,
(1)
21). Here, we assume that τd is governed by the radiative time
dp n+ + n− of the band-to-band photoluminescence transition. The analysis is
= −γh pN↑↓ + G+ + G− −
dt τd not sensitive to τd as long as it is much longer than the electron
capture and recombination time via the spin-filtering defects, that
Nc = N↑↓ + N+ + N− is, τd (γe N± )−1 and τd (γh N↑↓ )−1 . The analysis is insensitive
¬Pe (%)
to compare relative defect concentrations in different samples. The RTA γeNc = 0.15 ps¬1
10
fitting curves based on equation (1) are shown by the dashed lines
in Fig. 3, with γe /γh = 4 and γe Nc values given for each sample.
The analysis reveals that the defect concentration increases with 5
γeNc = 0.037 ps¬1
decreasing growth temperature and increasing N concentration
(not shown in Fig. 3). Post-growth rapid thermal annealing (RTA) 0
0 50 100 150 200
is found to significantly reduce the defect concentration by a factor
Continuous-wave laser power (mW)
of ∼2–15. The observed increase of Pe with excitation power clearly
manifests the spin-filtering effect. b 20
0
GaNAs MQWs, Tg = 580 °C
To identify the exact chemical nature of the spin-filtering
defects, we carried out a detailed study by optically detected 15
magnetic resonance (ODMR; for a review on ODMR, see, for γ eNc = 0.077 ps¬1
example, ref. 23). ODMR is ideally suited here because it monitors
¬Pe (%)
both the conduction electron concentration (via the intensity of 10
the band-to-band photoluminescence) and the spin polarization
of the electrons at the spin-filtering defects (via the electron 5 γ eNc = 0.031 ps¬1
spin resonance (ESR) intensity that is proportional to |N+ –N− |),
as well as the link between them (via ODMR as a change in
0
the intensity of the band-to-band photoluminescence induced 0 50 100 150 200
by the ESR transitions). Under σ x excitation, an equal number Continuous-wave laser power (mW)
of spin-up and spin-down conduction electrons are created. If c
the capture and annihilation of the electrons by the defects GaNAs epilayer
dominates over spin relaxation of both conduction and localized Tg = 420 ° C
electrons, no population difference between the two spin states
40
of the defects is expected, that is, |N+ –N− | = 0, rendering a
¬Pe (%)
vanishing ESR (thus ODMR) signal. Under σ ± excitation, on
the other hand, the dynamic spin polarization of the electrons γeNc = 0.074 ps¬1
at the defects leads to a sizable |N+ –N− | and thus ESR. The 20
ESR-induced spin flips of the spin-polarized defect electrons will
open up an otherwise spin-forbidden recombination channel (see
Fig. 4a). This will lead to a reduction of free carrier concentrations 0
0 50 100 150
and thus of the corresponding band-to-band photoluminescence Pulsed laser power (mW)
intensity, giving rise to a so-called ODMR signal. As examples,
we show in Fig. 4b typical ODMR spectra from GaNAs. As Figure 3 | Dependence of conduction electron spin polarization on the
expected, a vanishingly weak ODMR signal was observed under optical excitation power and the concentration of the spin-filtering
σ x excitation, whereas rather strong ODMR signals were observed defects. a,b, The symbols denote the Pe values determined from the
under σ ± excitation. These findings clearly show that a sizable photoluminescence polarization of the e–hh transition detected at the peak
spin polarization of the defect electrons was generated under σ ± position of the photoluminescence spectra under σ + continuous-wave
excitation, and a slight change of this polarization by the ESR laser excitation at 840 nm, obtained from the GaN0.012 As0.988 MQWs
directly affects the free carrier concentration, proving the role of the grown at 420 ◦ C (a) and the GaN0.011 As0.989 MQWs grown at 580 ◦ C (b).
defects in spin filtering. Results from both as-grown (filled circles) and RTA-treated (filled
A Gai self-interstitial is unambiguously identified by ODMR triangles) samples are shown. c, The time-integrated values of Pe from the
as the core of the spin-filtering defects. This conclusion is based GaN0.0076 As0.9924 epilayer grown at 420 ◦ C (circles), obtained by
on the following experimental facts. First, the observed multiple detecting the photoluminescence polarization at 950 nm (see Fig. 2a)
ODMR lines arise from a hyperfine structure derived from a strong under σ + pulsed laser excitation at 840 nm. The dashed lines are the
interaction between an unpaired localized electron spin (S = 1/2) simulated curves based on the best fit of equation (1) to the experimental
and the nuclear spin of an atom that has two isotopes with a nuclear results, yielding the fitting parameter γe Nc for each sample. The results
spin I = 3/2 and a 60/40 ratio of natural abundance. This gives rise clearly show that Pe increases with increasing excitation intensity until the
to four magnetic-dipole-allowed ODMR transitions (1mS = ±1 concentration of the photo-generated conduction electrons is comparable
and 1mI = 0) for each Ga isotope with an intensity ratio of 60/40 to that of the spin-filtering defects. With a further increase in the excitation
(see Fig. 4c). Ga is the only atom with such unique properties, that density, Pe saturates, and even slightly decreases under the
is, two naturally abundant isotopes, 69 Ga (60.4% abundant) and continuous-wave laser excitation, because the minority spins of the excess
71
Ga (39.6% abundant), and I = 3/2 for both isotopes. Second, conduction electrons can no longer be depleted by the defects. The
the ODMR spectra are isotropic with a rotation of magnetic field maximum value of Pe achievable in each sample is determined by its γe Nc
B with respect to the crystallographic axes. This finding reveals value and thus its defect concentration. Typically, higher Pe values were
that the electron wavefunction at the defects should be s-like. This obtained under the pulsed laser excitation with a similar average laser
is consistent with the observed strong hyperfine interaction, as power as compared to that under the continuous-wave laser excitation.
the s-like electron wavefunction results in a strong Fermi contact This can be attributed to the much higher excitation density during the
term. The involved defects should then be in the Ga2+ i charge state laser pulses and a discontinued supply of minority spins after the laser
occupied by a single unpaired electron with an electronic state pulses, when the pulsed laser excitation was used. All results were
of A1 symmetry, as the alternative GaAs antisite was predicted to obtained at room temperature and B = 0.
N¬ N¬ Exc: σσ x
B B
ODMR (arb.units)
Without ESR & ODMR With ESR & ODMR
Exp
Exc: σσ +
(2)
PL
Exc:
σσ+ Sim
(3)
Gai¬C
(1)
Magnetic field (mT)
c
ms mI
69Ga ¬C
+3/2
10 i +1/2 +1/2
¬1/2
¬3/2
d 0
¬3/2
¬10 ¬1/2 ¬1/2
Energy (GHz)
+1/2
ODMR (arb.units)
+3/2
+3/2
71Ga ¬C +1/2
10 i ¬1/2
+1/2 ¬3/2
0 ¬3/2
¬10 ¬1/2 ¬1/2
+1/2
+3/2
0 0.2 0.4 0.6 100 300 500
γ eNc (ps¬1) Magnetic field (mT)
Figure 4 | Identification of the spin-filtering defects by ODMR. a, Schematic diagrams of the expected ESR transitions under σ x and σ + excitation. In the
former case, an equal population of the two electron spin states of the Ga2+i defects results in vanishing ESR and ODMR signals as the ESR and ODMR
intensity is proportional to the population difference |N+ –N− |. Under σ + excitation, on the contrary, a sizable |N+ –N− | induced by dynamical electron spin
polarization leads to a detectable ESR transition between the two spin states. The ESR transition opens the otherwise spin-forbidden carrier capture and
recombination channel, leading to a reduced number of free carriers available for the band-to-band photoluminescence transition—an ODMR signal. The
numbers in the parentheses indicate the sequence of events. b, Typical ODMR spectra obtained by monitoring the total intensity of the band-to-band
photoluminescence from an RTA-treated GaN0.021 As0.979 epilayer, obtained at 3 K under σ x and σ + excitation at 850 nm. The microwave frequency used
is 9.2823 GHz. A simulated ODMR spectrum of the identified Gai defect (denoted by Gai –C) is also shown. c, Calculated energy levels associated with the
electronic and nuclear spin states of the Ga2+i defect. The allowed ESR transitions (1mS = ±1 and 1mI = 0) occur when the electron spin splitting matches
the microwave photon energy, and are marked by the vertical lines. The calculations and simulation were done with the aid of equation (2) by using the
spin Hamiltonian parameters given in the text. d, ODMR intensities from the identified Gai defects as a function of γe Nc . The latter are determined from the
rate equation analyses (equation (1)). The line is a guide to the eye.
possess a T2 -symmetry state24,25 . To quantitatively confirm this for a free neutral Ga atom25,27 , the localization of the electron
identification, we have carried out a detailed analysis of the ODMR wavefunctions at the Ga2+ i defects is estimated to be 20%
results by a spin Hamiltonian (as-grown) and 16% (RTA). The rather strong localization
shows that they are deep-level defects, a prerequisite for efficient
H = µB g B ·S + AS ·I (2) carrier recombination.
To further quantitatively confirm that the identified Gai
Here, µB is the Bohr magneton, g is the electronic g -factor defects are indeed responsible for the spin filtering, we examined
and A is the hyperfine parameter. g = 2.000 (2.005), their ODMR intensity as a function of the spin-filtering defect
A(69 Ga) = 620 (770) × 10−4 cm−1 and A(71 Ga) = 788 concentrations deduced by equation (1). The excellent correlation,
(1,000)×10−4 cm−1 are obtained from a best fit to the experimental shown in Fig. 4d, provides further support for the predominant role
data for the Gai defects in the RTA-treated (as-grown) samples. of these Gai defects in spin filtering that has led to the observed
The ratio A(71 Ga)/A(69 Ga) is in excellent agreement with that strong Pe at room temperature.
of their nuclear magnetic moments µ(71 Ga)/µ(69 Ga) = 1.27. An Further improvements in the efficiency of the defect-engineered
ODMR spectrum simulated by equation (2) using the obtained spin-filtering can be predicted by (1) increasing the ratio between
parameters is shown as Gai –C in Fig. 4b, showing excellent the capture rate of conduction electrons by the Gai defects
agreement with the experimental data. The difference between (∝γe Nc ) and the spin relaxation rate of conduction electrons
the Gai defects in the RTA and as-grown samples lies at their (1/τs ) and (2) increasing the ratio between γe Nc and the spin-
neighbouring atoms and geometric locations. By using a one- independent carrier recombination rate (1/τd ) via other defects.
electron linear combination of atomic orbital scheme26 and the Indeed, the spin relaxation of conduction electrons tends to
charge density of the 4s electron |ψ4s (0)|2 = 72.7 × 1024 cm−3 drive Pe towards zero value at thermal equilibrium (B = 0).
Spin-independent carrier recombination via other defects equally Received 8 September 2008; accepted 13 January 2009;
depletes conduction electrons of both spins, competing with spin published online 15 February 2009
filtering. The maximum Pe value is expected to approach 100%
when γe Nc 1/τs and 1/τd , independent of initial spin polarization References
1. Wolf, S. A. et al. Spintronics: A spin-based electronics vision for the future.
of the incoming electrons before being spin filtered. Future efforts Science 294, 1488–1495 (2001).
should therefore focus on prolonging the spin relaxation time 2. Žutić, I., Fabian, J. & Das Sarma, S. Spintronics: Fundamentals and applications.
of conduction electrons28,29 and increasing the concentration of Rev. Mod. Phys. 76, 323–410 (2004).
the spin-filtering defects without introducing other competing, 3. Awschalom, D. D. & Flatté, M. E. Challenges for semiconductor spintronics.
Nature Phys. 3, 153–159 (2007).
spin-independent carrier recombination centres. 4. Chappert, C., Fert, A. & Nguyen Van Dau, F. The emergence of spin electronics
So far there is a lack of an efficient room-temperature spin in data storage. Nature Mater. 6, 813–823 (2007).
filter—a key element for spintronics—despite intense efforts. This 5. Zhu, H. J. et al. Room-temperature spin injection from Fe into GaAs.
work demonstrates a significant development of our ability in Phys. Rev. Lett. 87, 016601 (2001).
drastically increasing spin-filtering efficiency at room temperature 6. Hammar, P. R. & Johnson, M. Detection of spin-polarized electrons injected
into a two-dimensional electron gas. Phys. Rev. Lett. 88, 066806 (2002).
by purposely introducing spin-filtering defects. Our approach offers 7. Jiang, X. et al. Highly spin-polarized room-temperature tunnel injector for
another attractive feature—the freedom and ease in switching semiconductor spintronics using MgO(100). Phys. Rev. Lett. 94, 056601 (2005).
the spin direction of the spin filters by reversing electron 8. Jonker, B. T. et al. Electrical spin-injection into silicon from a ferromagnetic
spins of spin-filtering defects either optically or electrically. The metal/tunnel barrier contact. Nature Phys. 3, 542–546 (2007).
proposed spin filtering not only can generate Pe or amplify 9. Fiederling, R. et al. Injection and detection of a spin-polarized current in a
light-emitting diode. Nature 402, 787–790 (1999).
weak Pe , but can also detect spin polarization of injected 10. Ohno, Y. et al. Electrical spin injection in a ferromagnetic semiconductor
electrons by monitoring the ratio between the numbers of heterostructure. Nature 402, 790–792 (1999).
electrons passing and being blocked by the spin filter. This 11. Potok, R. M. et al. Detecting spin-polarized currents in ballistic nanostructures.
work has thus demonstrated the potential of defect-engineered, Phys. Rev. Lett. 89, 266602 (2002).
switchable spin filters for creating, amplifying and detecting 12. Folk, J. A. et al. A gate-controlled bidirectional spin filter using quantum
coherence. Science 299, 679–682 (2003).
electron spin polarization at room temperature in a non-magnetic 13. Hauptmann, J. R. et al. Electric-field-controlled spin reversal in a quantum dot
semiconductor without requiring external magnetic fields, desirable with ferromagnetic contacts. Nature Phys. 4, 373–376 (2008).
for practical device applications. 14. Gajek, M. et al. Tunnel junctions with multiferroic barriers. Nature Mater. 6,
296–302 (2007).
Methods 15. Meier, F. & Zakharchenya, B. P. Optical Orientation (North-Holland, 1984).
Samples. The dilute nitride GaNAs samples were grown by gas-source as well 16. Weisbuch, C. & Lampel, G. Spin-dependent recombination and optical spin
as solid-source molecular beam epitaxy on (100)-oriented semi-insulating orientation in semiconductors. Solid. State. Commun. 14, 141 (1974).
GaAs substrates with a 2,500-Å-thick GaAs buffer. Two types of sample were 17. Miller, R. C., Tsang, W. T. & Nordland, W. A. Spin-dependent recombination
studied: (1) 7-period GaAs/GaNAs (200/70 Å) multiple-quantum-wells (MQWs) in GaAs. Phys. Rev. B 21, 1569–1575 (1980).
structures, which were grown either at Tg = 420 ◦ C with N composition [N ] = 1.2% 18. Paget, D. Optical-pumping study of spin-dependent recombination in GaAs.
and 2% or at 580 ◦ C with [N ] = 1.1%; and (2) GaNAs epilayers (about 0.1 µm Phys. Rev. B 30, 931–946 (1984).
thickness) with [N ] = 0.76–2.1%, grown at 420 ◦ C. Post-growth RTA was carried 19. Kalevich, V. K. et al. Spin-dependent recombination in GaAsN solid solutions.
out at 850 ◦ C for 10 s with a halogen lamp in a flowing N2 ambient. JETP Lett. 82, 455–458 (2005).
20. Lombez, L. et al. Spin dynamics in dilute nitride semiconductors at room
Dynamical polarization of the spin-filtering Gai defects. In our experiments, temperature. Appl. Phys. Lett. 87, 252115 (2005).
the initial spin polarization of conduction electrons was generated by 21. Lagarde, D. et al. Electron spin dynamics in GaAsN and InGaAsN structures.
circularly polarized light excitation at energy above the e–hh and e–lh Phys. Status Solidi A 204, 208–220 (2007).
band-to-band transition in GaNAs, taking advantage of the selection rules 22. Buyanova, I. A. & Chen, W. M. Physics and Applications of Dilute Nitrides
and the threefold higher oscillator strength of the e–hh optical transition as (Taylor & Francis Books, 2004).
compared with the e–lh transition15 . A spin-polarized conduction electron 23. Chen, W. M. Applications of optically detected magnetic resonance in
can be captured by the Ga2+ defect only if the first localized electron has a semiconductor layered structures. Thin Solid Films 364, 45–52 (2000).
i
spin orientation opposite to that of the conduction electron, owing to the 24. Baraff, G. A. & Schluter, M. Electronic structure, total energies, and abundances
Pauli exclusion principle. After the subsequent recombination between one of the elementary point defects in GaAs. Phys. Rev. Lett. 55, 1327–1330 (1985).
of two localized electrons and an unpolarized free hole (due to much faster 25. Thinh, N. Q. et al. Properties of Ga-interstitial defects in AlGaNP. Phys. Rev. B
spin relaxation), a half number of the Ga2+ are left with a localized electron 71, 125209 (2005).
i
with its spin orientation parallel to that of the conduction electrons. Such 26. Watkins, G. D. & Corbett, J. W. Defects in irradiated silicon. I. Electron spin
a continuous spin-dependent recombination process16–21 will dynamically resonance of the Si–A center. Phys. Rev. 121, 1001–1014 (1961).
polarize the spin of the first localized electrons at the Ga2+ towards that of 27. Koh, A. K. & Miller, D. J. Hyperfine coupling constants and atomic parameters
i
conduction electrons. for electron paramagnetic resonance data. Atom. Data Nucl. Data Tables 33,
235–253 (1985).
Experimental techniques. Both continuous-wave and time-resolved 28. Ohno, Y. et al. Spin relaxation in GaAs(110) quantum wells. Phys. Rev. Lett. 83,
photoluminescence experiments were carried out at room temperature and 4196–4199 (1999).
at zero magnetic field. Photo-excitation above the bandgap energy of GaNAs 29. D’yakonov, M. I. & Kachorovskii, V. Yu. Spin relaxation of two-dimensional
was provided by circularly polarized light (and linearly polarized light) from a electrons in noncentrosymmetric semiconductors. Sov. Phys. Semicond. 20,
Ti:sapphire laser at a wavelength of 840–855 nm, propagating along the direction 110–112 (1986).
normal to the sample surface. Laser power of up to 200 mW was used with a
typical spot size of about 0.1 mm. The resulting polarized photoluminescence Acknowledgements
was measured in a backscattering geometry. In time-resolved photoluminescence W.M.C. and I.A.B. gratefully acknowledge the support from Linköping University
experiments, the samples were excited by 1.5 ps pulses generated by a mode-locked through the Professor Contracts, the Swedish Research council (VR), the Swedish
Ti:sapphire laser with a repetition frequency of 80 MHz. The time-resolved Energy Agency, the Knut and Alice Wallenberg Foundation, the Wenner-Gren
photoluminescence was then recorded using a streak camera with an overall Foundations and the Swedish Foundation for International Cooperation in Research
time resolution of 8 ps. and Higher Education (STINT). The work at UCSD is partially supported by NSF
The ODMR experiments were done at 2.5–300 K with a modified ESR Grant No. DMR- 0606389.
spectrometer working at the X-band (∼9.3 GHz). Photoluminescence was
excited by a Ti:sapphire laser at a wavelength of 850 nm. The ODMR signals Additional information
were detected as spin-resonance-induced changes of the photoluminescence Reprints and permissions information is available online at http://npg.nature.com/
intensity monitored by a cooled Ge detector with a proper selection reprintsandpermissions. Correspondence and requests for materials should be
of optical filters. addressed to I.A.B. or W.M.C
Graphene, a single monolayer of graphite, has recently for technological applications5–7 . A considerable advantage of this
attracted considerable interest owing to its novel magneto- method is that insulating SiC substrates can be used so that transfer
transport properties1–3 , high carrier mobility and ballistic to another insulator is not required. However, the large-scale
transport up to room temperature4 . It has the potential for structural quality is limited at present by the lack of continuity
technological applications as a successor of silicon in the and uniformity of the grown film15,16 . On the Si-terminated (0001)
post Moore’s law era5–7 , as a single-molecule gas sensor8 , basal plane, vacuum annealing leads to small graphene domains
in spintronics9–11 , in quantum computing12 or as a terahertz typically 30–100 nm in diameter, whereas on the C-terminated
oscillator13 . For such applications, uniform ordered growth of (0001̄) face, larger domains (∼200 nm) of multilayered, rotationally
graphene on an insulating substrate is necessary. The growth of disordered graphene have been produced14 . The small-grain
graphene on insulating silicon carbide (SiC) surfaces by high- structure is due to morphological changes of the surface in the
temperature annealing in vacuum was previously proposed course of high-temperature annealing. Moreover, decomposition
to open a route for large-scale production of graphene-based of SiC is not a self-limiting process and, as a result, regions of
devices5,6 . However, vacuum decomposition of SiC yields different film thicknesses coexist, as shown by low-energy electron
graphene layers with small grains (30–200 nm; refs 14–16). microscopy15,16 (LEEM). Such inhomogeneous films do not meet
Here, we show that the ex situ graphitization of Si-terminated the demands of large-scale device production, which requires
SiC(0001) in an argon atmosphere of about 1 bar produces larger domains and tighter thickness control. Homogeneous film
monolayer graphene films with much larger domain sizes thickness is particularly important because the electronic structure
than previously attainable. Raman spectroscopy and Hall of the film depends strongly on the number of layers. For
measurements confirm the improved quality of the films thus example, although monolayer graphene is a gapless semiconductor,
obtained. High electronic mobilities were found, which reach a forbidden gap can be induced in bilayer graphene and tuned by
µ = 2,000 cm 2 V−1 s−1 at T = 27 K. The new growth process an external electrostatic potential21,22 .
introduced here establishes a method for the synthesis of Here we demonstrate a method of preparing graphene on
graphene films on a technologically viable basis. SiC(0001) that results in a significantly improved film quality. In
The successful development of graphene-based electronic Fig. 1, we compare samples prepared by vacuum annealing with
devices depends on a large-scale availability of the material. samples produced by ex situ annealing under an argon atmosphere.
Several methods for graphene production have been proposed. Figure 1a–c shows the morphology of the 6H–SiC(0001) surface
Mechanical exfoliation17 leads to isolated, high-quality crystals before and after the formation of a graphene monolayer by
with dimensions only in the 10 µm range, which questions annealing in ultrahigh vacuum (UHV) as determined by atomic
the practicality of this method. Large, high-quality graphene force microscopy (AFM) and LEEM. The initial 6H–SiC(0001)
islands were grown on the surface of transition metals such surface in Fig. 1a, obtained after hydrogen etching, is characterized
as Ru(0001) (ref. 18), but this technique requires transfer to by wide, highly uniform, atomically flat terraces. The step direction
an insulating substrate, with methods that have yet to be and terrace width (of the order of 300–700 nm) are determined by
developed. Liquid-phase exfoliation19 yields high quantities of the incidental misorientation of the substrate surface with respect
monolayer and few-layer graphene. Although this method—in to the crystallographic (0001) plane. The step height is 1.5 nm,
contrast to graphene oxide reduction20 —prevents the formation which corresponds to the dimension of the 6H–SiC unit cell in the
of defects, the electrical properties are similar to graphene oxide direction perpendicular to the surface (c axis). On defect-free areas
reduction, indicating poor transport at contacts between the of the sample, the terraces typically extend undisturbed over 50 µm
individual graphene sheets. in length. The morphology of the surface covered with a monolayer
The preparation of single-layer graphene by the thermal of graphene prepared by vacuum annealing is shown in Fig. 1b.
decomposition of silicon carbide (SiC) has been proposed as a The surface obviously undergoes significant modifications; it is now
viable route for the synthesis of uniform, wafer-size graphene layers covered with small pits up to 10 nm in depth, and the original
1 Lehrstuhlfür Technische Physik, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erwin-Rommel-Str. 1, 91058 Erlangen, Germany, 2 Advanced Light
Source, Lawrence Berkeley National Laboratory, One Cyclotron Road, Berkeley, California 94720, USA, 3 Department of Molecular Physics,
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Faradayweg 4-6, 14195 Berlin, Germany, 4 Lehrstuhl für Angewandte Physik,
Friedrich-Alexander-Universität Erlangen-Nürnberg, Staudtstr. 7, 91058 Erlangen, Germany, 5 Sandia National Laboratories, Surface & Interface Sciences
Department, PO Box 5800, Albuquerque, New Mexico 87185, USA, 6 National Center for Electron Microscopy, Lawrence Berkeley National Laboratory,
One Cyclotron Road, Berkeley, California 94720, USA. *e-mail: thomas.seyller@physik.uni-erlangen.de.
a b c
720 nm 1.0 µm
1 µm E = 2.9 eV
0 ML 1 ML 2 ML
d e
8.0 µm
10 µm
E = 5.2 eV
f g i A A
1 1
2 2
Vertical position
2.0 µm 2.0 µm
B B
1 1 A B
1 2 1 32 1 2 1
2 2 40
3 3
Height (nm)
30
4 µm 1 1
E = 5.2 eV
20
0 5 10 15
h Energy (eV)
0 10
Energy (eV)
5 0
0 2 4 6 8 10
10 Distance (µm)
15
Horizontal position
Figure 1 | Morphological changes of 6H–SiC(0001) during graphene growth. a, Initial surface after H-etching imaged by AFM. The step height is 15 Å.
b, AFM image of graphene on 6H–SiC(0001) with a nominal thickness of 1 ML formed by annealing in UHV at a temperature of about 1,280 ◦ C. c, LEEM
image of a UHV-grown graphene film on SiC(0001) with a nominal thickness of 1.2 monolayers. The image contrast is due to the locally different layer
thickness. Light, medium and dark grey correspond to a local thickness of 0, 1 and 2 ML, respectively. d, AFM image of graphene on 6H–SiC(0001) with a
nominal thickness of 1.2 ML formed by annealing in Ar (p = 900 mbar, T = 1,650 ◦ C). e, LEEM image of a sample equivalent to that of d revealing
macro-terraces covered with graphene up to 50 µm long and at least 1 µm wide. f, Close-up LEEM image revealing monolayer coverage on the terraces and
bilayer/trilayer growth at the step edges. g,h, Electron reflectivity spectra (grey-scale images) taken at the positions indicated by the blue lines in f.
Monolayer, bilayer and trilayer graphene are readily identified by the presence of 1, 2 or 3 reflectivity minima, respectively. i, Close-up AFM images of the
film shown in d. In the right-hand-side image, the z scale was adjusted such that the terraces appear at the same height. The profile shows that small
depressions 4 and 8 Å in height exist at the step edges due to second and third layer nucleation.
steps are hardly discernible any longer. This indicates that graphene image shown in Fig. 1c. The irregularly shaped graphene islands
growth is accompanied by substantial changes in the morphology are at most a few hundred nanometres in size, in agreement with
of the substrate itself, leading to a considerable roughening. As a X-ray diffraction14 . Moreover, monolayer graphene areas coexist
consequence of this roughening, the graphene layer acquires an with graphene
√ √ bilayer islands as well as with uncovered regions of
inhomogeneous thickness distribution as can be seen in the LEEM the (6 3×6 3) buffer layer23 .
SiC
S2
S1
c d
G-peak
(1,592 ± 5) cm¬1 2D-peak
0 (2,706 ± 5) cm¬1
D-peak
(1,356 ± 5) cm¬1
Binding energy (eV)
54 cm¬1
ED
0.5
37 cm¬1
1.0
(2,717 ± 5) cm¬1
(1,596 ± 5) cm¬1
1.5
1.5 1.6 1.7 1.8 1.9 1,400 1,600 1,800 2,600 2,800
Electron momentum Kll (Ŭ1) Raman shift (cm¬1)
Figure 2 | Atomic and electronic structure of ex-situ-grown monolayer graphene. a, LEED pattern at 74 eV showing the diffraction spots due
√ √
to the SiC(0001) substrate (blue arrows) and the graphene lattice (red arrows). The extra spots are due to the (6 3 × 6 3) interface layer. b, C1s
√ √
core-level spectrum measured at a photon energy of 700 eV. The spectrum contains contributions from the SiC substrate (marked SiC), the (6 3 × 6 3)
interface layer (marked S1 and S2) and from the graphene layer (G) residing on top of the interface layer. c, π-bands probed by ARPES in the vicinity of the
K-point of the hexagonal Brillouin zone measured along the 0 K-direction. The position of the Dirac energy (ED ) at 0.45 eV below the Fermi energy is
consistent with previous reports on UHV-grown graphene on SiC(0001). Faint features marked by yellow arrows signal the presence of small regions of
bilayer graphene in agreement with the LEEM results. d, Comparison of Raman spectra of Ar-grown (red) and UHV-grown (blue) epitaxial graphene on
6H–SiC(0001). The spectra of the D- and G-line shown here are corrected for the emission of the substrate by subtraction of a reference spectrum26
(see Supplementary Information).
In stark contrast to the low quality resulting from vacuum energy shows stripes that follow in width and orientation the
graphitization (Fig. 1b), films grown under 900 mbar of argon have macro-terraces with a contrast that is determined by the graphene
a greatly improved surface morphology, as demonstrated by the layer thickness15,16 . Hence, we can unambiguously conclude that
AFM image in Fig. 1d. Large continuous terraces are found, extend- except for narrow stripes at the edges, the large atomically
ing over large distances parallel to the step edges. Step bunching is flat macro-terraces are homogeneously covered with a graphene
manifested by the formation of macro-terraces that are a factor of monolayer. The domain size of monolayer graphene is significantly
5–8 times wider than the original terraces. The macro-steps are par- larger than that of the vacuum-annealed samples as a comparison
allel to the original steps, increase in step height by the same factor between Fig. 1c,f shows. In fact, the domain size seems to be
and reach average heights of 8–15 nm. Parallel to the steps, uninter- limited only by the length and width of the SiC terraces. Narrower,
rupted macro-terraces more than 50 µm long have been observed. darker regions at the downward edges of the terraces correspond
The thickness distribution of the graphene film grown under an to bilayer and in some cases trilayer graphene (see region 3 in
argon atmosphere is determined by LEEM as shown in Fig. 1e,f. Fig. 1f). In the AFM image, these regions (see Fig. 1i) appear
A series of spatially resolved LEEM I –V spectra taken along a as small depressions of around 0.5 nm and 1 nm amplitude
vertical and a horizontal line in Fig. 1f is shown in Fig. 1g,h. The located at the very edge of the macro-step. This indicates that
layer thickness is easily determined from the number of minima the nucleation of new graphene layers starts at step edges of
in the individual spectra; the LEEM image taken at a particular the substrate surface.
So far, one of the fundamental limitations of organic satisfied by Lu3 N@C80 ’s LUMO energy7 . Cyclic voltammetry and
photovoltaic (OPV) device power conversion efficiencies Osteryoung square-wave voltammetry (OSWV), shown in Fig. 1
(PCEs) has been the low voltage output caused by a molecular and Table 1, confirm that the reduction potential and therefore
orbital mismatch between the donor polymer and acceptor the LUMO levels of the Lu3 N@C80 and its methano derivatives
molecules. Here, we present a means of addressing the are 204 and 280 mV, respectively, more negative than that of
low voltage output by introducing novel trimetallic nitride [6,6]-phenyl-C61 -butyric methyl ester (C60 -PCBM), the most
endohedral fullerenes (TNEFs) as acceptor materials for use common fullerene acceptor used in OPV devices so far. The open
in photovoltaic devices. TNEFs were discovered in 1999 by circuit voltage in OPVs is correlated to the difference between
Stevenson et al.1 ; for the first time derivatives of the TNEF the donor’s highest occupied molecular orbital (HOMO) and the
acceptor, Lu3 N@C80 , are synthesized and integrated into OPV acceptor’s LUMO (refs 7–10). Therefore, the increase in reduction
devices. The reduced energy offset of the molecular orbitals potential is representative of the expected Voc gain from Lu3 N@C80 -
of Lu3 N@C80 to the donor, poly(3-hexyl)thiophene (P3HT), based OPV devices9,11 .
reduces energy losses in the charge transfer process and Similar to empty-cage fullerenes, the solubility of pristine
increases the open circuit voltage (Voc ) to 260 mV above Lu3 N@C80 is not high enough for solution processing of
reference devices made with [6,6]-phenyl-C61 -butyric methyl OPVs. Thus, an exohedral functionalization of Lu3 N@C80 was
ester (C60 -PCBM) acceptor. PCEs >4% have been observed required to allow its incorporation into solution-processed bulk-
using P3HT as the donor material. This work clears a path heterojunction OPV devices. Here, we report the first methano
towards higher PCEs in OPV devices by demonstrating that derivatives of Lu3 N@C80 that have been isolated and fully char-
high-yield charge separation can occur with OPV systems that acterized. It is important to emphasize that the chemical reactivity
have a reduced donor/acceptor lowest unoccupied molecular of Lu3 N@C80 differs from empty-cage fullerenes. Owing to these
orbital energy offset. differences in reactivity5,12–14 , the standard protocol used for the
The trimetallic nitride endohedral fullerenes (TNEFs) used in synthesis of C60 -PCBM (ref. 15) had to be significantly modified
this study are Ih -C80 fullerenes incarcerating rare-earth metals to obtain the Lu3 N@C80 -PCBX (X = M,B,H,O) analogues. For
bound in a trimetallic nitride cluster. The spherically symmetric example, the reagents had to be increased to 25 times the amounts
C80 and trimetallic nitride cluster do not independently exist, but used for empty-cage fullerenes, the reaction was run at 120 ◦ C
when combined together form a stable molecule. Owing to the instead of 70 ◦ C and the highest yield was obtained after 25 min
large variety of metals that can be used to make the endohedral rather than 22 h, see Fig. 2 and Supplementary Methods. The
cluster, these TNEF nanomaterials are multifunctional, and have Lu3 N@C80 -PCBX family of derivatives—PCBM (methyl), PCBB
been sought for their unique properties in applications such as (butyl), PCBH (hexyl) and PCBO (octyl)—have proven to have
magnetic resonance imaging contrast agents2 and optoelectronics3 . thermal stability up to 300 ◦ C in a thermal gravimetric analyser, see
Recent theoretical4 and experimental data5 suggest that the lowest Supplementary Fig. S1. The absorption spectra of bare Lu3 N@C80
unoccupied molecular orbital (LUMO) energies for this type of (Fig. 3) resembles closely its PCBX derivatives, which suggests
fullerene are much higher than the LUMO energy of empty-cage that this class of TNEF derivatives have a fulleroid character16 .
species with respect to the Fermi level. The higher LUMO provides Electrochemical characterization of these Lu3 N@C80 -PCBX deriva-
a path towards higher Voc , and therefore, higher efficiencies in tives also showed the kinetically reductive irreversible behaviour
organic photovoltaic (OPV) devices. Here, Lu3 N@C80 was chosen of the pristine Lu3 N@C80 and they retained the same reduction
because its LUMO energy is closer to the commonly used OPV potential advantage compared to C60 -PCBM, but their solubility in
donor polymer poly(3-hexyl)thiophene’s (P3HT) LUMO level than organic solvents and miscibility in P3HT varied significantly. The
are those of other available fullerenes such as C60 (refs 6,7). ability to vary the solubility and miscibility by exchanging the X
Photophysics and OPV device measurements demonstrate that portion of the PCBX functional group on TNEF molecules and not
the minimum LUMO offset needed for exciton dissociation is affect the molecular orbitals can serve as a vital tool for designing
1 Georgetown University, 37th and O st. NW, Washington, District of Columbia 20057, USA, 2 Luna Innovations Incorporated, 521 Bridge Street, Danville,
Virginia 24541, USA, 3 Department of Chemistry and Pharmacy & Interdisciplinary Center for Molecular Materials (ICMM), Friedrich-Alexander-
Universität Erlangen-Nürnberg, Egerlandstr. 3, 91058 Erlangen, Germany, 4 National Renewable Energy Laboratory, 1617 Cole Blvd, Golden, Colorado
80401, USA, 5 Center for Polymer and Organic Solids, University of California, Santa Barbara, California 93106, USA, 6 Department of Chemistry and
Biochemistry (or Center for Polymers and Organic Solids) University of California at Santa Barbara, Santa Barbara, California 93117, USA.
*e-mail: dreesm@lunainnovations.com.
(2) Lu3N
OH H+ O
O O Hexanol O O
O p-toluenesulphonyl hydrazide
CH3CN
O
O
(3) Lu3N NO
NH
O S O
O
CH3
O NaOCH3
(4) Pyridine
Lu3N
o-DCB O
N2 O
Lu3N O
+
0 ¬0.5 ¬1.0 ¬1.5 ¬2.0 ¬2.5 ¬3.0 N⊕ O ∼120 ° C
Lu3N
V versus Fc/Fc+ N
TNEF and P3HT was confirmed for the first time by means of
1 µA subnanosecond photolytic experiments. Specifically, we contrasted
the photophysical behaviour in spin-coated films with that in
solution by exciting either at 387 nm (fullerenes) or at 540 nm
(P3HT) and comparing the resulting photospectra with those of
[C60]PCBM
the corresponding radical cation and radical anion produced in
radiolytic experiments, see Supplementary Fig. S3. Initially the
P3HT singlet excited-state features are discernible in photoexcited
1 µA films of P3HT/Lu3 N@C80 -PCBH or P3HT/C60 -PCBM—see Fig. 4
and Supplementary Fig. S4. However, they are in both cases
ultrashort: they decay in less than 550 fs (>2.0 × 1012 M−1 ) to
form a new photoproduct. There is a decisive difference between
[Lu3N@C80]PCBH the photoproducts of the two acceptors: in the near-infrared
0 ¬0.5 ¬1.0 ¬1.5 ¬2.0 ¬2.5 ¬3.0 region, the photoproducts absorb with maxima at 665, 890
V versus Fc/Fc+ and 1,020 nm for P3HT/C60 -PCBM and 665, 890 and 1,025 nm
for P3HT/Lu3 N@C80 -PCBH. The first two maxima (665 and
Figure 1 | TNEF electrochemical properties to establish LUMO levels. 890 nm) are in excellent agreement with those noted for the
a, Cyclic voltammograms of the reductive behaviour of (1) C60 -PCBM, one-electron oxidized radical cation of P3HT in dichloromethane
(2) Lu3 N@C80 , (3) Lu3 N@C80 -PCBM, (4) Lu3 N@C80 -PCBH in 0.5 M (600 and 855 nm—Supplementary Fig. S3a); the last ones (1,020
n-Bu4 NPF6 /o-DCB (o − DCB : 1,2 − dichlorobenzene) with ferrocene as the or 1,025 nm) correspond to the one-electron reduced C60 -PCBM
internal standard, 100 mV s−1 scan rate. b, OSWV measurements of and Lu3 N@C80 -PCBH radical anions generated in toluene, acetone
C60 -PCBM and Lu3 N@C80 in 0.05 M n-Bu4 NPF6 /o-DCB. and 2-propanol (see Supplementary Fig. S3b). The visible portion
of the transient absorption spectra, on the other hand, is in
the optimal composite morphology for high-performance OPV both cases dominated by strong bleaching of the ground state,
devices. Lu3 N@C80 -PCBH was chosen among the synthesized revealing minima at 500 nm. These fingerprints again corroborate
Lu3 N@C80 -PCBX derivatives for this study because of its similarities the attributes seen during the P3HT oxidation. In other words,
in solubility and miscibility to C60 -PCBM in the processing both Lu3 N@C80 -PCBH and C60 -PCBM give rise to the rapid
of P3HT-based OPV devices. Characterization of space-charge- formation of the radical ion pair states, under excitation of
limited devices further confirmed that Lu3 N@C80 -PCBH possesses either the donor or the acceptor. These radical ion pair states
similar charge-carrier mobility to C60 -PCBM (4.0 × 10−4 for were found to be stable on the timescale of our investigation
Lu3 N@C80 -PCBH versus 1.4 × 10−3 cm2 V−1 s−1 for C60 -PCBM), (up to 3,000 ps). Complementary nanosecond experiments (see
see Supplementary Fig. S2. Supplementary Fig. S5) confirm the remarkable stability of the
Photophysical studies were carried out to understand the radical ion pair states in both films with lifetimes exceeding
excited-state interaction of the Lu3 N@C80 derivatives with the the experimental time window of our apparatus (1.0 ms). These
P3HT donor polymer. Charge transfer between photoexcited transient absorption measurements with P3HT/Lu3 N@C80 -PCBH
0.08
a P3HT.+ 1∗P3HT
0.0016
0.06
0.0008
P3HT.+
Absorbance
0.04
ΔOD (a.u.)
0.02 0
P3HT.+
0
¬0.0008 Lu3N@C80.¬
Lu3N@C80
0.0004
300 400 500 600 700 800 900
516 nm
Wavelength (nm) 0 1,000 nm
ΔOD (a.u.)
Figure 3 | Absorption spectra of Lu3 N@C80 and its methano derivatives.
¬0.0004
Inset: Spectra for all four methano derivatives with their parent molecule
Lu3 N@C80 overlaid and magnified in the absorption onset region.
¬0.0008
60
5
50
J (mA cm¬2) 40
EQE (%)
0
30
20
¬5
10
¬10 0
¬0.5 0 0.5 1.0 400 500 600 700
V (V) Wavelength (nm)
c d 7
0.5 6
5
0.4
Intensity (a.u.)
4
Absorbance
0.3
3
0.2 2
1
0.1
0
0
300 400 500 600 700 800 900 4.0 4.5 5.0 5.5 6.0 6.5 7.0
Wavelength (nm) 2θ (°)
Figure 5 | Blend-film electronic and optical properties. a, J–V curves of P3HT/Lu3 N@C80 -PCBH (triangles) PCE = 4.2%, Voc = 810 mV,
Jsc = 8.64 mA cm−2 and FF = 0.61 and P3HT/C60 -PCBM (squares and dashed lines) PCE = 3.4%, Voc = 630 mV, Jsc = 8.9 mA cm−2 and FF = 0.61 blend
devices. Filled symbols show the dark curves and open symbols show devices under simulated Air Mass 1.5 (100 mW cm−2 ). b, EQE P3HT/Lu3 N@C80 -
PCBH (dashed) and P3HT/C60 -PCBM (solid) blend devices. c, Absorption of P3HT/Lu3 N@C80 -PCBH (solid) and P3HT/C60 -PCBM (dashed) blend films.
d, X-ray diffraction of the P3HT (100) peak normalized to thickness for a 1:1 blend film of P3HT/Lu3 N@C80 -PCBH (solid) and a 1:0.8 reference film of
P3HT/C60 -PCBM (dashed) blend.
device. Photophysical experiments confirm the efficient charge 4. Campanera, J. M., Bo, C., Olmstead, M. M., Balch, A. L. & Poblet, J. M.
transfer between P3HT and the Lu3 N@C80 -PCBH. Optimization of Bonding within the endohedral fullerenes Sc3 N@C78 and Sc3 N@C80 as
the P3HT/Lu3 N@C80 -PCBH active-layer morphology has resulted determined by density functional calculations and reexamination of the crystal
structure of {Sc3 N@C78 }·Co(OEP)·1.5(C6 H6 )·0.3(CHCl3 ). J. Phys. Chem. A
in OPV devices with higher Voc values and a similar photocurrent 106, 12356–12364 (2002).
and fill factor compared to P3HT/C60 -PCBM reference devices. 5. Cardona, C. M., Elliott, B. & Echegoyen, L. Unexpected chemical and
The higher overall PCE of the P3HT/Lu3 N@C80 -PCBH devices electrochemical properties of M3 N@C80 (M = Sc, Y, Er). J. Am. Chem. Soc. 128,
is attributed to a better positioned LUMO level that captures 6480–6485 (2006).
more of the energy associated with each absorbed photon. This 6. Koster, L. J. A., Mihailetchi, V. D. & Blom, P. W. M. Ultimate efficiency
of polymer/fullerene bulk heterojunction solar cells. Appl. Phys. Lett. 88,
work demonstrates that reducing the donor/acceptor LUMO offset
0935111 (2006).
by using TNEF acceptor materials can lead to enhanced OPV 7. Scharber, M. C. et al. Design rules for donors in bulk-heterojunction
performance through the Voc . Using the improved LUMO level solar cells—Towards 10% energy-conversion efficiency. Adv. Mater. 18,
offset of our Lu3 N@C80 -PCBH acceptor materials and combining 789–794 (2006).
it with previously reported low-bandgap donor polymers21,22 , OPV 8. Brabec, C. J. et al. Origin of the open circuit voltage of plastic solar cells.
efficiencies greater than 10% may now be feasible6,7 . With this Adv. Funct. Mater. 11, 374–380 (2001).
9. Kooistra, F. B. et al. Increasing the open circuit voltage of bulk-heterojunction
we have introduced an entirely new class of acceptor materials solar cells by raising the LUMO level of the acceptor. Org. Lett. 9,
for organic solar cells that show a straightforward path towards 551–554 (2007).
commercially viable devices. 10. Brabec, C. J. et al. The influence of materials work function on the open circuit
voltage of plastic solar cells. Thin Solid Films 403/404, 368–372 (2002).
Received 11 July 2008; accepted 6 January 2009; 11. Lenes, M. et al. Fullerene bisadducts for enhanced open-circuit voltages and
published online 8 February 2009 efficiencies in polymer solar cells. Adv. Mater. 20, 2116–2119 (2008).
12. Campanera, J. M., Bo, C. & Poblet, J. M. Exohedral reactivity of trimetallic
References nitride template (TNT) endohedral metallofullerenes. J. Org. Chem. 71,
1. Stevenson, S. et al. Small-bandgap endohedral metallofullerenes in high yield 46–54 (2006).
and purity. Nature 401, 55–57 (1999). 13. Cardona, C. M., Kitaygorodskiy, A. & Echegoyen, L. Trimetallic nitride
2. MacFarland, D. K. et al. Hydrochalarones: A novel endohedral metallofullerene endohedral metallofullerenes: Reactivity dictated by the encapsulated metal
platform for enhancing magnetic resonance imaging contrast. J. Med. Chem. cluster. J. Am. Chem. Soc. 127, 10448–10453 (2005).
51, 3681–3863 (2008). 14. Rodriguez-Fortea, A., Campanera, J. M., Cardona, C. M., Echegoyen, L.
3. Jones, M. A. G., Taylor, R. A., Ardavan, A., Porfyrakis, K. & Briggs, G. A. D. & Poblet, J. M. Dancing on a fullerene surface: Isomerization of
Direct optical excitation of a fullerene-incarcerated metal ion. Chem. Phys. Lett. Y3 N@(N-ethylpyrrolidino-C80 ) from the 6,6 to the 5,6 regioisomer.
428, 303–306 (2006). Angew. Chem. Int. Ed. 45, 8176–8180 (2006).
15. Hummelen, J. C. et al. Preparation and characterization of fulleroid and 22. Peet, J. et al. Efficiency enhancement in low-bandgap polymer solar cells by
methanofullerene derivatives. J. Org. Chem. 60, 532–538 (1995). processing with alkane dithiols. Nature Mater. 6, 497–500 (2007).
16. Lukoyanova, O. et al. ‘Open rather than closed’ malonate methano-fullerene
derivatives. The formation of methanofulleroid adducts of Y3 N@C80 . J. Am. Acknowledgements
Chem. Soc. 129, 10423–10430 (2007). This material is based on work supported by the Air Force Office of Scientific Research
17. Li, G. et al. ‘Solvent annealing’ effect in polymer solar cells based on under Contract No. FA9550-06-C-0010. Any opinion, findings and conclusions or
poly(3-hexylthiophene) and methanofullerenes. Adv. Funct. Mater. 17, recommendations expressed in this material are those of the authors and do not
1636–1644 (2007). necessarily reflect the views of the Air Force Office of Scientific Research. This material is
18. Erb, T. et al. Correlation between structural and optical properties of also based on work supported by the National Science Foundation under Grant
composite polymer/fullerene films for organic solar cells. Adv. Funct. Mater. No. 0348955 and Grant No. IIP-0740454. Additional thanks to The Cluster of Excellence
15, 1193–1196 (2005). ‘Engineering of Advanced Materials’ and the Alexander von Humboldt Foundation for
19. Peet, J. et al. Method for increasing the photoconductive response generous support (S.G.S.).
in conjugated polymer/fullerene composites. Appl. Phys. Lett. 89,
252105 (2006). Additional information
20. Zhokhavets, U., Erb, T., Gobsch, G., Al-Ibrahim, M. & Ambacher, O. Relation The authors declare competing financial interests: details accompany the full-text HTML
between absorption and crystallinity of poly(3-hexylthiophene)/fullerene films version of the paper at www.nature.com/naturematerials. Supplementary Information
for plastic solar cells. Chem. Phys. Lett. 418, 347–350 (2006). accompanies this paper on www.nature.com/naturematerials. Reprints and permissions
21. Muhlbacher, D. et al. High photovoltaic performance of a low-bandgap information is available online at http://npg.nature.com/reprintsandpermissions.
polymer. Adv. Mater. 18, 2931–2931 (2006). Correspondence and requests for materials should be addressed to M.D.
1 Chemical Sciences and Engineering Division, Argonne National Laboratory, 9700 South Cass Avenue, Argonne, Illinois 60439, USA, 2 Center for
Nanoscale Materials, Argonne National Laboratory, 9700 South Cass Avenue, Argonne, Illinois 60439, USA, 3 Department of Chemical Engineering,
School of Engineering & Applied Science, Yale University, 9 Hillhouse Avenue, New Haven, Connecticut 06520, USA, 4 Materials Science Division,
Argonne National Laboratory, 9700 South Cass Avenue, Argonne, Illinois 60439, USA, 5 Energy Systems Division, Argonne National Laboratory,
9700 South Cass Avenue, Argonne, Illinois 60439, USA. † Present Address: Max-Planck-Institut für Metallforschung, Stuttgart, Germany.
*e-mail: vajda@anl.gov; curtiss@anl.gov.
Selectivity Activity
a CO2 d
16.3% Pt8¬10 /SnO/Al2O3 /AAO, 400 °C
CO Pt8¬10 /SnO/Al2O3 /AAO, 500 °C
Propylene
83.7% 0% Pt8¬10 /Al2O3 /AAO, 550 °C
b CO2
15.7%
VOx /Al2O3, 1.4 V nm¬1, 390 °C
Catalyst
Propylene CO VOx /Al2O3, 8.0 V nm¬1, 390 °C
64% 20.3% VOx /Al2O3, 16.6 V nm¬1, 390 °C
VOx /Al2O3, 34.2 V nm¬1, 390 °C
c CO2
32% Pt /monolith, 400 °C
Figure 2 | Catalyst activity and selectivity. a–c, Selectivity of the Pt8−10 -based catalysts at various temperatures and support compositions: SnO/Al2 O3
at 400 ◦ C (a), SnO/Al2 O3 at 500 ◦ C (b) and Al2 O3 at 550 ◦ C (c). d, TOFs of propylene produced on the Pt8−10 catalysts (green) and reference ODH
catalysts (grey) expressed as number of propylene molecules formed per metal atom. Pt monolith and vanadia data from refs 29 and 22, respectively. See
Supplementary Information for more details.
Pt4 + C3H8(g)
0.24 eV 0.42
0.54
Pt4---C3H8
0.37
0.95
Pt4-(H,C3H7)
Pt4-(2H,C3H6)
Figure 3 | Reaction path. Diagram of primary reaction steps in channel A (see formula (1)) from DFT calculations for the dehydrogenation of propane on a
Pt4 cluster leading to formation of propylene adsorbed on the cluster. Energies (in eV) of the equilibrium structures are relative to the reactants. Energy
barriers for the transition state structures are relative to the preceding equilibrium structure (‘true’ barriers). The first barrier corresponds to breaking of
the first C–H bond (on the CH2 group) and the second barrier corresponds to breaking of the second C–H bond (on a CH3 group). A number of other
reaction steps involving hydrogen migration are not included in this diagram. The dotted lines in the structures indicate partial bonds. See Supplementary
Information for more details of structures and energies and other reaction pathways.
shape of the Pt clusters. The supported clusters, similar to those below). In addition, we carried out calculations to investigate the
used in this study, showed no evidence of agglomeration over a stability of a positively charged Pt8 cluster on alumina and found
temperature range of 20–400 ◦ C, and they maintained a three- that it withdraws electrons from the surface and becomes negatively
dimensional structure. At the higher temperatures used in the charged as in the case of the supported neutral cluster (see below).
catalytic testing in this study, there was also no evidence of any In a sense, these highly uniform clusters can be considered an ideal
change in cluster size, as there was no change in selectivity or activity model of a single-site catalyst on technologically relevant supports,
after 30 h of testing. We have also carried out density functional in which all active sites closely resemble each other2,19,20 .
theory (DFT) calculations14 to investigate the structure of a Pt8 The catalyst tests were carried out under atmospheric pressure
cluster on a θ -alumina surface. The structure (Fig. 1) indicates that in a flow reactor at temperatures from 400 to 550 ◦ C by using
the cluster maintains its three-dimensional structure, as is found 10 s.c.c.m. total flow of reactants in argon carrier gas with 2.63 and
in the X-ray studies. The Pt cluster forms Pt–O bonds with the 2.73 mol% of oxygen and propane, respectively. The temperature
surface, resulting in significant charge transfer to the cluster. The range was chosen to give a direct comparison with the performance
binding of the cluster to the surface (∼3 eV) is consistent with of reported Pt-and VOx -based ODH catalysts. The measured
the stability of the subnanometre Pt8 clusters on alumina, but turnover frequencies (TOFs) are shown in Fig. 2, along with the
does not significantly affect the cluster’s chemical reactivity (see highest reported values for platinum and vanadia. The TOFs were
which results in poorer overlap between the adsorbate and the 7. Lemire, C., Meyer, R., Shaikhutdinov, S. & Freund, H.-J. Do quantum size
reaction site orbitals in the transition state for breaking the C–C effects control CO adsorption on gold nanoparticles? Angew. Chem. Int. Ed. 43,
bond compared with that for the C–H bond. This argument has 118–121 (2004).
8. Wei, J. & Iglesia, E. Mechanism and site requirements for activation and
been put forward by Blomberg et al. for other surface reactions24 . chemical conversion of methane on supported Pt clusters and turnover rate
Computations indicate that CH4 (ref. 25) and H2 (ref. 26) have a comparisons among noble metals. J. Phys. Chem. B 108, 4094–4103 (2004).
small barrier and no barrier, respectively, for dissociation on a Pt4 9. Hvolbaek, B. et al. Catalytic activity of Au nanoparticles. Nano Today 2,
cluster, which supports the overlap explanation. 14–18 (2007).
10. Hutchings, G. J., Scurrell, M. S. & Woodhouse, J. R. Oxidative coupling of
To our knowledge, the work reported here is the first
methane using oxide catalysts. Chem. Soc. Rev. 18, 251–283 (1989).
investigation of size-preselected Pt clusters under realistic high- 11. Labinger, J. A. & Bercaw, J. E. Understanding and exploiting C–H bond
temperature catalytic conditions. It has revealed a very high activity activation. Nature 417, 507–509 (2002).
of subnanometre Pt-cluster-based catalysts for the ODH of propane 12. Cavani, F., Ballarini, N. & Cericola, A. Oxidative dehydrogenation of ethane
to propylene. Combined with quantum chemical studies, this work and propane: How far from commercial implementation? Catal. Today 127,
113–131 (2007).
has shown that the high activity is due to the under-coordination 13. Benz, L. et al. Landing of size-selected Ag+ n clusters on single crystal TiO2
of the Pt in the clusters and that the clusters favour the scission of (110)-(1 × 1) surfaces at room temperature. J. Chem. Phys. 122, 081102 (2005).
C–H bonds relative to C–C or C=C bonds. Some recent work in our 14. Lee, S., Fan, C., Wu, T. & Anderson, S. L. CO oxidation on Aun /TiO2
laboratory demonstrates that small gold clusters (Au6−10 ) are highly catalysts produced by size-selected cluster deposition. J. Am. Chem. Soc. 126,
active for propylene epoxidation, thus providing further evidence 5682–5683 (2004).
15. Winans, R. E. et al. Reactivity of supported platinum nanoclusters studied by in
for the unique catalytic properties of subnanometre clusters27 . In situ GISAXS: Clusters stability under hydrogen. Top. Catal. 39, 145–149 (2006).
the future, size-selected clusters stabilized on appropriate supports 16. Yoon, B. et al. Charging effects on bonding and catalyzed oxidation of CO on
with uniform surface chemistry hold great promise for design of Au8 clusters on MgO. Science 307, 403–407 (2005).
new catalytic materials. It will be a challenging task to scale up the 17. Sadykov, V. A. et al. Oxidative dehydrogenation of propane over monoliths at
production of size-selected clusters by more conventional chemical short contact times. Catal. Today 61, 93–99 (2000).
18. Pellin, M. J. et al. Mesoporous catalytic membranes: Synthetic control of pore
methods, but there are very encouraging efforts suggesting that this size and wall composition. Catal. Lett. 102, 127–130 (2005).
will ultimately be possible2,3 . 19. Bell, A. T. The impact of nanoscience on heterogeneous catalysis. Science 299,
1688–1691 (2003).
Methods 20. Somorjai, G. A., Contreras, A. M., Montano, M. & Rioux, R. M. Cluster
Production of narrow cluster size distributions. The clusters emerging from the chemistry and dynamics special feature: Clusters, surfaces, and catalysis.
continuous beam cluster source possess identical velocity. Owing to the variation Proc. Natl Acad. Sci. 103, 10577–10583 (2006).
in their kinetic energy with size, in addition to the single mass selection on a 21. Xu, Y., Shelton, W. A. & Schneider, W. F. Thermodynamic equilibrium
quadrupole mass filter, narrow distributions of clusters with one to four sizes can be compositions, structures, and reaction energies of Ptx Oy (x = 1–3) clusters
isolated using a quadrupole deflector operated in energy filter mode. predicted from first principles. J. Phys. Chem. B 110, 16591–16599 (2006).
22. Argyle, M. D., Chen, K., Bell, A. T. & Iglesia, E. Effect of catalyst structure
Determination of the fraction of clusters in the pores acting as active catalysts. on oxidative dehydrogenation of ethane and propane on alumina-supported
As only 10.5% of the facing area of the membrane was exposed to clusters, and to vanadia. J. Catal. 208, 139–149 (2002).
the propane feed, a multiplication factor of 9.5 is applied to obtain the corrected 23. Redfern, P. C. et al. Quantum chemical study of mechanisms for oxidative
propane conversion rate on the Pt-cluster-coated fraction of the AAO membrane. dehydrogenation of propane on vanadium oxide. J. Phys. Chem. B 110,
By calculating the relative pore opening area to the total surface area of the 8363–8371 (2006).
membrane, 28% of the Pt8−10 particles, corresponding to 252 ng of Pt metal, enter 24. Blomberg, M. R. A., Siegbahn, P. E. M., Nagashima, U. & Wennerberg, J.
the pores. The rest forms metallic platinum on the face of the AAO membrane as Theoretical study of the activation of alkane C–H and C–C bonds by different
confirmed by X-ray photoemission spectroscopy. Metallic Pt is known to possess transition metals. J. Am. Chem. Soc. 113, 424–433 (1991).
orders of magnitude lower TOFs for ODH of propane28 . Thus, 252 ng of Pt was 25. Xiao, L. & Wang, L. Methane activation on Pt and Pt4 : A density functional
used for the calculation of the TOFs. theory study. J. Phys. Chem. B 111, 1657–1663 (2007).
26. Cruz, A., Bertin, V., Poulain, E., Benitez, J. I. & Castillo, S. Theoretical study of
Activation barrier calculation from the experimental data. Using Arrhenius law the H2 reaction with a Pt4 (111) cluster. J. Chem. Phys. 120, 6222–6225 (2004).
and assuming that the conversion rate is roughly proportional to the fraction 27. Lee, S. et al. Selective propene epoxidation on immobilized Au6−10 clusters: The
of activated C–H bonds, an estimate was obtained of the activation barrier effect of hydrogen and water on selectivity and activity. Angew. Chem. Int. Ed.
from two available experimental values for Pt8−10 /SnO/Al2 O3 /AAO catalyst at (in the press, 2008) <http://dx.doi.org/10.1002/anie.200804154>.
different temperatures. 28. Yu, C., Xu, H., Ge, Q. & Li, W. Properties of the metallic phase of zinc-doped
platinum catalysts for propane dehydrogenation. J. Mol. Catal. A 266,
Theoretical methods. The calculations were done using the B3LYP functional 80–87 (2007).
with the LANL2DZ basis set for Pt and 6-31G∗ for Al, C, H and O. The Pt(111) 29. Silberova, B., Fathi, M. & Holmen, A. Oxidative dehydrogenation of ethane
results were obtained with plane-wave calculations using the RPBE functional. See and propane at short contact time. Appl. Catal. A 276, 17–28 (2004).
Supplementary Information for more details.
Acknowledgements
Received 21 July 2008; accepted 12 January 2009; The work at Argonne National Laboratory was supported by the US Department
published online 8 February 2009 of Energy, BES-Chemical Sciences, BES-Materials Sciences, and BES-Scientific User
Facilities under Contract DE-AC-02-06CH11357 with UChicago Argonne, LLC,
References Operator of Argonne National Laboratory. S.V. gratefully acknowledges the support by
1. Xu, Z. et al. Size-dependent catalytic activity of supported metal clusters. the Air Force Office of Scientific Research. We acknowledge grants of computer time at
Nature 372, 346–348 (1994). the Laboratory Computing Resource Center (LCRC) at Argonne National Laboratory,
2. Gates, B. C. Supported metal clusters: Synthesis, structure, and catalysis. the National Energy Research Scientific Computing Center (NERSC) at Lawrence
Chem. Rev. 95, 511–522 (1995). Berkeley National Laboratory and the Molecular Science Computing Facility (MSCF) at
3. Argo, A. M., Odzak, J. F., Lai, F. S. & Gates, B. C. Observation of ligand effects Pacific Northwest National Laboratory. The authors are indebted to E. Iglesia and P. Stair
during alkene hydrogenation catalysed by supported metal clusters. Nature for valuable discussions, A. Holmen for providing the exact dimensions of the monolith
415, 623–623 (2002). used in their studies of Pt-based catalysts and thank J. Moore for carrying out X-ray
4. Fu, Q., Saltsburg, H. & Flytzani-Stephanopoulos, M. Active Nonmetallic photoemission spectroscopy analysis of the Pt/AAO sample.
Au and Pt species on ceria-based water-gas shift catalysts. Science 301,
935–938 (2003). Additional information
5. Campbell, C. T. The active site in nanoparticle gold catalysis. Science 306, Supplementary Information accompanies this paper on www.nature.com/naturematerials.
234–235 (2004). Reprints and permissions information is available online at http://npg.nature.com/
6. Chen, M. S. & Goodman, D. W. The structure of catalytically active gold on reprintsandpermissions. Correspondence and requests for materials should be
titania. Science 306, 252–255 (2004). addressed to S.V. or L.A.C.
Mesoporous materials with tunable non-oxidic framework To produce the germanium-based chalcogenides, we used
compositions can exhibit new kinds of functionality including surfactant-templated coupling of (Ge9 )4− ions with the elemental
internal surfaces with high polarizability. As the chemical chalcogens S, Se and Te. The self-assembly of amphiphilic surfac-
and physical characteristics of the framework components can tant N -eicosane-N ,N -dimethyl-N -(2-hydroxyethyl) ammonium
induce useful catalytic, absorption and optoelectronic features, bromide (EDMHEAB) in formamide/ethylenediamine mixture so-
the mesoporous structure can promote fast mass diffusion ki- lution provides a suitable template for the growth of the mesostruc-
netics and size-selective transport of guest molecules1 . So far, tured frameworks. The strong reducing character of (Ge9 )4− favours
synthetic efforts have resulted in mesoporous metal chalco- the reduction of S, Se or Te atoms, producing mesostructured
genides on using structure-directing moulds of soft or hard germanium-linked chalcogenide frameworks (equation (1)).
templates. These include ordered mesoporous II–VI semicon-
ductors (such as CdS (refs 2,3), ZnS (ref. 4) and CdTe (ref. 5)). EDMHEAB + Ge4−
9 + 2N Q → (EDMHEA)x Ge9 Qn
Recently, template-free synthetic routes for high-surface-area 1
chalcogenide aerogels have been reported6,7 . Here, we describe −→ NU–GeQ-N (1)
NH4 NO3
a novel kind of porous materials based on germanium-rich
chalcogenide networks and ‘soft’ highly polarizable surfaces. The new family of Ge-rich chalcogenides NU–GeQ-1 and
We demonstrate that these materials can exhibit excellent NU–GeQ-2 (Q = S, Se and Te) was obtained after removal of
selectivity for separating hydrogen from carbon dioxide and the surfactant from the pores, through exchange with ammonium
methane. These highly polarizable mesoporous structures have nitrate (NH4 NO3 ) following mild heating at 80 ◦ C.
important implications for membrane-based gas separation Energy-dispersive X-ray spectroscopy (EDS) of the samples
process technologies including hydrogen purification. indicates an atomic ratio of Q:Ge9 ∼ 2 for NU–GeQ-1 and close to 4
Amphiphilic surfactants with polar head and nonpolar tail for NU–GeQ-2 materials. These ratios are very close to the nominal
groups can template the metal-ion-mediated crosslinking poly- composition. The EDS spectra did not show any signal from other
merization of Zintl clusters (Ge4 S4− 10 ) into mesostructured metal elements such as potassium or halides, suggesting the complete
chalcogenides8–11 . These materials possess nanoscale pore peri- polymerization reaction of (Ge9 )4− clusters. Thermogravimetric
odicity of hexagonal or cubic morphology. However, the pore analysis measurements indicated a ∼14–25 wt% of the surfactant
structure of these solids is not accessible because of the inability to content in as-prepared samples, whereas after the template removal
remove the surfactant. only a ∼6–7 wt% of the surfactant remaining in the pores (see
We have shown that Ge4− and (Ge9 )4− Zintl ions can result in Supplementary Figs S1–S3). The inorganic residue at 600 ◦ C was
ordered mesostructured germanium and binary intermetallic alloy found to be crystalline Ge and GeTe (in the case of NU–GeTe-1,
semiconductors12–15 . Unlike the linking polymerization reactions Supplementary Fig. S4). The analytical data and the estimated
of the (Ge9 )4− ions with metal ions, the new reaction described chemical formulae of as-prepared and mesoporous samples are
here involves polymerizing the (Ge9 )-clusters with chalcogens given in Table 1.
(S, Se and Te) in what seems to be a redox process. The The X-ray diffraction (XRD) patterns of the as-prepared
resulting materials, NU–GeQ-N (NU: Northwestern University), NU–GeQ-N materials show an intense broad diffraction peak
possess ‘Ge9 Qx ’ (Q = S, Se, Te) Ge-rich framework compositions at low 2θ scattering angles of 1.5◦ –3◦ and a broader feature at
with periodically ordered pores in hexagonal as well as cubic ∼3.5◦ –5.5◦ , which according to transmission electron microscopy
symmetry. This pore symmetry is preserved after template removal (TEM) could be attributed respectively to the (100) diffraction
to give mesoporous germanium-linked chalcogenide materials with peak and the overlapped pair of (110) and (200) peaks of two-
accessible pore channels and high surface areas. Because of the ‘soft’ dimensional hexagonal p6mm symmetry (Fig. 1a). Interestingly,
atoms (S, Se and Te), the polarizability of the internal surface of the XRD pattern of as-prepared NU–GeSe-2 and NU–GeTe-2
these materials is anticipated to be higher than those of metal oxides, shows a main diffraction peak associated with a weak shoulder
porous carbons and organic polymers and therefore provides an (2θ ∼ 2.3◦ ) and a broad feature, respectively. These peaks according
entirely new medium through which to study the diffusion and to TEM could be attributed respectively to the distinct and
separation of gasses. the overlapped pair of (211) and (220) diffraction peaks of
1 Department of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, Illinois 60208-3113, USA, 2 Materials Science Division, Argonne
National Laboratory, Chicago, Illinois 60439, USA. *Present address: Department of Materials Science and Technology, University of Crete, 71003
Heraklion, Greece. † e-mail: m-kanatzidis@northwestern.edu.
Table 1 | Analytical data of as-prepared and mesoporous NU–GeQ-1 and NU–GeQ-2 materials.
†
Sample Atomic ratio* Surfactant Chemical formula Colour Energy gap
(Q:Ge9 ) (wt%) (eV)
NU–GeS-1 1.9 7.2 (25.2) (EDMHEA)0.15 Ge9 S1.9 Red–orange 1.87
NU–GeS-2 4.0 6.9 (23.7) (EDMHEA)0.16 Ge9 S4.0 Brown 1.65
NU–GeSe-1 2.2 6.3 (14.1) (EDMHEA)0.15 Ge9 Se2.2 Red–brown 1.63
NU–GeSe-2 4.1 5.1 (22.6) (EDMHEA)0.14 Ge9 Se4.1 Brown 1.52
NU–GeTe-1 1.8 6.6 (16.7) (EDMHEA)0.16 Ge9 Te1.8 Light black 1.48
NU–GeTe-2 3.9 6.3 (16.5) (EDMHEA)0.21 Ge9 Te3.9 Black 1.20
20 nm 20 nm 20 nm 20 nm
c d 022 022
224
224
20 nm 20 nm 20 nm 20 nm
e f 022
022
224
224
50 nm 50 nm 20 nm 20 nm
Figure 2 | TEM images of mesoporous NU–GeQ-N. a–f, TEM images of hexagonal mesoporous NU–GeS-1 (a), NU–GeS-2 (b), NU–GeSe-1 (c) and
NU–GeTe-1 (e) viewed along the pore channel axis (left image in each part) and perpendicular to the pore channel axis (right image in each part) and TEM
images of cubic mesoporous NU–GeSe-2 (d) and NU–GeTe-2 (f) taken along the [110] direction (left image in each part) and along the [100] direction
(right image in each part). Insets in each panel show the corresponding FFT images.
provide a unique opportunity to study the gas adsorption CH4 and H2 of the NU–GeQ-1 chalcogenides indicate a significantly
and gas separation processes in this system as a function of higher affinity for CO2 and CH4 compared with hydrogen at 0 ◦ C
surface polarizability (for example, in going from the S to (Fig. 4a and Supplementary Fig. S11). Analysis of the adsorption
Se to Te analogue). data using the ideal adsorption solution theory24 reveals that
We examined the ability of NU–GeQ-1 materials to selectively NU–GeS-1 and NU–GeSe-1 have CO2 /H2 separation factors of
absorb carbon dioxide (CO2 ) and methane (CH4 ) from a gas ∼15–16 in the low-pressure limit, whereas the NU–GeTe-1 has
mixture with hydrogen (H2 ), which is relevant to hydrogen a separation factor of ∼39 in the low-pressure limit (Fig. 4b).
purification processes. Generally, hydrogen is produced by steam Similarly, a CH4 /H2 separation factor of ∼4–7 was determined for
reforming of natural gas or methanol cracking22 and its separation NU–GeS-1 and NU–GeSe-1 and a separation factor of ∼28 was
from gas mixtures containing CO2 , CO, N2 and CH4 is a determined for NU–GeTe-1 (see Supplementary Fig. S12).
technologically important procedure in industrial feedstock and The ‘soft’ character of the chalcogenide atoms in the framework
next-generation fuel sources23 . The adsorption isotherms for CO2 , is expected to favour strong dispersion/induced-dipole interactions
Figure 3 | Nitrogen adsorption–desorption isotherms of mesoporous NU–GeQ-1 and NU–GeQ-2 materials. a, NU–GeS-N. b, NU–GeSe-N. c, NU–GeTe-N.
NU–GeQ-1: black lines; NU–GeQ-2: red lines. The filled and open symbols are the adsorption and desorption data, respectively. Inset: Non-local density
function theory pore size distribution.
a 1.4 b 100
NU¬GeS-1
Adsorbed amount (mmol g¬1)
1.2 - 1 NU¬GeSe-1
Te 80 NU¬GeTe-1
Ge -1
Selectivity CO2/H2
1.0 ¬ e NU¬GePbTe-1
NU eS
0.8 ¬G CO2 60
NU
0.6 1
eS- 40
¬G
NU
0.4
e-1
ePbT 20
0.2 NU¬G
H2
0 0
0 200 400 600 800 1,000 0 100 200 300 400 500 600
p (mm Hg) p (mm Hg)
c 30 d 1
NU¬GeS-1 NU¬GeTe-1 NU¬GePbTe-1
NU¬GeSe-1
25
NU¬GeTe-1
Relative intensity (a.u.)
0
NU¬GePbTe-1
20
qst (KJ mol¬1)
¬1
15 10
qst (KJ mol¬1)
8
10 6 ¬2 CH4
4 CO2
2
5 CO
0
0 0.05 0.10 0.15 0.20 ¬3 H2
Amount adsorbed (mmol g¬1)
0
0 0.2 0.4 0.6 0.8 1.0 0 2 4 6 8 10 0 2 4 6 8 10
Amount adsorbed (mmol g¬1) Time (min) Time (min)
Figure 4 | Adsorption isotherms, heat of adsorption and gas mixture separation curves of mesoporous NU–GeQ-N. a, Adsorption isotherms for CO2
(filled symbols) and H2 (open symbols) of mesoporous NU–GeS-1 (black), NU–GeSe-1 (red), NU–GeTe-1 (blue) and NU–GePbTe-1 (green symbols) at 0 ◦ C.
The corresponding red lines are fits to the data. b, The CO2 /H2 selectivities predicted by applying the ideal adsorption solution theory to the CO2 and H2
single-component isotherms at 0 ◦ C of NU–GeQ-1 and NU–GePbTe-1 samples. c, The loading dependence of isosteric heat of the CO2 adsorption that is
determined by fitting the isotherms at −10 and 0 ◦ C to appropriate virial equations (see Supplementary Information). Inset: The loading dependence of
isosteric heat of adsorption for H2 of the NU–GePbTe-1 sample. d, The gas mixture separation curves (H2 /CH4 /CO2 /CO 20:22:18:5 v/v in N2 ) from
NU–GeTe-1 and NU–GePbTe-1 solids at 0 ◦ C. The relative intensities for each gas were obtained using a mass spectrometer gas analyser that could detect
the ion peaks at m/z+ = 16 (CH4 ), 44 (CO2 ), 12 (CO) and 2 (H2 ).
with guest molecules. In this case, the separation process relies heat of adsorption for probe molecules with different polariz-
on the differences between the adsorption strengths of adsorbates ability. The isosteric heat of adsorption (qst ) for CO2 , CH4 and
and the surface and is more selective for the strongly adsorbing CH3 CH3 (α(CH3 CH3 ) = 4.23 cm−3 ) of NU–GeQ-1 was deter-
molecule. Indeed the strength of these interactions is mostly mined to be ∼17–20, ∼17–29 and ∼23–41 kJ mol−1 at the limit
determined by the polarizability (α) of the adsorbate/adsorbent of zero coverage, respectively (see Fig. 4c and Supplementary In-
system. The NU–GeTe-1, which contains the more polarizable formation). These values of qst indicate favourable adsorption
telluride compared with the selenide NU–GeSe-1 and sulphide interactions between these molecules and the chalcogenide sur-
NU–GeS-1, exhibits higher solubility separation of the more face, which depend strongly on the polarizability of adsorbed
polarizable CO2 and CH4 molecules over H2 (polarizability (α): molecules as well as the porous framework. As the polariz-
α(CO2 ) = 2.51 > α(CH4 ) = 2.45 > α(H2 ) = 0.79 cm−3 ). ability of probe molecules increases and the soft character of
To examine the role of dispersion/induced-dipole interactions chalcogenide composition increases, the heat of adsorption between
on the adsorption selectivity process, we also determined the the two also increases.
25. Lin, H., Wagner, E. V., Freeman, B. D., Toy, L. G. & Gupta, R. P. Acknowledgements
Plasticization-enhanced hydrogen purification using polymeric membranes. These studies were supported primarily by the Nanoscale Science and Engineering
Science 311, 639–642 (2006). Initiative of the National Science Foundation under NSF Award Number EEC-0647560.
26. Verweij, H., Lin, Y. S. & Dong, J. Microporous silica and zeolite membranes We thank Peter C. Stair for the use of a mass-spectrometer gas analyser. This work made
for hydrogen purification. MRS Bull. 31, 756–764 (2006). use of the J.B. Cohen X-ray Diffraction facility and the Electron Probe Instrumentation
27. Ockwig, N. W. & Nenoff, T. M. Membranes for hydrogen separation. Center (EPIC) and Keck Interdisciplinary Surface Science (Keck-II) facility of NUANCE
Chem. Rev. 107, 4078–4110 (2007). Center at Northwestern University.
28. Verweij, H., Lin, Y. S. & Dong, J. Microporous silica and zeolite membranes
for hydrogen purification. MRS Bull. 31, 756–764 (2006).
29. Wang, K. & Stiefel, E. I. Toward separation and purification of olefins Additional information
using dithiolene complexes: An electrochemical approach. Science 291, Supplementary Information accompanies this paper on www.nature.com/naturematerials.
106–109 (2001). Reprints and permissions information is available online at http://npg.nature.com/
30. Broxon, T. J. & Chung, R.P.-T. Micellar bound Meisenheimer complexes. reprintsandpermissions. Correspondence and requests for materials should be
J. Org. Chem. 55, 3886–3890 (1990). addressed to M.G.K.
There are many structural and optical similarities between a liquid and a plastic flow. Thus, it is non-trivial to distinguish
between them at high pressures and temperatures, and a detailed description of the transformation between these phenomena
is crucial to our understanding of the melting of metals at high pressures. Here we report a shear-induced, partially disordered
viscous plastic flow from body-centred-cubic tantalum under heating before it melts into a liquid. This thermally activated
structural transformation produces a unique, one-dimensional structure analogous to a liquid crystal with the rheological
characteristics of Bingham plastics. This mechanism is not specific to Ta and is expected to hold more generally for other
metals. Remarkably, this transition is fully consistent with the previously reported anomalously low-temperature melting curve
and thus offers a plausible resolution to a long-standing controversy about melting of metals under high pressures.
H
ow and at what temperature metals melt at high pressures founded on intuitive phenomenological modelling and shear
are fundamental questions that have wide scientific impli- modulus analysis, Ross et al.21 have proposed a visco-elastic
cations, including our understanding of the Earth’s interior. phase (similar to a supercooled liquid22 ) at higher pressures (near
Despite dramatic advancements in experimental and theoretical 150 GPa) to bridge the low- and high-temperature melting curves,
methods, a remarkable controversy remains for the melting of while maintaining that the DAC melting is the true solid–liquid
metals at high pressures1–21 . Often1–6 , but not always7–9 , the melting boundary. Whether or not melting is true, it strongly signals a
curve of a metal obtained from laser-heated diamond-anvil cell transition from a b.c.c. crystal to a less ordered structure whose
(DAC) experiments1–6 is much flatter than those obtained by shock- identity remains a mystery.
wave experiments10 , piston–cylinder-apparatus measurements11 In searching for a thermally activated disordered structure,
and calculations12–17 . For example, extrapolating the Ta DAC melt- we investigated the role of shear in producing viscous plastic
ing curve reported in ref. 1 to 300 GPa (crosses in Fig. 1a) yields flow before melting. This is particularly relevant because we
a melting temperature that is approximately 5,000 K lower than have evidence from ab inito energy landscapes that significant
the shock-melting point10 (solid circle). Substantial variations also thermal softening is present at temperatures near DAC melting.
exist among calculations12–17 , but the majority of them disagree Coincidentally, a hot viscous plastic flow could also have the
qualitatively with the DAC melting slope. The underlying source appearance of liquid-like motion. Two key distinctions between
of this discrepancy is unlikely to be material specific, as it exists viscous plastic flow and liquid are (1) the absence or presence
for a majority of the transition metals that have been studied of long-range thermally diffusive atoms23 and (2) whether
experimentally1–6 . In addition, the flat DAC melting curve contrasts or not it supports a shear at rest. It is notable that DAC
sharply with the classical Lindemann melting law12 (red dashed experiments do not involve measuring atomic mobility1–6 .
curve). In contrast, a DAC melting curve consistent with shock Interestingly, Ross et al.21 found that the shear modulus has a
melting was recently reported for body-centred-cubic (b.c.c.) lead discontinuous change to a finite value from b.c.c. to the proposed
in ref. 9. In this study, X-ray signal scattered by the liquid was used visco-elastic phase.
as the detection criterion. It suggested that the previous melting Previously, non-hydrostatic stress and thermal pressure in DAC
criteria using onset-of-speckle detection and disappearance of solid had been studied primarily for insulators, which are poor thermal
X-ray signal alone are not always reliable, even though the previous conductors2,24–26 . In particular, dynamic recrystallization has been
DAC experiments reproduced the correct melting temperature at proposed as an explanation for DAC low-temperature melting of
ambient pressure1,4 . In comparison, the shock-melting criterion10 MgO (ref. 20). In addition, the possibility of plastic deformation was
based on sound-speed measurements works reasonably well for also hypothesized24 . However, previous searches27–29 for mechanical
identifying a solid-to-liquid boundary. However, both theory and instability of b.c.c. Ta by experiments (at low temperatures) and
DAC are often needed to provide more detailed identifications. calculations (with b.c.c. symmetry) were not successful. It is also
Several hypotheses18–21 have been proposed to explain the known that shear plays an important role in disordering transition
melting discrepancy of metals. For instance, Belonoshko et al. of colloidal crystals30 , viscosity of non-Newtonian fluids and the
have advocated solid-to-solid transition as the explanation for rheological behaviour of Bingham plastics31,32 . Later, we shall
the observed DAC melting18,20 . In particular, they have recently show that this thermally activated transition results in a novel
presented ab inito evidence of a b.c.c.-to-face-centred-cubic (f.c.c.) structure, different from that of the usual plastically deformed b.c.c.
transition in molybdenum (Mo) and suggested it as a candidate18 . metals. Instead, it matches the rheological description of a Bingham
Their study shed important light on the stability of Mo at plastic31,32 (for example a toothpaste or a tar), which behaves like
elevated temperatures and pressures, which might be connected a rigid solid under low shear stresses but flows like a viscous fluid
to the observed shock transition at 210 GPa (ref. 19). Intriguingly, under high shear stresses31,32 .
Lawrence Livermore National Laboratory, Livermore, California 94551, USA. *e-mail: wu5@llnl.gov.
a 12,000 b ε yz
Shock (ref. 10)
8,000
z[111]
Temperature (K)
MGPT (ref. 34)
DAC (ref. 1)
6,000 c
0.30
4,000
〈ΔR[111]〉
0.25
2,000 Plastic flow (MGPT)
B.c.c. (MGPT)
0.20
0
0 100 200 300
2,500 3,000 3,500
Pressure (GPa)
T (K)
Figure 1 | Predicted MD boundary between b.c.c. and plastic flow, as well as previously reported melting curves of Ta. a, The green dotted curve shows
the boundary between b.c.c. (blue triangles) and plastic flow (green diamonds). b, The atoms shown in a simulation cell are those identified by a central
symmetry parameter as disordered after transforming into a plastic flow. c, Averaged atomic displacements h1R[111] i along [111] as a function of
temperature near 30 GPa.
Unfortunately, the exact magnitude of shear in the DAC is the transition temperature was independent of the external stress
unknown2,25,26 . However, its presence is indisputable, even with once it was large enough to overcome the deformation barrier. We
pressure media, because compression by two diamonds is uniaxial. shall show later that the low deformation barrier of Ta is linked to
So far, the role of shear on melting of metals has not been seriously the flat energy landscape along the Burgers path36 .
considered, perhaps for several reasons. First, DAC experiments The technique we used to identify the disordering transition is
have demonstrated that pressure media have no significant effect based on the widely used method of identifying amorphization37
on melting temperatures, although they do influence X-ray patterns by monitoring changes in the radial distribution function g (r).
near melting6 . Second, shear is expected to dissipate quickly in The main characteristic feature of the disordered structure is the
metals near melting. Indeed, we find that the timescale of shear appearance of the shoulder feature in g (r), shown in Fig. 2c.
relaxation falls in between those of melt and plastic flow. Thus, shear The merging of peaks reflects that the third- and fourth-nearest-
is not important for determining the solid-to-liquid boundary, but neighbour distances in the {1̄10} planes become indistinguishable.
it is critical in inducing plastic flow, particularly when significant Note that this structure is different from an amorphous or a
thermal softening is present. Later, we shall show that the transition supercooled liquid, which possesses only short-range order21,22 .
temperature is approximately independent of external stresses after Instead, it has a partial disordered–partial crystalline structure
it exceeds a threshold value. We suggest that a combination of with a long-range order in the [111] direction. In fact, it shares
deviatoric and thermal stresses in the DAC is likely to exceed the structural similarity with the smectic phase of a liquid crystal, or
transformation threshold, even with the use of pressure media. one-dimensional polymer. However, the interchain interactions
Investigation of structural transformations at high pressure of Ta are expected to be much stronger than those of liquid
requires a very accurate interatomic potential, making Ta an ideal crystals and polymers. We have chosen to call it a one-dimensional
prototype for this purpose. In particular, Ta is one of only a few plastic flow because, as shown in Fig. 2, it looks like a standard
metals for which a highly accurate quantum-based, many-body high-temperature b.c.c. crystal (blue panels, b) in the {111} and
interatomic model-generalized pseudopotential theory (MGPT) {112̄} planes, whereas it is similar to a glass (red panels, d) in
potential has been developed for applications across a wide range the {1̄10} planes. In addition, analysis of the angular distribution
of temperatures, pressures and strains33 . Specifically, the melting function during melting shows that the [111] chains are the last
curve (solid red curve in Fig. 1a), established in ref. 34 using crystalline structure to survive before melting (see Supplementary
the MGPT potential33 , is in excellent agreement with the high- Information, Fig. S1). This is consistent with {1̄10} [111] being
temperature shock-melting10 point (black circle) as well as with the the primary Ta slip system35 in deformation and [111] chains
ambient-pressure melting point. Using the same MGPT potential, having the strongest atomic bonding. Below, we shall show that
we conducted a series of molecular dynamics (MD) simulations the origin of this transition is linked to the observed anisotropy
on heated b.c.c. Ta crystals with and without shear. Details of the of the energy landscape, which leads to thermal softening of {1̄10}
simulations are provided in the Methods section. planes under shear.
Here we report our finding of a thermally activated and Shear provides the critical driving force of this transition, the
shear-induced structural transformation to an anisotropic signature of a Bingham plastic31,32 . To make a Bingham plastic flow,
Bingham-like plastic flow (green dotted curve) that coincides with the driving shear stress has to exceed the yield stress. Without shear,
the DAC melting curve (see Fig. 1a). In contrast, under hydrostatic we found no structural transformation before melting. For instance,
compression, the same MGPT potential33 gave the solid–liquid Fig. 2b shows that unsheared b.c.c. Ta remains as b.c.c. at 3,250 K.
boundary (solid red line in Fig. 1a) at much higher temperatures34 . Under sufficient shear, Ta becomes partially disordered at the same
For each fixed volume, initial strain was applied along a b.c.c. temperature (Fig. 2c). This result explains why the DAC produces
slip35 system of {1̄10} [111] . After heating the system to the the correct melting temperature at ambient pressure, a stress-free
targeted temperature and equilibrating it for 40 ps, we analysed the state. The threshold stress reflects the materials’ resistance to flow,
structure to determine whether it remained b.c.c. (blue triangles) or which is material specific and depends strongly on pressure. Rough
transformed into a plastic flow (green diamonds). We found that estimates of threshold strain of Ta fall between 10% and 13% for a
{111} plane
–
–
[112]
–
[111]
{112} plane
–
[110]
[111]
–
{110} plane
–
[112]
0
2 4 6 2 4 6 2 4 6 2 4 6
r (Å) r (Å) r (Å) r (Å)
Figure 2 | Structural properties compared between b.c.c., one-dimensional plastic flow and glass. Planar projections of Ta atomic coordinates are shown
in 8 Å × 8 Å regions, and the corresponding radial distribution functions g(r) near 30 GPa. a, Different symbols indicate atoms from different layers of an
unsheared b.c.c. crystal at 298 K. b (blue panels), A b.c.c. crystal heated to 3,250 K without shear. c (green panels), A one-dimensional (1D) plastic flow
induced by shear and heating. d (red panels), A supercooled isotropic glass. The arrow in c indicates the shoulder feature in g(r) of a plastic flow. The third
and fourth peaks of g(r) of a b.c.c. crystal correspond to the distances inside a {1̄10} plane, shown by arrows in a.
pressure range of 15–65 GPa. In addition, we found that the flowing 800 K to 3,250 K, we would obtain a value of 8.9 GPa. However,
transition occurs substantially faster than shear dissipation. For a much smaller value for NaCl was also reported to be 1.2 GPa
example, the blue curve in Fig. 3a shows that it takes ∼1.4 ps to at 3,250 K and a pressure of 15.1 GPa (ref. 2). Interestingly, the
transform from b.c.c. Ta to the plastic flow at 3,250 K and 17 GPa. higher (8.9 GPa) and the lower values (1.2 GPa) coincide closely
In comparison, the stress of the simulation cell relaxed at a much with the MD stresses before and after thermal relaxation (9 GPa
slower rate (green curve) before reaching a steady state of 1.75 GPa, and 1.75 GPa, respectively). However, we should exercise extreme
and was roughly unchanged up to 375 ps. The finite stress value caution when making such quantitative comparisons between
is consistent with viscous plastic flow rather than liquid. Before stresses obtained using different methods, materials and conditions.
exceeding the threshold value, the external stress influences the As MD simulation is known to overestimate the yield strength of a
transition temperature, as Fig. 3b shows. For instance, no transition material by as much as an order of magnitude20 , we expect an MD
takes place before melting if the external stress is less than 5 GPa at stress to be an upper bound. Qualitatively, we suggest that the initial
pressures near 17 GPa. However, once the external stress is larger stress in a DAC is likely to be larger than the threshold stress, even
than the threshold stress (∼9 GPa at a pressure of 17 GPa), the with soft pressure media, thereby explaining why pressure media do
transition temperature no longer changes with stress. not influence the observed transition temperatures.
Unfortunately, the actual range of the initial stress in a DAC On transition, disordered atoms, identified by the central
Ta sample is unknown. In fact, the previously reported values symmetry parameter, increase significantly in population, and
of thermal stress in DAC are fairly scattered2,25,26 . For instance, appear throughout the system (Fig. 1b). In contrast, the plastic
a significant thermal pressure (>10 GPa) has been reported for deformation of a b.c.c. metal is usually propagated by well-localized
MgO (ref. 25). For NaCl, if we extrapolated a reported temperature screw dislocations, which are accompanied by possible debris38 .
dependence of thermal pressures (∼0.003 GPa K−1 ; ref. 26) from Also, a decrease of the viscosity is observed along the slip direction.
a a δ
8
b.c.c. → h.c.p.
θ : 109.74° → 120.0° θ
δ : 0 → 0.0833
6
Shear stress (GPa)
4
b
Ta
2
h.c.p. 5,000
ΔE (K)
3,250 K 3,500
1,500
0
0 40 80 120 160 200 b.c.c.
Time (ps)
108 8
2 0.0
b θ ( 11 4
°) 0.0
116 δ
4,200 0
18. Belonoshko, A. B. et al. Molybdenum at high pressure and temperature: 36. Burgers, W. G. On the process of transition of the cubic-body-centered
Melting from another solid phase. Phys. Rev. Lett. 100, 135701 (2008). modification into the hexagonal-close-packed modification of zirconium.
19. Hixson, R.A., Boness, D.A., Shaner, J.W. & Moriarty, J. A. Acoustic velocities Physica 1, 561–586 (1934).
and phase transitions in molybdenum under strong shock compression. 37. Szczgielska, A., Burian, A., Duber, S., Dore, J. C. & Honkimaki, V. Radial
Phys. Rev. Lett. 62, 637–640 (1989). distribution function analysis of the graphitization process in carbon materials.
20. Belonoshko, A. B. & Dubrovinsky, L. S. A simulation study of induced J. Alloys Compounds 328, 231–236 (2001).
failure and recrystallization of a perfect MgO crystal under non-hydrostatic 38. Jaime, M., Cai, W. & Bulatov, V. V. Dynamic transitions from smooth to rough
compression: Application to melting in the diamond-anvil cell. Am. Mineral. to twinning in dislocation motion. Nature Mater. 3, 158–163 (2004).
82, 441–451 (1997). 39. Wang, Y. et al. Amorphouslike diffraction pattern in solid metallic titanium.
21. Ross, M., Errandonea, D. & Boehler, R. Melting of transition metals at high Phys. Rev. Lett. 95, 155501 (2005).
pressure and the influence of liquid frustration: The early metals Ta and Mo. 40. Hattori, T. et al. Does bulk metallic glass of elemental Zr and Ti exist?
Phys. Rev. B 76, 184118 (2007). Phys. Rev. Lett. 96, 255504 (2006).
22. Jakse, N., Le Bacq, O. & Pasturel, A. Predication of the local structure of liquid 41. Streitz, F. H., Glosli, J. N. & Patel, M. V. Beyond finite-size scaling in
and supercooled tantalum. Phys. Rev. B 70, 174203 (2004). solidification simulations. Phys. Rev. Lett. 96, 225701 (2006).
23. Wolf, D., Okamoto, P. R., Yip, S., Lutsko, J. F. & Kluge, M. Thermodynamic 42. Rodney, D. & Martin, G. Dislocation pinning by small interstitial loops: A
parallels between solid-state amorphization and melting. J. Mater. Res. 5, molecular dynamics study. Phys. Rev. Lett. 82, 3272–3275 (1999).
286–301 (1990). 43. Allen, M. P. & Tildesley, D. J. Computer Simulations of Liquids
24. Belonoshko, A. B., Ahuja, R. & Johansson, B. Molecular dynamics of LiF (Clarendon, 1987.).
melting. Phys. Rev. B 61, 11928–11935 (2000). 44. Gonze, X. et al. First-principles computation of material properties: The
25. Fiquet, G, Andrault, D., Itié, J. P., Gillet, P. & Richet, P. X-ray diffraction of ABINIT software project. Comput. Mater. Sci. 25, 478–492 (2002).
periclase in a laser-heated diamond-anvil cell. Phys. Earth Planet. Inter. 95, 45. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation
1–17 (1996). made simple. Phys. Rev. Lett. 77, 3865–3868 (1996).
26. Vinod, K., Malik, V. S., Sharma, S. K. & Srivastava, S. K. Temperature 46. Troullier, N. & Martins, J. L. Efficient pseudopotentials for plane-wave
dependence of thermal pressure of NaCl and KCl crystals. J. Phys. Chem. Solids calculations. Phys. Rev. B 43, 1993–2006 (1991).
68, 32–35 (2007). 47. Monkhorst, H. J. & Pack, J. D. Special points for Brillouin-zone integrations.
27. Foata-Prestavoine, M., Robert, G., Nadal, M. H. & Bernard, S. First-principles Phys. Rev. B 13, 5188–5192 (1976).
study of the relations between the elastic constants, phonon dispersion curves,
and melting temperatures of b.c.c. Ta at pressures up to 1000 GPa. Phys. Rev. B Acknowledgements
76, 104104 (2007). The authors would like to thank J. A. Moriarty for providing the MGPT Ta potential,
28. Gulseren, O. & Cohen, R. E. High-pressure thermoelasticity of M. Ross, N. C. Holmes, W. J. Evans, M. J. Lipp, M. Tang, R. Gee and D. A. Orlikowski
body-centred-cubic tantalum. Phys. Rev. B 65, 064103 (2002). for useful discussions and K. Kline and J. McInnis for their contributions in preparation
29. Orlikowski, D., Soderlind, P. & Moriarty, J. A. First-principles thermoelasticity of the manuscript and figures. This work was carried out under the auspices of the US
of transition metals at high pressure: Tantalum prototype in the quasiharmonic Department of Energy by Lawrence Livermore National Laboratory under Contract
limit. Phys. Rev. B 74, 054109 (2006). DE-AC52-07NA27344.
30. Butler, S. & Harrowell, P. The shear-induced disordering transition
in a colloidal crystal: Nonequilibrium Brownian dynamic simulations. Author contributions
J. Chem. Phys. 103, 4653–4671 (1995). C.J.W. originated the central idea, performed and analysed the Ta MD simulations
31. King, R. P. Introduction to Practical Fluid Flow 117 and the Ta ab initio electronic structure calculations, carried out code development
(Butterworth-Heinemann, 2002). (central symmetry) and wrote the paper; P.S. contributed in ab initio electronic structure
32. Bingham, E. C. An investigation of the laws of plastic flow. US Bureau of calculations and manuscript editing; J.N.G. carried out MD code development; J.E.K.
Standards Bulletin 13, 309–353 (1916). contributed in manuscript editing; all contributed to discussions.
33. Moriarty, J. A. et al. Quantum-based atomistic simulation of materials
properties in transition metals. J. Phys. Condens. Matter 14, 2825–2857 (2002). Additional information
34. Moriarty, J. A. et al. Supplemental Proceedings: Volume 1: Minerals, Metals Supplementary Information accompanies this paper on www.nature.com/naturematerials.
and Materials under Pressure for the 2008 TMS Annual Meeting, New Orleans, Reprints and permissions information is available online at http://npg.nature.com/
LA, 313. reprintsandpermissions. Correspondence and requests for materials should be
35. Hirth, J. P & Lothe, J. Theory of Dislocation 275 (Krieger, 1992). addressed to C.J.W.
Domain walls may play an important role in future electronic devices, given their small size as well as the fact that their location
can be controlled. Here, we report the observation of room-temperature electronic conductivity at ferroelectric domain walls
in the insulating multiferroic BiFeO3 . The origin and nature of the observed conductivity are probed using a combination of
conductive atomic force microscopy, high-resolution transmission electron microscopy and first-principles density functional
computations. Our analyses indicate that the conductivity correlates with structurally driven changes in both the electrostatic
potential and the local electronic structure, which shows a decrease in the bandgap at the domain wall. Additionally, we
demonstrate the potential for device applications of such conducting nanoscale features.
C
orrelated oxide systems are an exciting and challenging
area of condensed-matter research, with their interacting effects at the nanoscale.
and competing charge, spin, orbital and lattice degrees Here we describe the observation of a new behaviour—
of freedom forming new electronic and magnetic phases1,2 . room-temperature electrical conductivity—at ferroelectric domain
These phases can be controlled through stress, optical excitation walls in the prototypical multiferroic BiFeO3 (BFO). BFO
and electric and magnetic fields and have great potential for is a room-temperature G-type antiferromagnet (TN ∼ 650 K)
applications in the fields of spintronics, information storage and and a rhombohedral ferroelectric (TC ∼ 1,103 K), with a large
communications. Among the correlated oxides, the multiferroics, spontaneous ferroelectric polarization (∼90 µC cm−2 ) along the
which show more than one type of ferroic order in the same phase, pseudocubic 111 directions16,17 . Such rhombohedral ferroelectrics
are attracting particular interest3–6 . The defining characteristic of possess 71◦ , 109◦ and 180◦ domain walls forming on {101},{100}
a ferroic material is an order parameter (electric polarization and planes that satisfy the requirement that ±h ± k + l = 0,
in ferroelectrics, magnetization in ferromagnets or spontaneous respectively18 . All three wall orientations have been observed
strain in ferroelastics) that has different, energetically equivalent in BFO (refs 19,20).
orientations, the orientation of which can be selected using an Epitaxial BFO films (∼100 nm thick) were grown using laser
applied field. This often leads to the appearance of domains molecular-beam epitaxy in (111), (110) and (100) orientations,
of differently oriented regions, separated by domain walls, using carefully controlled single-crystal SrTiO3 substrates. A thin
coexisting in a sample7 . Such domain walls will become more 50 nm layer of epitaxial SrRuO3 was used as a bottom electrode
technologically important as the dimensions of individual elements for electrical-contact purposes. Ferroelectric domains were im-
in devices continue to shrink. aged using piezoresponse force microscopy (PFM) as described
Although the morphology and properties of domains and their previously19 . Controlled ferroelectric domain patterns were written
walls have been studied for more than 50 years, in recent times there using PFM by applying a d.c. voltage to the probe tip. Local electrical
has been increasing focus on novel functionality at domain walls8–12 . conductivity was measured using high-resolution conductive
For example, it has been predicted theoretically that the ferroelectric atomic force microscopy (c-AFM) (details of measurements are
walls in magnetoelectric multiferroics can be ferromagnetic even provided in the Methods section) by applying a bias voltage (below
if the domains themselves are antiferromagnetic9–11 . Conversely, the polarization switching voltage) between the conductive AFM
spin rotation across ferromagnetic domain walls in insulating tip and the bottom electrode of the sample.
ferromagnets can induce a local polarization in the walls of 100-nm-thick epitaxial films grown on (110) surfaces show a
an otherwise non-polar material5,12 . Experimentally, unusual two-variant ferroelectric domain structure in the as-grown states
functional properties of domain walls have also been observed: with average domain sizes between 5 and 10 µm. On electrical
preferential doping along domain walls has been reported to switching at high fields, all three variations of domain walls can
induce two-dimensional superconductivity in WO3−x (ref. 13) be created21 ; therefore, we begin our study on these (110)-oriented
and enhanced resistivity in phosphates14 , whereas in paraelectric, films. The typical morphology is shown in Fig. 1a. We measured
non-polar SrTiO3 the ferroelastic domain walls seem to be the RMS roughness of the films to be ∼0.5 nm and there are
ferroelectrically polarized15 . Local probes, particularly those based no observable surface features, either before or after switching,
1 Department of Physics, University of California, Berkeley, 94720 California, USA, 2 Materials Sciences Division, Lawrence Berkeley National Laboratory,
94720 California, USA, 3 Department of Materials Science and Engineering, University of California, Berkeley, 94720 California, USA, 4 Department of
Materials Science and Engineering, National Chiao Tung University, HsinChu, 30010, Taiwan, 5 Institute of Structure Physics, Triebenberg Laboratory,
Technische Universität Dresden, DE-01062, Germany, 6 Center for Nanophase Materials Science, Oak Ridge National Laboratory, 37831 Tennessee, USA,
7 Forschungszentrum Dresden-Rossendorf, 01314 Dresden, Germany, 8 Department of Earth Sciences, University of Cambridge, Cambridge CB2 3EQ, UK,
9 Materials Department, University of California, Santa Barbara, 93106 California, USA. *These authors contributed equally to this work.
† e-mail: jseidel@berkeley.edu.
a b
D
1 µm
c d 0.3 pA
D D
0.15
0
180°
U 71°
U
109°
Figure 1 | Conductivity at different wall types. a, Topography of a BFO(110) thin film, r.m.s. roughness of 0.5 nm. b, Out-of-plane PFM image of a written
domain pattern in a monodomain BFO(110) film showing the out-of-plane polarization component of the domains to be either down, labelled as ‘D’
(white), or up, labelled as ‘U’ (black). c, In-plane PFM image of a written domain pattern in a monodomain BFO(110) film showing all three types of domain
wall, that is, 71◦ (blue), 109◦ (red) and 180◦ (green), as inferred from the combination of both out-of-plane and in-plane PFM images. In these images, both
the out-of-plane (U or D) component and the in-plane projection of the polarization direction (shown as an arrow) are also labelled. d, Corresponding
c-AFM image showing conduction at both 109◦ and 180◦ domain walls; note the absence of conduction at the 71◦ domain walls.
a b
+V 1 µm
BFO ¬2 V
Ground
SRO
STO ¬1.5 V
¬1 V
c 500 d 500
On wall
Off wall
400 400
300 300
Current (fA)
Current (fA)
200 200
100 100
0 0
¬2.5 ¬2.0 ¬1.5 ¬1.0 ¬05 0 0.5 1.0 1.5 2.0 2.5 0 50 100 150 200
Applied field (V) Time (s)
Figure 2 | Conduction at domain walls. a, Schematic illustration of the experimental c-AFM set-up. b, (Top) Out-of plane PFM image of a written 180◦
domain in a monodomain BFO(110) sample and (lower) corresponding c-AFM current maps for −1, −1.5 and −2 V sample bias made with a Pt-coated tip.
c, I–V curves taken both on the domain wall (black) and off the domain wall (red) reveal Schottky-like behaviour. d, Time dependence of the current both
on the wall (black) and off the wall (red) at an applied sample bias of −2 V. Results are qualitatively similar for N-doped diamond tips.
3.9
Pll
c 0.45
0.40
0.35
1 3 5 7 9 11 13 15 17 19 21 23 0 2 4 6 8 10 12 14 16 18 20 22 24
Unit cell
Figure 3 | Structural analysis of domain walls. a, Schematic diagram of 109◦ domain wall and exit-wave-reconstructed HRTEM image of a 109◦ domain
wall imaged along the [010] zone axis. b, Extracted a and c lattice parameters for each unit cell across the domain wall. c, Extracted Fe-ion displacement
relative to the Bi lattice for each unit cell across the domain wall. A close-up (upper panel) reveals an increase in the component of polarization
perpendicular to the domain wall.
I (nA)
SrRuO3 and STO is SrTiO3 , is shown in Fig. 3a. Analysis of this 0
image enables us to determine the lattice parameter in the plane
¬0.5
of the film (a) ([100]) and the lattice parameter out of the plane
of the film (c) ([001]) (Fig. 3b). As expected, the in-plane lattice ¬1.0
parameter is slightly smaller and the out-of-plane lattice parameter
larger than the values in bulk BFO (3.96 Å; ref. 16) owing to the ¬1.5
strain inherent in the epitaxial films. In addition, we find that ¬10 ¬5 0 5 10
both the in-plane and out-of-plane film lattice parameters are Bias (V)
unchanged in the vicinity of the domain wall. We also extract the
relative displacement of the Fe ion with respect to the Bi sublattice, Figure 4 | Proof of concept for device application. a, Schematic illustration
and resolve it into components parallel ([001]) and perpendicular (left) of in-plane electrode structure (green) and how scanning probe tips
([100]) to the domain wall (which lies in a (100) plane running can be used to controllably create conductive domain-wall features
vertically across the image and labelled with a dashed orange line) between electrodes. The images on the right show AFM (top) and
by quantitative analysis of the HRTEM data (Fig. 3c); this distance out-of-plane PFM (bottom) contrast for this written domain area on a
is representative of the local polarization. The component of the BFO(110) sample. b, I–V characteristics of the devices measured between
displacement parallel to the domain wall (along [001]) decreases in the two in-plane electrodes show that the current can be incrementally
magnitude to zero at the centre of the domain wall before changing controlled through creating or erasing the conducting domain walls.
to the same magnitude (but opposite sign) on the other side of the
wall, reflecting the change in polarization orientation of the domain. To confirm that our calculated structure is consistent with our
Interestingly, the perpendicular displacement component (along TEM data, we analysed the layer-by-layer polarization, defined as
[100]) shows a small increase at the domain wall; we return to the the sum over the bulk Born effective charges29 multiplied by the
implications of this later. Similar studies were completed on a 71◦ displacements of the ions from their centrosymmetric reference
domain wall as well (Supplementary Information, Fig. S2). Again, positions in each layer. As expected, the local polarization in the
only minor variation in lattice parameters was observed across the middle of the domain is close to the value calculated for bulk BFO
domain wall. In this case a similar step in Fe-ion displacement is using the same computationally and experimentally determined
observed parallel to the domain wall, but we are unable to resolve a lattice parameters (∼93 µC cm−2 ; refs 29–32), confirming that our
step in the perpendicular component across the wall. supercell is large enough to capture the essential physics. Consistent
To investigate the influence of these structural changes on the with our TEM analysis, we find an abrupt change in the parallel
electronic properties, we next performed a density functional study polarization component across the domain wall and a small change
of the structure and electronic properties for all three ferroelectric in the normal component at the domain wall.
domain-wall variants (details in the Methods section). First we Our calculations indicate that this small change in the normal
performed full structural optimizations of the ionic positions with component of the polarization across the 109◦ domain wall leads
the lattice parameters fixed to their experimental bulk values; in to a step in the electrostatic potential (planar and macroscopically
particular, the oxygen polyhedral rotations around the polar axis, averaged) of 0.15 eV across the domain wall (Table 1); a similar
which are known to have a profound effect on both the magnetic28 step was computed and explained previously across 90◦ domain
and electronic29 properties and cannot be easily extracted from the walls in PbTiO3 (ref. 22). Such a potential step should enhance
HRTEM data, were accurately calculated. As the sense of the oxygen the electrical conductivity by causing carriers in the material
rotations around the polar axis is independent of the direction to accumulate at the domain wall to screen the polarization
of polarization along the axis28 we studied two scenarios: first discontinuity. Our calculations for the 180◦ domain wall also
we initialized the sense of rotation to be continuous across the yield a variation in the normal component of the polarization,
domain boundary and second we changed the rotation sense when and a corresponding potential step of 0.18 eV (Table 1). The
the polarization direction changed. As expected, we found that normal component results from the polarization rotating towards
domain walls with continuous oxygen rotations are considerably successive adjacent corners of the perovskite unit cell, through a 71◦
lower in energy, because this avoids formation of an antiphase and than a 109◦ change in the polarization direction before reaching
boundary associated with the octahedral rotations. In addition, we the reversed polarization16 . This behaviour is in striking contrast to
investigated domain-wall configurations centred at both the Bi–O the 180◦ polarization reversal in tetragonal ferroelectrics22 , where
and Fe–O planes and found the Bi–O walls to be slightly lower the polarization changes in only one direction within the wall
in energy, confirming the findings of the HRTEM analysis. Our plane and no normal component occurs. The 71◦ wall, however,
lowest-energy calculated configuration for the 109◦ domain wall has exhibits no anomaly in the perpendicular component of the ion
a domain-wall energy of 206 mJ m−2 . displacement at the centre of the wall, again consistent with our
9. Přívratská, J. & Janovec, V. Pyromagnetic domain walls connecting 28. Ederer, C. & Spaldin, N. A. Weak ferromagnetism and magnetoelectric
antiferromagnetic non-ferroelastic magnetoelectric domains. Ferroelectrics 204, coupling in bismuth ferrite. Phys. Rev. B 71, 060401(R) (2005).
321–331 (1997). 29. Neaton, J. B. et al. First-principles study of spontaneous polarization in
10. Přívratská, J. & Janovec, V. Spontaneous polarization and/or magnetization multiferroic BiFeO3 . Phys. Rev. B 71, 014113 (2005).
in non-ferroelastic domain walls: Symmetry predictions. Ferroelectrics 222, 30. Li, J. F. et al. Dramatically enhanced polarization in (001), (101), and (111)
23–32 (1999). BiFeO3 thin films due to epitiaxial-induced transitions. Appl. Phys. Lett. 84,
11. Goltsev, V. et al. Structure and interaction of antiferromagnetic domain walls 5261 (2004).
in hexagonal YMnO3 . Phys. Rev. Lett. 90, 177204 (2003). 31. Ravindran, P. et al. Theoretical investigation of magnetoelectric behavior in
12. Mostovoy, M. Ferroelectricity in spiral magnets. Phys. Rev. Lett. 96, BiFeO3 . Phys. Rev. B 74, 224412 (2006).
067601 (2006). 32. Lebeugle, D. et al. Electric-field induced spin flop in BiFeO3 single crystals at
13. Aird, A. & Salje, E. K. H. Sheet superconductivity in twin walls: Experimental room temperature. Phys. Rev. Lett. 100, 227602 (2008).
evidence of WO3−x . J. Phys. Condens. Matter 10, L377–L380 (1998). 33. Coene, W. M. J., Thust, A., Op de Beeck, M. & Van Dyck, D.
14. Bartels, M. et al. Impurity-induced resistivity of ferroelastic domain walls in Maximum-likelihood method for focus-variation image reconstruction
doped lead phosphate. J. Phys. Condens. Matter 15, 957–962 (2003). in high resolution transmission electron microscopy. Ultramicroscopy 64,
15. Zubko, P. et al. Strain-gradient-induced polarization in SrTiO3 single crystals. 109–135 (1996).
Phys. Rev. Lett. 99, 167601 (2007). 34. Thust, A., Coene, W. M. J., Op de Beeck, M. & Van Dyck, D. Focal-series
16. Kubel, F. & Schmid, H. Structure of a ferroelectric and ferroelastic monodomain reconstruction in HRTEM: Simulation studies on non-periodic objects.
crystal of the perovskite BiFeO3 . Acta Crystallogr. B 46, 698–702 (1990). Ultramicroscopy 64, 211–230 (1996).
17. Lebeugle, D. et al. Very large spontaneous electric polarization in BiFeO3 35. Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio
single crystals at room temperature and its evolution under cycling fields. total-energy calculations using a plane-wave basis set. Phys. Rev. B 54,
Appl. Phys. Lett. 91, 022907 (2007). 11169–11186 (1996).
18. Streiffer, S. K. et al. Domain patterns in epitaxial rhombohedral ferroelectric 36. Anisimov, V. I., Aryasetiawan, F. & Liechtenstein, A. I. First-principles
films. I. Geometry and experiments. J. Appl. Phys. 83, 2742–2753 (1998). calculations of the electronic structure and spectra of strongly correlated
19. Zavaliche, F. et al. Multiferroic BiFeO3 films: Domain structure and systems: The LDA + U method. J. Phys. Condens. Matter 9, 767–808 (1997).
polarization dynamics. Phase Transit. 79, 991–1017 (2006).
20. Béa, H. et al. Influence of parasitic phases on the properties of BiFeO3 epitaxial Acknowledgements
thin films. Appl. Phys. Lett. 87, 072508 (2005).
The work at Berkeley is supported by the Director, Office of Science, Office of Basic
21. Cruz, M. P. et al. Strain control of domain-wall stability in epitaxial BiFeO3
Energy Sciences, Materials Sciences Division of the US Department of Energy under
(110) films. Phys. Rev. Lett. 99, 217601 (2007).
contract No DE-AC02-05CH1123. The authors from Berkeley would like to acknowledge
22. Meyer, B. & Vanderbilt, D. Ab initio study of ferroelectric domain walls in
the support of the National Center for Electron Microscopy, Lawrence Berkeley National
PbTiO3 . Phys. Rev. B 65, 104111 (2002).
Laboratory. J.S. acknowledges support from the Alexander von Humboldt Foundation.
23. Stemmer, S., Streiffer, S. K., Ernst, F. & Rühle, M. Atomistic structure
Y.H.C. would also like to acknowledge the support of the National Science Council,
of 90◦ domain walls in ferroelectric PbTiO3 thin-films. Phil. Mag. A 71,
R.O.C., under contract No NSC 97-3114-M-009-001. A.R. and S.G. acknowledge
713–724 (1995).
support from Deutsche Forschungsgemeinschaft through FOR 520 and Deutsche
24. Floquet, N. & Valot, C. Ferroelectric domain walls in BaTiO3 : Structural
Akademische Austauschdienst through GE 1202/5-1, and N.A.S. acknowledges support
walls model interpreting fingerprints in XRPD diagrams. Ferroelectrics 234,
from NSF DMR Award No DMR-0605852 and the Miller Institute for Basic Research
107–122 (1999).
in Science, UC Berkeley.
25. Jia, C.-L. et al. Atomic-scale study of electric dipoles near charged and
uncharged domain walls in ferroelectric films. Nature Mater. 7, 57–61 (2008).
26. Wu, X. & Vanderbilt, D. Theory of hypothetical ferroelectric superlattices Additional information
incorporating head-to-head and tail-to-tail 180 degree domain walls. Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Phys. Rev. B 73, 020103(R) (2006). Reprints and permissions information is available online at http://npg.nature.com/
27. Allen, L. J., McBride, W., O’Leary, N. L. & Oxley, M. P. Exit wave reconstruction reprintsandpermissions. Correspondence and requests for materials should be
at atomic resolution. Ultramicroscopy 100, 91–104 (2004). addressed to J.S.
Graphene shows promise as a future material for nanoelectronics owing to its compatibility with industry-standard lithographic
processing, electron mobilities up to 150 times greater than Si and a thermal conductivity twice that of diamond. The electronic
structure of graphene nanoribbons (GNRs) and quantum dots (GQDs) has been predicted to depend sensitively on the
crystallographic orientation of their edges; however, the influence of edge structure has not been verified experimentally. Here,
we use tunnelling spectroscopy to show that the electronic structure of GNRs and GQDs with 2–20 nm lateral dimensions varies
on the basis of the graphene edge lattice symmetry. Predominantly zigzag-edge GQDs with 7–8 nm average dimensions are
metallic owing to the presence of zigzag edge states. GNRs with a higher fraction of zigzag edges exhibit a smaller energy gap
than a predominantly armchair-edge ribbon of similar width, and the magnitudes of the measured GNR energy gaps agree with
recent theoretical calculations.
T
he surface structure of bulk, crystalline semiconductors has edge structure. Predominantly zigzag-edge GQDs with 7–8 nm
profound consequences on the development and manufac- average dimensions are metallic and diverge from the Eg –L scaling
turing of electronic devices. Bardeen1 explained that the pres- law owing to the presence of metallic zigzag edge states, which
ence of surface states results in binding of free carriers and induces spatially decay into the graphene interior with a 1.0–1.2 nm decay
the formation of Schottky barriers at semiconductor–metal inter- length. In addition to GQDs, we study the electronic structure of
faces. A deeper understanding of surface states has enabled scientists GNRs with 2–3 nm widths and 20–30 nm lengths. GNRs with a
and engineers to optimize the performance of integrated circuits higher fraction of zigzag edges exhibit a smaller energy gap than
for 50 years2 . Analogous to the surface states that exist in bulk a predominantly armchair-edge ribbon of similar width and the
crystals, the edge structure of nanometre-sized, two-dimensional magnitudes of the measured GNR energy gaps agree with recent
materials such as graphene, a one-atom-thick layer of carbon, can theoretical calculations.
significantly influence their electronic structure. Unlike previous studies of micrometre-sized, mechanically
Researchers have reported experimental transport exfoliated graphene monolayers on SiO2 (refs 17,18), nanometre-
measurements3–8 and theoretical studies9–11 of graphene quan- sized graphene monolayer samples do not typically exhibit the
tum dots (GQDs) and nanoribbons (GNRs) elucidating their 2.5 Å spaced, hexagonal graphene lattice in high-resolution STM
remarkable promise for future nanoelectronic applications. In topographs. As shown in Fig. 1, we typically observe either a
spite of theoretical calculations that predict a localized metal- 4.1 Å hexagonal lattice (Fig. 1a–c) or 2.5 Å triangular lattice patterns
lic state for the zigzag edge12 , all transport measurements of (Fig. 1d,e) in the STM topograph of the 3-Å-tall monolayer samples.
GQDs (ref. 3) and GNRs (refs 4–7) reveal only semiconducting Owing to the close proximity of the graphene edges in our samples,
behaviour. Furthermore, the electronic properties of the graphene we believe that the electron wavefunction scatters off the edges
are independent of crystallographic orientation4 in contrast to and the resultant interference pattern generates the observed STM
theoretical predictions9–12 . Recent theoretical studies show that topographic contrasts for the GQDs and GNRs. Our √interpretation
√
transport effects such as Coulomb blockade13 or a mobility is supported by previous STM studies where 3 × 3 R30◦
gap induced by edge disorder14,15 may affect the accuracy of hexagonal superstructures were observed near graphite terrace
bandgaps measured under transport conditions and explain the edges19–21 and graphene lattice defects22 , as well as the topographs
independence of energy gap and crystallographic orientation. By of 10–15-nm-wide GNRs defined on a highly oriented pyrolytic
probing the local electronic structure of GQDs and GNRs using graphite (HOPG) substrate through ambient STM lithography23 .
ultrahigh-vacuum scanning tunnelling microscopy (UHV-STM), Figure 1a and b show the STM topograph and topographic
we detect that the crystallographic orientation of the graphene edges derivative, respectively, of a 2.3-nm-wide, 20-nm-long GNR.
significantly influences the electronic properties of nanometre-sized Figure 1b delineates the hexagonal pattern along the GNR and
graphene. Enabled by the development of an atomically clean, Fig. 1c shows a line contour taken from Fig. 1a, the location
in situ deposition method16 , we experimentally determine the of which is designated by the grey line. From Fig. 1c, the
energy gap (Eg )–size (L) relation for GQDs with 2–20 nm lateral nearest-neighbour spacing between the hexagons is 4.1 ± 0.3 Å.
dimensions and correlate the Eg measurements with the GQD The hexagonal pattern with a 4.1 Å average lattice spacing is
1 Beckman Institute for Advanced Science and Technology, University of Illinois, 405 North Mathews Avenue, Urbana, Illinois 61801-2325, USA,
2 Department of Materials Science and Engineering, University of Illinois, 1304 West Green Street, Urbana, Illinois 61801-2920, USA, 3 Department of
Electrical and Computer Engineering, University of Illinois, 1406 West Green Street, Urbana, Illinois 61801-2918, USA. *e-mail: ritter@engineering.uiuc.edu.
a d
e
1.5
Height (Å)
c 1.0
Height (Å)
4.1 Å
0.5 0.5 2.5 Å
0 0
0 5 10 15 20 25
0 10 20 30 40 50
Position (Å)
Position (Å)
Figure 1 | Atomic-resolution imaging √ of GQDs and GNRs. a,b, STM topograph (a) and topographic derivative image (b) of a 2.3-nm-wide GNR exhibiting
a hexagonal superstructure with a 3ao (4.3 Å) lattice constant. The white hexagonal lattice is used to show the hexagonal superstructure. c, Topographic
cross-section taken from a along nearest-neighbour hexagons as indicated by the grey line. The distance between the centre of adjacent hexagons is
4.1 ± 0.3 Å, consistent with the lattice constant of the hexagonal superstructure. d, STM topograph and inset topographic derivative images of a GQD with
≈7 nm lateral dimensions exhibiting triangular lattice symmetry. A white hexagonal lattice is overlaid on the derivative image to highlight the orientation of
the C lattice. e, Line contour taken across the grey line in d showing the 2.5 Å spacing between nearest-neighbour hexagons. Imaging parameters: −0.5 V,
100 pA (a,b), −2 V, 50 pA (d). The scale bars in a and d represent 2 nm.
d e g
h i
j k
l m n
Figure 3 | STM topographs of the GQDs included in the Eg –L plot in Fig. 2. The samples are labelled in order from the smallest (a) to the largest (n). For
the GQDs where the graphene lattice was resolved, the topographic spatial derivative image is included to highlight the atomic-scale contrast. On the basis
of the atomic-scale resolution images, we label the crystallographic orientation at the graphene edges as follows: armchair (red), zigzag (blue) and
unassigned (black). All of the samples were exfoliated from the graphite crystal under ambient conditions except d, j and k, which were exfoliated in UHV
and H-passivated. The scale bars represent 2 nm. Data from a, e and f were taken from ref. 16.
our experiments. We observe the triangular symmetry for GQDs The experimental observations suggest that the source of the
that conform to the atomically flat contours of the surface, unlike triangular lattice symmetry is intrinsic to the electronic structure
the 1 nm height variations and 30-nm-long corrugations for rippled of the graphene because the atomic lattice of the graphene and
monolayer films deposited on SiO2 substrates17,18 . Furthermore, the the chemically inert Si(100)-2×1:H substrate are incommensurate.
triangular pattern has been observed on samples with heights of Although we are uncertain of the origin of the triangular symmetry
≈3 Å that do not cross step edges, unlike the sample in Fig. 1d. In in nanometre-sized monolayer samples, one potential source of the
addition, the pattern is extended across the entire graphene piece, triangular pattern is the edges, which could induce an interference
which suggests it is not caused by a local charge on the underlying pattern. We believe detailed simulations of graphene monolayers
Si due to charged dangling bonds24 or underlying dopant atoms25 . with different edge structures supported on the H–Si(100) surface
a b
Zigzag Mixed
(1)
(1) Zigzag
Armchair (2)
(2)
Unassigned (3)
(3)
0 0
¬0.5 6 ¬0.5
0 10 20 30 40 50 60 0 10 20 30 40
Position (Å) Position (Å)
e 80 f 80
40 (1) 40 (1)
dl/dV at V = 0 V (pA V¬1)
40 40 (2)
(2)
0 0
80 80
40 (3) 40 (3)
0 0
0 10 20 30 40 50 0 10 20 30 40
Position (Å) Position (Å)
Figure 4 | Comparison of a zigzag- and mixed-edge GQD using spatially resolved tunnelling spectroscopy. a, Predominantly zigzag-edge GQD from the
Eg –L plot in Fig. 2, which exhibits metallic character. The fraction of zigzag edges (rzig ) is 0.65. b, STM topograph of a 5 nm GQD with a mixture of both
zigzag and armchair edges (rzig = 0.44). Although zigzag edges are present, they are shorter than the sample in a. c, dI/dV − V spectra, obtained with
0.42 Å spatial resolution, plotted as a function of position across the green line in a. d, dI/dV − V spectra, obtained with 0.60 Å spatial resolution, plotted
as a function of position across the green line in b. e, Constant voltage, dI/dV versus position contours corresponding with the three numbered lines in a. In
general, we observe an increase in the differential conductance at the edges oriented along the zigzag direction. The spatial decay of the zigzag edge states
into the graphene interior prevents the observation of the expected 0.15 eV energy gap for this ≈8 nm sample. Line (2) is plotted along the solid green line
in a and does not include the dotted green line, which delineates the low conductance region at the left edge of the spectra map in c. f, Constant voltage,
dI/dV versus position contours recorded along the three lines in b. In contrast to the zigzag GQD, the differential conductivity does not increase near the
edges of the mixed-edge GQD and the magnitude of the differential conductivity is substantially lower than the zigzag GQD. The scale bars in a and b
represent 2 nm. STS setpoint: −2 V, 0.1 nA.
may provide insight into the origin of the triangular lattice in STM and black distinguishes edges that were unassigned. The samples
topographs of nanographene. shown in Fig. 3d,j,k were exfoliated in UHV and H-passivated (see
By using scanning tunnelling spectroscopy (STS), we measured the Methods section).
the size-dependent energy gap for 14 GQD samples. Each For the ten GQDs exhibiting an energy gap, the experimental
dI /dV (V /I ) − V spectrum used for the energy gap determination data were modelled with a power law, which resulted in
represents the average of 5–50 (depending on the particular sample) Eg (eV) = 1.57 ± 0.21 eV nm/L1.19±0.15 for the least-squares fit. The
individual I –V spectra from the interior of the GQDs. The length metallic samples were not included when modelling the exper-
scale for each sample was calculated by averaging the minimum imental Eg –L relation because the fit would be skewed owing
lateral dimension of the GQD and the results were compiled into to the limitations of room-temperature STS. The experimen-
the energy gap versus size plot shown in Fig. 2 (see Supplementary tal data are in close proximity to the predicted scaling trend
Information for more details). For the nine samples where the Eg (eV) = 1.68 eV nm/L (refs 26,27) resulting from quantum con-
graphene lattice was resolved, the samples are labelled armchair or finement and the linear dispersion of a graphene monolayer. We be-
zigzag on the basis of the edge symmetry present in the highest lieve that the reasonable correlation between simple theory and our
fraction (>0.5). Although the graphene was not atomically resolved experiment results from the nature of the STS measurement, which
at the edges, the crystallographic orientation of the edges was is not influenced by transport effects such as Coulomb blockade13 or
classified by determining the crystallographic orientation from the mobility gaps14,15 , and the enhanced spatial resolution of the STM.
atomic-scale contrast within the interior of the GQD (ref. 21). The Of the metallic GQDs in Fig. 2, two are predominantly zigzag
STM topographic images of each GQD composing the Eg –L plot edged (Fig. 3l,m) and the largest (Fig. 3n) has a 10 nm minimum
in Fig. 2 are shown in Fig. 3. The edge structure is represented by lateral dimension. The metallic GQD with 7 nm lateral dimensions
the coloured lines, where blue represents zigzag, red is armchair (Fig. 3j) and unresolved edge structure will be discussed below.
b e 100
1,000 Zigzag edge
0Å
0Å 80 2.5 Å
800 4.3 Å
dl/dV (pA V¬1)
c f 103
1,000
Armchair edge
0Å
800
4.3 Å
Zigzag edge (b)
8.5 Å ξ = 1.17 ± 0.14 nm
600
12.8 Å
Zigzag edge (e)
102 ξ = 0.95 ± 0.11 nm
400
200
Figure 5 | Detection of zigzag-edge state for GQDs. a, STM topograph of a ≈15 nm GQD with the edge lattice symmetry labelled. b,c, Comparison of
dI/dV spectra collected near the zigzag edge (b) and armchair edge (c) of the GQD sample. Whereas the zigzag edge in b exhibits a localized energy state
at −0.33 eV, a peak is not detected near the armchair-terminated edge in c. d,e, STM topograph (d) and dI/dV spectra (e) from the right zigzag edge of
the green line in Fig. 4a, which shows the spatial decay of the zigzag edge state at 0.29 eV into the interior of the graphene. The zigzag edge state is located
at 0.29 eV compared with the −0.33 eV position for the zigzag edge state of the sample in b. The energetic shift in the zigzag edge state peak is probably
due to charge transfer from the different chemical functional groups bound to the graphene edges of the respective samples. f, dI/dV zigzag-state peak
height plotted as a function of position on a semi-logarithmic scale for the samples in b and e. The decay length (ξ ) for the edge states is extracted from
the slope of the line. For f, the error of the length was determined by the spatial resolution of the measurement. The dI/dV standard deviation was
calculated from individual spectra used to generate the spatially averaged spectra in b and e. The scale bars in a and d represent 2 nm.
For the largest GQD (Fig. 3n), we believe that the reduced energy samples. The colour scale represents the value of the differential
gap induced by quantum confinement is not resolved owing to the conductance for Fig. 4c,d.
60.1 eV constraints of our room-temperature STS measurements. The dI /dV spectra maps for the GQDs in Fig. 4a,b exhibit
The two zigzag-edged GQDs diverge from the 0.15 eV expected pronounced differences. In Fig. 4c, the right end of the spectra map
energy gaps based on the experimental fit in Fig. 2. The metallic shows higher conductance with a localized state at 0.29 V near the
character of the zigzag-edged, 7–8 nm samples suggests that the pristine edge terminated along the zigzag direction. The left end is
fraction of zigzag edges influences the measured energy gap. also terminated along the zigzag direction, but there is a reduced
From Fig. 4, we gain more insight into the effect of conductance near the 3 Å topographic protrusion. Although the
zigzag-terminated edges and the measured energy gap by compar- identity of the topographic protrusion cannot be determined with
ing the spatially resolved, differential conductivity of a metallic, certainty, we speculate the protrusion may be a cluster of C atoms
zigzag-edge GQD (rzig = 0.65) and a semiconducting GQD with from the exfoliation process on the basis of its height, reduced local
a lower fraction of zigzag edges (rzig = 0.44). To quantify the density of states and spatial location 2 nm from the edge. However,
relative concentration of zigzag edges, rzig (ref. 28) represents the we cannot eliminate the possibility that localized protrusions at
ratio of zigzag edges to the total length of resolved zigzag and the edge result from ambient functionalization of the edges during
armchair edges, excluding edges that were unassigned. Figure 4a the ex situ exfoliation of graphite. Although the sample crosses a
and b show the STM topographs of a metallic, zigzag GQD and monatomic Si step edge, multiple 0.42 Å spaced, spatially resolved
a mixed-edge GQD, respectively. The longest, continuous zigzag tunnelling spectra maps across the GQD reveal that the graphene
edge for the monolayer in Fig. 4a is 4.7 nm, whereas the maximum electronic structure does not vary as the sheet smoothly traverses
zigzag edge length for the mixed-edge sample is 1.8 nm. Figure 4c,d the Si terrace edge.
shows spatially resolved maps of the local graphene electronic In Fig. 4d, the differential conductance spatial map for the
structure where the dI /dV − V tunnelling spectra are plotted mixed-edge GQD exhibits a uniform 0.3–0.4 eV energy gap across
as a function of distance across the green lines of the respective the entire piece even at the edges. The general features of the
a
Armchair
Zigzag b
Unassigned
b c 2
Armchair (rzig = 0.15)
6
Eg = 0.38 eV
4
Wavg = 2.9 ± 0.3 nm 1
0 0
¬1.0 ¬0.5 0 0.5 1.0 ¬0.4 ¬0.2 0 0.2 0.4
Sample bias (V) Sample bias (V)
d 10 e 2
Zigzag (rzig = 0.95)
dl/dV (V/l) (arb. units)
dl/dV (V/l) (arb. units)
8 Eg = 0.14 eV
6
Wavg = 2.3 ± 0.3 nm
4
2
0 0
¬1.0 ¬0.5 0 0.5 1.0 ¬0.2 ¬0.1 0 0.1 0.2
Sample bias (V) Sample bias (V)
f g
10 Zigzag (rzig = 0.67)
dl/dV (V/l) (arb. units)
8 Eg = 0.12 eV
Wavg = 3.3 ± 0.3 nm 1
6
2
0 0
¬1.0 ¬0.5 0 0.5 1.0 ¬0.2 ¬0.1 0 0.1 0.2
Sample bias (V) Sample bias (V)
Figure 6 | Tunnelling spectroscopy of three 20–30-nm-long, 2–3-nm-wide GNRs. a, STM topograph of the three GNRs with the edge symmetries
labelled. b–g, Normalized dI/dV spectra for a predominantly armchair (b–c) and two zigzag (d,e;f,g) GNRs presented in a. The insets of b, d and f show the
STM topograph of the nanoribbon that corresponds with the tunnelling spectroscopy measurements. The dotted regions in b, d and f correspond with the
higher resolution dI/dV(V/I) − V spectra in c, e and g. A higher concentration of zigzag edges tends to decrease the energy gap because the predominantly
zigzag-edged, 2.3-nm-wide nanoribbon exhibits a smaller energy gap (0.14 eV) than the 2.9-nm-wide armchair nanoribbon (0.38 eV). The scale bar in a
represents 2 nm.
spectra maps in Fig. 4c and d were reproducible across the surface spectra map shown in Fig. 4c. For the zigzag GQD in Fig. 4e,
of the respective sheets as shown in Fig. 4e and f, which correspond we observe an increase in the differential conductance within
with the lines labelled (1)–(3) in Fig. 4a and b, respectively. 1–2 nm from the edge of the graphene. For the mixed-edge
Figure 4e,f shows the value of the differential conductance at GQD, however, the differential conductance at 0 V does not
0 V as a function of position across both GQDs. Line (2) in increase near the graphene edges and the zero-bias, differential
Fig. 4e is recorded along the solid green line in Fig. 4a and does conductance is significantly lower compared with the zigzag
not include the region labelled by the dotted green line, which GQD. The increased conductance near the edges in Fig. 4c,e is
represents the low conductance region at the left end of the suggestive of the localized zigzag edge state, which has been
5. Chen, Z., Lin, Y., Rooks, M. J. & Avouris, P. Graphene-nanoribbon electronics. 24. Piva, P. G. et al. Field regulation of single-molecule conductivity by a charged
Physica E 40, 228–232 (2007). surface atom. Nature 435, 658–661 (2005).
6. Li, X., Wang, X., Zhang, L., Lee, S. & Dai, H. Chemically derived, ultrasmooth 25. Liu, L., Yu, J. & Lyding, J. W. Atom-resolved three-dimensional mapping of
graphene nanoribbon semiconductors. Science 319, 1229–1232 (2008). boron dopants in Si(100) by scanning tunneling microscopy. Appl. Phys. Lett.
7. Lin, Y.-M., Perebeinos, V., Chen, Z. & Avouris, P. Conductance quantization in 78, 386–388 (2001).
graphene nanoribbons. Preprint at <http://arxiv.org/abs/0805.0035> (2008). 26. Berger, C. et al. Electronic confinement and coherence in patterned epitaxial
8. Adam, S., Cho, S., Fuhrer, M. S. & Das Sarma, S. Density inhomogeneity driven graphene. Science 312, 1191–1196 (2006).
percolation metal–insulator transition and dimensional crossover in graphene 27. Geim, A. K. & Novoselov, K. S. The rise of graphene. Nature Mater. 6,
nanoribbons. Phys. Rev. Lett. 101, 046404 (2008). 183–191 (2007).
9. Son, Y.-W., Cohen, M. L. & Louie, S. G. Energy gaps in graphene nanoribbons. 28. Nakada, K., Fujita, M., Dresselhaus, G. & Dresselhaus, M. S. Edge state
Phys. Rev. Lett. 97, 216803 (2006). in graphene ribbons: Nanometer size effect and edge shape dependence.
10. Barone, V., Hod, O. & Scuseria, G. E. Electronic structure and stability of Phys. Rev. B. 54, 17954–17960 (1996).
semiconducting graphene nanoribbons. Nano Lett. 6, 2748–2754 (2006). 29. Klusek, Z. et al. Observation of local electron states on the edges of circular
11. Son, Y.-W., Cohen, M. M. & Louie, S. G. Half-metallic graphene nanoribbons. pits on hydrogen-etched graphite surface by scanning tunneling spectroscopy.
Nature 444, 347–349 (2006). Appl. Surf. Sci. 161, 508–514 (2000).
12. Fujita, M., Wakabayashi, K., Nakada, K. & Kusakabe, K. Peculiar localized state 30. Zhang, Z. Z., Chang, K. & Peeters, F. M. Tuning of energy levels and optical
at zigzag graphite edge. J. Phys. Soc. Jpn. 65, 1920–1923 (1996). properties of graphene quantum dots. Phys. Rev. B 77, 235411 (2008).
13. Sols, F., Guinea, F. & Castro Neto, A. H. Coulomb blockade in graphene 31. Datta, S. S., Strachan, D. R., Khamis, S. M. & Johnson, A. T. C. Crystallographic
nanoribbons. Phys. Rev. Lett. 99, 166803 (2007). etching of few-layer graphene. Nano Lett. 8, 1912–1915 (2008).
14. Querlioz, D. et al. Suppression of the orientation effects on bandgap in 32. Lyding, J. W., Skala, S., Hubacek, J. S., Brockenbrough, R. & Gammie, G.
graphene nanoribbons in the presence of edge disorder. Appl. Phys. Lett. 92, Variable-temperature scanning tunneling microscope. Rev. Sci. Instrum. 59,
42108 (2008). 1897–1902 (1988).
15. Gunlycke, D., Areshkin, D. A. & White, C. T. Semiconducting graphene 33. Feenstra, R. M. Tunneling spectroscopy of the (110) surface of direct-gap III–V
nanostrips with edge disorder. Appl. Phys. Lett. 90, 142104 (2007). semiconductors. Phys. Rev. B 50, 4561–4570 (1994).
16. Ritter, K. A. & Lyding, J. W. Characterization of nanometer-sized, mechanically 34. Albrecht, P. M. & Lyding, J. W. Ultrahigh-vacuum scanning tunneling
exfoliated graphene on the H-passivated Si(100) surface using scanning microscopy and spectroscopy of single-walled carbon nanotubes on
tunneling microscopy. Nanotechnology 19, 015704 (2008). hydrogen-passivated Si(100) surfaces. Appl. Phys. Lett. 83, 5029–5031 (2003).
17. Ishigami, M., Chen, J. H., Cullen, W. G., Fuhrer, M. S. & Williams, E. D.
Atomic structure of graphene on SiO2 . Nano Lett. 7, 1643–1648 (2007). Acknowledgements
18. Stolyarova, E. et al. High-resolution scanning tunneling microscopy imaging This work was supported by the Office of Naval Research under grant number
of mesoscopic graphene sheets on an insulating surface. Proc. Natl Acad. Sci. N000140610120 and by the National Science Foundation grant number NSF ECS
104, 9209–9212 (2007). 04-03489. K.A.R. acknowledges support from a NDSEG fellowship. We thank J. Koepke
19. Kobayashi, Y., Fukui, K., Enoki, T. & Kusakabe, K. Edge state on for assistance with a portion of the data collection, L. Ruppalt for providing the code for
hydrogen-terminated graphite edges investigated by scanning tunneling the normalized dI /dV calculations and P. Albrecht, P. Dollfus, D. Querlioz, A. Rockett,
microscopy. Phys. Rev. B 73, 125415 (2006). M. Sztelle and J. Weaver for helpful discussions.
20. Kobayashi, Y., Fukui, K., Enoki, T., Kusakabe, K. & Kaburagi, Y. Observation
of zigzag and armchair edges of graphite using scanning tunneling microscopy Author contributions
and spectroscopy. Phys. Rev. B 71, 193406 (2005). K.A.R and J.W.L. conceived the experiments. K.A.R. carried out the experiments,
21. Nimi, Y. et al. Scanning tunneling microscopy and spectroscopy of the analysed the data and wrote the manuscript. J.W.L. provided technical support for the
electronic local density of states of graphite surfaces near monoatomic step instrumentation, discussed the data and commented on the manuscript.
edges. Phys. Rev. B 73, 085421 (2006).
22. Rutter, G. M. et al. Scattering and interference in epitaxial graphene. Science Additional information
317, 219–222 (2007). Supplementary Information accompanies this paper on www.nature.com/naturematerials.
23. Tapasztó, L., Dobrik, G., Lambin, P. & Biró, L. P. Tailoring the atomic structure Reprints and permissions information is available online at http://npg.nature.com/
of graphene nanoribbons using scanning tunneling microscope lithography. reprintsandpermissions. Correspondence and requests for materials should be
Nature Nanotech. 3, 397–401 (2008). addressed to K.A.R.