ARTICLE

Received 3 Sep 2012 | Accepted 5 Mar 2013 | Published 9 Apr 2013 DOI: 10.1038/ncomms2695

A strong electro-optically active lead-free

ferroelectric integrated on silicon
Stefan Abel1, Thilo Stöferle1, Chiara Marchiori1, Christophe Rossel1, Marta D. Rossell2, Rolf Erni2,
Daniele Caimi1, Marilyne Sousa1, Alexei Chelnokov3, Bert J. Offrein1 & Jean Fompeyrine1

The development of silicon photonics could greatly benefit from the linear electro-optical
properties, absent in bulk silicon, of ferroelectric oxides, as a novel way to seamlessly connect
the electrical and optical domain. Of all oxides, barium titanate exhibits one of the largest
linear electro-optical coefficients, which has however not yet been explored for thin films on
silicon. Here we report on the electro-optical properties of thin barium titanate films
epitaxially grown on silicon substrates. We extract a large effective Pockels coefficient of
reff ¼ 148 pm V  1, which is five times larger than in the current standard material for electro-
optical devices, lithium niobate. We also reveal the tensor nature of the electro-optical
properties, as necessary for properly designing future devices, and furthermore unambigu-
ously demonstrate the presence of ferroelectricity. The integration of electro-optical active
films on silicon could pave the way towards power-efficient, ultra-compact integrated devices,
such as modulators, tuning elements and bistable switches.

1 IBM Research—Zurich, 8803 Rüschlikon, Switzerland. 2 Electron Microscopy Center, Empa, Swiss Federal Laboratories for Materials Science and Technology,

8600 Dübendorf, Switzerland. 3 CEA, LETI, MINATEC, DOPT, 38054 Grenoble Cedex 9, France. Correspondence and requests for materials should be
addressed to S.A. (email: [email protected]).

NATURE COMMUNICATIONS | 4:1671 | DOI: 10.1038/ncomms2695 | www.nature.com/naturecommunications 1

& 2013 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2695

T
he presence of functional properties in many perovskite 107
materials1,2 has pushed the development of integrating Out-of-plane In-plane Si ) Cubic

106 04
such crystals directly on silicon (Si) to realize a variety of (0 Tetrag.
O )
O ) BT (002
new devices3. Since the first methods to epitaxially deposit BT (001

Intensity (a.u.)
10 5
0 )/
)/
perovskite thin films on Si were established4, the growth process 0 (20
(10
104
has been carefully optimized5 and single crystalline layers were Si )
44 45 46
O
fabricated even on large-scale 800 substrates6. However, the 103 (2
20 BT (003
)
O )
0)/ BT (004
interest in this research field has mainly been motivated by the ( 3 0
0 0)/
102 (4
use of oxides in the electronic domain, such as memory
applications7, gate dielectrics in field-effect transistors8 or all- 101
oxide electronics9. Meanwhile, the steadily improving 20 40 60 80 100
performance of complementary metal oxide semiconductor 2 (°)
(CMOS) technology increases the difficulty to handle inter- or
intra-chip data communication by means of electrical wiring10
because of power constraints in combination with increased
bandwidth demands. In this context, devices working in the BTO
optical domain have emerged as a mature field of research11, and
building blocks required for a functional photonic network such
as detectors12 or modulators13 have already been integrated on Si. STO
BTO
However, today’s silicon photonics technology does not exploit
SiOx STO
the linear electro-optical (EO) properties available in oxides,
which for decades fuelled the progress in telecommunication Si Si
based on fibre optics to extremely high modulation speeds with
well-established devices14. In recent work, micro-scale EO devices Figure 1 | Structural analysis. (a) XRD diagram of 130 nm BTO on 4 nm
have been presented15. These devices were based on LNO STO on Si; the inset shows a magnification of the BTO (200)/(002) peak
(LiNbO3), which can be integrated as high-quality films obtained with out-of-plane and in-plane geometries. The vertical lines
on Si only via complex and costly layer-bonding approaches16. correspond to bulk values of tetragonal (solid) and cubic (dashed) BTO
On the other hand, a non-vanishing Pockels effect has also been (ref. 35). (b) Cross-sectional high-angle annular dark-field STEM image of
reported directly in Si, by exploiting strain effects that break the the interface region showing Si, SiOx, STO and BTO (scale bar, 5 nm), and a
crystalline symmetry17,18. However, these approaches are both corresponding low-magnification cross-section of the sample in dark-field
limited to relatively small EO coefficients (rLNOE30 pm V  1 mode (scale bar, 200 nm). The red square indicates the area from which
(ref. 14), rSiE1.7 pm V  1 (ref. 18) compared with those the high-resolution micrograph was recorded. Defects penetrate mainly
of alternative oxides (for example, barium titanate (BTO), perpendicularly to the interface. (c) Schematic of the domain structure in
rBTO41,000 pm V  1) (ref. 19), which drastically narrows the the BTO film with randomly oriented spontaneous polarization associated
range of potential applications. with the c-axis direction (white arrows in blue cuboids). a.u., arbitrary unit.
Our work demonstrates that large Pockels coefficient can be
obtained in high-quality oxide thin films epitaxially grown on Si.
This enables a flexible way to develop novel devices in silicon
photonic circuits that benefit from the established EO know-how strained films, for example by reducing their thickness, an
of fibre optics. BTO (BaTiO3) is the material of choice because of in-plane-oriented c-axis is expected in our thick films that are
its strong EO properties in bulk form, the earlier demonstration fully relaxed at the growth temperature. This orientation, which
of EO devices using BTO thin films on oxide substrates20,21, and has to be considered carefully when interpreting any EO
its compliance with possible future regulations excluding lead response, results from biaxial tensile strain after cooling, due
from products. Additionally, bulk BTO is a ferroelectric, which is to the different thermal expansion coefficients of BTO and
an appealing class of materials for bistable electrical and optical Si (ref. 24). As the template for the epitaxial growth is a cubic
applications. This ferroelectric behaviour is, as shown in our Si substrate, two equivalent orientations of the elongated
study, maintained even in thin films. tetragonal c-axis exist. By forming domains in which the c-axis
is rotated by 90° within the plane of the film, strain fields
originating from the mismatch between the cubic Si and
Results tetragonal BTO lattice are compensated. The existence of two
Film growth and structural characterization. The EO properties orthogonal domains is evidenced by the presence of the
of BTO are tightly connected with its crystalline structure, which convoluted [200]BTO and [002]BTO diffraction peaks in the
is described in a non-centrosymmetric tetragonal P4mm space grazing-incidence XRD data (Fig. 1a, inset). As no intentional
group compatible with a non-vanishing linear Pockels effect22. miscut was present in the substrate, the population ratio of these
Although the tetragonal phase is stable in bulk BTO crystals at domains is assumed to be 1:1 for symmetry reasons. The
room temperature, polycrystalline BTO thin films tend to epitaxial relationship between BTO, STO and Si is visualized in
stabilize in a cubic symmetry23, which originates from the high-resolution scanning transmission electron microscopy
formation of small grains, without Pockels effect. We suppressed (STEM) analysis (Fig. 1b, left). Low-resolution images reveal
such grain formation by depositing our 130-nm-thick BTO film that defects in the BTO develop mainly perpendicularly to the
epitaxially on a Si substrate covered by a 4-nm-thin strontium interface (Fig. 1b, right), likely because of strain compensation at
titanate (STO) buffer layer. Out-of-plane and grazing-incidence the interfaces of orthogonal domains. From both XRD and
in-plane X-ray diffraction (XRD) measurements (Fig. 1a) STEM analyses we infer that the different domains penetrate the
confirmed the epitaxial relationship and showed that the film to form a structure as shown in Fig. 1c. The arrows indicate
crystal symmetry is tetragonal with its long tetragonal c-axis the spontaneous polarization present in tetragonal BTO along
parallel to the sample surface. While an out-of-plane oriented the crystalline c-axis. This ferroelectric structure of the film is
c-axis can be obtained by growing highly compressively critical for the EO analysis presented here.

2 NATURE COMMUNICATIONS | 4:1671 | DOI: 10.1038/ncomms2695 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2695 ARTICLE

Qualitative determination of Pockels effect. The EO char- power modulation DP because in that case, DP would be
acterization was performed at the standard telecom wavelength of proportional to the absolute transmitted power, in contrast to the
l ¼ 1.55 mm, which is compatible with the silicon photonic observed EO response. The linear dependence of DP on the
technology. The EO properties of the film were determined by applied field (Fig. 2b, inset) further precludes the quadratic EO
investigating the polarization changes of a laser beam transmitted Kerr effect to be dominant and thus confirms the presence of the
between a pair of electrodes separated by a small, micrometre- linear Pockels effect in our sample.
scale gap. Applying a potential difference between two electrodes
Dependence of EO response on measurement geometry. To
results in an electric field E within the gap, modifying the
further explore the tensor nature of the Pockels effect19, the
refractive index n(E) of BTO (ref. 21)
dependence of d on the angle between the applied electric field,
1 incident polarization, and crystallographic orientation has to be
nðEÞ ¼ n0  rn30 E ð1Þ considered carefully. The derivation of a field-normalized angle d0
2
from data sets, as shown in Fig. 2, is described in the Methods
section. In general, applying an electric field to an anisotropic
In this simplified equation that neglects any tensor material with a linear EO effect results in a change of the
properties19, r is the Pockels coefficient and n0 the refractive birefringence Dn by D(Dn) and a rotation of the optical axis by
index when no electric field is applied. To gain sensitivity by using Dg (Fig. 3a). In our measurement geometry, both effects cause a
a lock-in detection scheme, the applied voltage is modulated by rotation of the polarization of the transmitted beam, but the
Va.c. ¼ 3 V (peak-to-peak) at a frequency of f ¼ 17.3 kHz around a magnitude of d0 depends on the incident polarization: If Dg is
d.c. offset voltage Voff ¼ 25 V, which is sufficiently high to align dominant, d0 is maximal when the incident polarization is parallel
the ferroelectric domains. The modification of the refractive index to the optical axis of the crystal (yi ¼ 0°), which is experimentally
due to the applied field leads to a rotation of the linear observed when the applied electric field E is oriented along the
polarization of the transmitted beam by an angle d (Fig. 2a). After [101]BTO direction (Fig. 3b). If D(Dn) is dominant, on the other
passing through an analyzer oriented at an angle ya relative to the hand, d0 is maximal for |yi|B45°. We also observe this regime in
polarization of the incident beam, the transmitted power our measurements, when the E is oriented along the [001]BTO/
PBcos2(ya  d) (ref. 19) depends on the induced rotation of [100]BTO axes. By modelling the EO response of BTO for a single
the polarization. For small angles d, the power variations are domain25 (Fig. 3b, inset), the rotation of its optical axis is
expected to be proportional to the derivative dP/dya (ref. 25). The predicted to vanish when E is parallel to [001]BTO (E||[001]BTO),
measured response of our sample is in excellent agreement with whereas a change in the birefringence remains. In contrast, a
such a dependence (Fig. 2b), confirming the existence of an EO strongly enhanced rotation of the optical axis is expected for
effect. We can rule out a varying absorption as the origin of the E||[101]BTO. Both features of the model agree well with our
experimental results. The dependence of the EO response on the
orientation of the electrodes also precludes any polarization-
Analyzer Detector dependent absorption effect from the edges of the metallic pads.
λ/4 Plate Such effect would be independent of the orientation of the
a
 electrodes with respect to the BTO crystalline axis, which is in
Incident Sample
beam Electrodes contrast to the data shown in Fig. 3.
E
i
E
Ferroelectric properties of the BTO film. To quantify the ele-
Voltmeter
ments of the Pockels tensor, one must take into account not only a
a.c. + d.c.
voltage source single domain but the overall ferroelectric domain structure of the
Trigger
BTO film (Fig. 1c): because of the anti-centrosymmetry of the
Lock-in amplifier
D(Dn) and Dg functions (Fig. 3b, inset), the induced rotation of
the polarization by d0 will be reversed for a domain that is rotated
Applied voltage Va.c. (V) by 180° and exposed to the identical electric field. The EO
response of two anti-parallel domains will thus cancel out. How-
|ΔP | /P (10–4)

0 4 8
60 P Fit 1.5 ever, anti-parallel domain pairs can be eliminated by flipping their
ΔP
Power P (10–6 W)

40 0.0 polarization when the applied electric field component E[001]

along the crystallographic [001]BTO axis is above the coercive field
20 Ec. Hence, a ferroelectric hysteresis loop would be expected in the
0 optical measurements of d0 on varying the offset of the applied
x 2x10
5 electric field Eoff. This is indeed observed experimentally (Fig. 4).
–20
–45 0 45 90 135 180 225
As expected, the random domain orientation after heating
a (°) the sample above the Curie temperature of BTO (Tc ¼ 124 °C
(ref. 26)) initially shows no EO response at Eoff ¼ 0. By increasing
Figure 2 | Principle of EO characterization. (a) Schematics of Eoff, a strong enhancement in d0 is visible, resulting from the
measurement setup indicating the variation of the polarization states when switching of the ferroelectric domains, which eliminates anti-
an electric field (green) is applied by an a.c./d.c. voltage source. jE is the parallel domains. The hysteresis loop monitored during the sweep
angle between the BTO [h00] and [00h] crystallographic axes and the of Eoff (Fig. 4) is a clear signature of the ferroelectric nature of the
applied electric field E. The small rotation d of the polarization results in a film. Once poled, the ferroelectric domain structure remains stable
modulation of the transmitted power at the detector, which is recorded with for several days at least, as can be seen from the small relaxation of
a lock-in amplifier. (b) Transmitted power P and variation in power DP due the remanent EO response (Fig. 4, inset). From the shape of the
to the EO effect as a function of the analyzer position ya, taken for yi ¼ 0° hysteresis, a coercive field of Ec ¼ 5  105 V m  1 is extracted,
across a 17-mm-wide gap oriented at jE ¼ 45° with an applied voltage offset which is an order of magnitude larger than in bulk BTO. Such an
of Voff ¼ 25 V. Inset in b shows the normalized linear power change |DP|/P increase is typically observed for ferroelectric thin films because of
at ya ¼ 45° and Voff ¼ 25 V as a function of the peak-to-peak amplitude Va.c.. domain pinning and finite depolarization fields27.

NATURE COMMUNICATIONS | 4:1671 | DOI: 10.1038/ncomms2695 | www.nature.com/naturecommunications 3

& 2013 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2695

No field Applied field 1.5×10–8 8

(°/(V m–1))
′ ×10–9
[001]BTO
Δnz–Δnx Δγ 1.0×10–8 4
nz n’z
[101]BTO =Δ(Δn) 0

′ (°/(V m–1))
5.0×10–9 0 40 80
Optical axis

E nz Time (h)
nx E
n’x 0.0
[100]BTO Direction
nx –5.0×10–9
Up
Down
<0 — >0 [001] –1.0×10–8
30 E || [101]BTO 0° E Down (initial)
Δ(Δn)
Δ [101] –1.5×10–8
E || [001]BTO
20 –5.0×106 –2.5×106 0.0 2.5×106 5.0×106
270° 90°
′ (10–9 °/(V m–1))

Fit Offset field Eoff (V m–1)

[100]
10 x10
180° Figure 4 | Optical response as a function of the offset field Eoff. The data
2·′max

0 set is collected in three successive loops measured at 25 °C after heating

the film above its bulk Curie temperature (green) and sweeping Eoff up
–10 (red) and down (blue) again. The finite slope at large |Eoff| values results
from nonlinear EO effects. The inset shows the relaxation of the remanent
–20 [100] [100] [100]
[001] [101] [001] [101] [001] d0 value at room temperature.
–135 –90 –45 0 45 90 135
Incident polarization i (°)

domain population. Therefore, the modulation of the electric field

by Ea.c. ¼ ±0.2  106 V m  1 ) Eoff (Methods section) for the
lock-in detection induces only a negligible error in the
determination of d0 and thus of the EO constants. More
15 generally, if domain switching contributed to the EO response, a
′max 30
′max (10–9 °/(V m–1))

i(′max) 15
strong enhancement of d0 for the regime of |Eoff þ Ea.c.|oEc would
i(′max) (°)

10 be expected, because the relative change of the domain population

0
is the strongest for small electric offset fields. This behaviour
–15 would also be expected for the unpoled sample and is contrary to
5
[100] [100] –30 our observations of a vanishing EO signal in that regime (Fig. 4),
[001] [101] [001]
0 –45 thus excluding domain switching to be a major contribution to d0 .
0 20 40 60 80
E (°)

Figure 3 | Experimental and simulated EO response. (a) Refractive index Discussion

ellipse in BTO, showing the refractive index nx and nz along different We can also utilize the hysteresis behaviour for the quantitative
crystalline directions (left). Applying an electric field E changes the determination of the EO constants: indeed, because the offset in
birefringence D(Dn) (centre) and rotates the optical axis (right) depending the electric field E applied during the measurements (Fig. 3b) was
on the orientation of E (not to scale). (b) Field-normalized d0 as function of significantly above the coercive field, we assume the ferroelectric
the angle of incident polarization for two orientations of the applied field E. domains to be aligned to the electric field component E[001].
The dashed vertical lines correspond to the main BTO crystalline directions. Domains orthogonal to E experience no component along their
The inset shows the simulated birefringence change D(Dn) and the angle of long [001] axis, thus keep their anti-parallel configuration (Fig. 3c)
the optical axis Dg in BTO in arbitrary units. (c) Schematics indicating the without any net contribution to the EO response. Taking
different ferroelectric domain configurations for various orientations of E these domain structures into account, we can calculate the
(green), with domains oriented towards the field (red) and randomly individual components of the reduced 6  3 Pockels tensor
oriented ones (blue). (d) Maximum rotation angle d0 max and corresponding r (ref. 19), which reduces to only three different elements for
angle of incident polarization yi(d0 max) for various values of jE. The error the P4mm symmetry class of tetragonal BTO (ref. 22). For
bars show the standard deviation from several measurements on different E||[001]BTO, the optical properties are determined by r13 and r33.
electrode pads. The simulated curves (dashed lines) are based on the Considering dBrc ¼ r33 (neo/no)3r13, we obtain rc ¼ 30 pm V  1 as
Pockels tensor that was experimentally determined (details in the c-axis Pockels coefficient. This rc value19 is more important for
Supplementary Note 2). the performance of EO devices than the individual components r13
and r33 (ref. 20), which are correlated and cannot be
independently determined in our measurement geometry. The
third tensor element, r42, describes the rotation of the optical axis
The hysteresis in the optical signal qualitatively excludes any as observed at E||[101]BTO. A detailed analysis allows us to extract
influences from the Si-substrate on the EO response of the sample. r42 ¼ 105 pm V  1 (see Supplementary Note 1). For comparison
As recently shown, non-vanishing Pockels coefficients are induced with previous work21,30, we calculated an effective value
in Silicon if strain-gradients are present18,28. However, Si remains reff ¼ 148 pm V  1 by neglecting the tensor nature of r and
non-ferroelectric and thus no effects of poling are visible, in simply considering the field-induced phase shift between two
clear contrast to our observations. From the measurements orthogonal components of the transmitted light, as described in
shown in Fig. 4 we can furthermore exclude that ferroelectric the Methods section. Our results represent a major improvement
domain switching29 has a significant contribution on the EO compared with the limited number of published data on EO
response presented in Figs 2 and 3. These measurements were properties of lead-free oxides integrated on silicon31–33.
performed in the saturation regime of the hysteresis loop The Pockels tensor and our model with multiple domains are
(|Eoff|43  106 V m  1, see Methods section). The saturation validated by systematically varying the orientation of the applied
indicates that small changes in the applied field hardly change the electric field and comparing the experimental results with

4 NATURE COMMUNICATIONS | 4:1671 | DOI: 10.1038/ncomms2695 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2695 ARTICLE

simulations using the previously determined values for r (Fig. 3d, Transmission electron microscopy. Cross-section samples for transmission
more details in the Supplementary Note 2). When the electric electron microscopy were cut parallel to the (110) planes of the Si substrate and
mechanically polished to a thickness of about 20 mm, followed by final ion milling
field is oriented close to the [101]BTO orientation, the EO under grazing incidence until electron transparency was reached. High-angle
response originates from a rotation of the optical axis (yiB0°), annular dark-field STEM was carried out using a conventional JEOL 2200FS TEM/
and the magnitude of the response approaches a maximum. STEM microscope operated at 200 kV. The probe semi-convergence angle was set
These characteristics are expected for a large r42 coefficient when to 10.8 mrad and the inner semi-detection angle of the annular dark-field detector
the electric field is not parallel to the [001]BTO direction (Fig. 3b, was 100 mrad.
inset). The situation is reversed when E is aligned to the [001]BTO
and [100]BTO axes: The EO response is minimal and determined EO measurements. For the EO characterization (schematics in Fig. 2a, for more
by the change in birefringence (|yi|B45°) rather than by the details see Supplementary Fig. S1), a single-mode diode laser from New Focus
operating at lE1.55 mm was first polarized with a Glan-laser prism (extinction
rotation of the optical axis, in excellent agreement with the ratio o5  10  6). A successive half-wave plate mounted on a motorized stage
simulation. Finally, the identical EO response when rotating the (Newport URS50BPP) was used to set the incident polarization, before focusing the
electric field by 90° matches the picture of equally distributed beam with a spot size of B15 mm into the gap of two electrodes. Owing to the
orthogonal domains inferred from the structural characterization birefringence of BTO, the transmitted light is generally slightly elliptically polar-
ized. A quarter-wave plate after the sample and a second Glan-laser prism, both
(Fig. 1c). Thus, the Pockels tensor and domain structure derived mounted on motorized stages, were iteratively rotated to minimize the power of the
in our study are consistent with the experimentally observed EO transmitted light on an InGaAs detector. For the condition of minimal transmis-
behaviour for various orientations of the applied electric field. sion, the light polarization after the quarter-wave plate is linear. After applying a
In conclusion, we have shown the direct integration of a voltage offset of Voff ¼ 40 V for B1 min to align the ferroelectric domains, a
ferroelectric BTO film on silicon exhibiting a strong linear EO sinusoidal, alternating voltage of Va.c. ¼ 3.0 V (peak-to-peak amplitude) at
f ¼ 17.3 kHz with Voff ¼ 25 V was applied during the measurements. No electrical
effect with an effective Pockels coefficient of reff ¼ 148 pm V  1. breakdown was observed for these voltages (see I–V curve, Supplementary Fig. S2).
This value exceeds previous data reported for integrated LNO by The transmitted power P and the power variations DP were recorded using a lock-
at least a factor of five, and for strained Si by a factor of 100. By in amplifier while rotating the analyzer ya in small steps. The rotation d was
combining EO measurements with EO simulations, we developed determined by fitting the signal DP/P using DP ¼ d  dP/dya þ A P. In this equa-
tion, PBcos2(ya  d)Ecos2ya corresponds to the transmission between two
a method to investigate the tensor nature of the Pockels effect of polarizers19, and A to background contributions, which are proportional to the
multi-domain thin films, which is essential for properly designing transmitted power and caused by interference effects between the domains. With
future optical devices. The strong EO response renders the films this fitting procedure, a rotation in the polarization down to d ¼ 10  5 ° can be
perfectly suitable as base material for novel applications in the detected, which is significantly better than the mechanical resolution of the rotation
stages. d was field-normalized by the amplitude of the applied a.c.-field Ea.c., d0 ¼ d/
domain of silicon photonics such as tuning or bistable switching. Ea.c. The magnitude of the applied electric field E is calculated by approximating the
The latter application is especially appealing because of the long field in the gap of width d by E ¼ V d  1. The Pockels coefficient rc is determined at
remanence demonstrated in the films, which also makes BTO jE ¼ 0° by rc ¼ Gl/(p n3 Ea.c. d n) (ref. 30), where G ¼ 2  dmax is the induced
layers attractive for transistors with ferroelectric gates and similar phase shift in the sample and n ¼ 0.5 is the relative volume fraction of the domains
parallel to the electric field, which contribute to the EO response. dmax corresponds
application in the electronic domain. Besides the new opportu- to the maximum rotation angle when varying the incident polarization (Fig. 3b).
nities for engineering photonic circuits, the integration of The effective Pockels coefficient reff is calculated at jE ¼ 45° in the same way using
functional oxides such as BTO also opens exciting challenges n ¼ 1. Before measuring the hysteresis loops, the sample was heated to 250 °C for
for material scientists to tailor the EO properties of those oxides B10 min to ensure a depolarized domain state in the sample, and then cooled
by strain engineering or fabrication of superlattice structures34, down to room temperature. Each data point during the hysteresis measurement
corresponds to a sweep in ya, taking B5 min, which yields a typical duration of
which could ultimately lead to another boost of their EO B5 h for the measurement of one loop.
properties.
EO simulations. The calculation is based on an algorithm described in ref. 25 using
Methods the ordinary and extraordinary refractive indices of bulk BTO (no ¼ 2.268 and
Sample preparation. The sample was prepared in a molecular beam epitaxy neo ¼ 2.297 at l ¼ 1,550 nm). The EO properties of BTO thin films are typically
chamber with a base pressure of 5  10  10 mbar. A 4-nm-thick STO (SrTiO3) film reduced by an order of magnitude20,21 compared with bulk values (r13 ¼ 8, r33 ¼ 80,
was deposited on a high-resistive, 500-mm-thick Si(001) substrate using a similar r42 ¼ 1,000 pm V  1 at l ¼ 1,500 nm) (ref. 22). Therefore, the bulk values were
process as described in ref. 8. It serves as epitaxial seed layer for the subsequent scaled by a factor of 0.1 in our simulations, resulting in a good qualitative
130-nm-thick BTO layer, which was grown by alternate exposure to Ba and Ti agreement of the simulated EO response with our experimental results.
fluxes in atomic oxygen at a pressure of 10  5 mbar and a temperature of 650 °C.
The Ba and Ti fluxes were calibrated with a quartz microbalance. An interfacial
SiOx layer can develop during the BTO growth step because of the combination of
References
high temperature and high oxygen pressure, but does not disrupt the epitaxial 1. Warusawithana, M. P. et al. A ferroelectric oxide made directly on silicon.
relationship between Si, STO and BTO as this relationship is established during the Science 324, 367–370 (2009).
initial growth of STO on Si. After the growth, the backside of the wafer was 2. Baek, S. H. et al. Giant piezoelectricity on Si for hyperactive MEMS. Science
polished for optical transmission measurements. The film thickness was 334, 958–961 (2011).
determined by spectroscopic ellipsometry (Woollam VASE). Electrode pairs of 3. Reiner, J. W. et al. Crystalline oxides on silicon. Adv. Mater. 22, 2919–2938
300 nm Pt/10 nm Ti with varying orientation between the BTO crystallographic (2010).
axes and the electrode gap (0°–90°) with an accuracy of ±1° were fabricated on 4. McKee, R. A. et al. Molecular beam epitaxy growth of epitaxial barium silicide,
the samples using standard optical lithography techniques. Unless stated barium oxide, and barium titanate on silicon. Appl. Phys. Lett. 59, 782–784
differently, the optical measurements were performed on devices with electrodes (1991).
having a gap size of B7 mm. 5. Norga, G. J. et al. Phase of reflection high-energy electron diffraction
oscillations during (Ba,Sr)O epitaxy on Si(100): a marker of Sr barrier integrity.
Appl. Phys. Lett. 87, 262905 (2005).
X-ray analysis. XRD measurements were carried out on a Bruker Discovery D8 6. Gu, X. et al. Commercial molecular beam epitaxy production of high quality
system equipped with a rotating anode generator. A symmetric Ge[220] 4-bounce SrTiO3 on large diameter Si substrates. J. Vac. Sci. Technol. B 27, 1195 (2009).
monochromator was used on the primary beam for the out-of-plane measure- 7. Niu, G. et al. Epitaxy of BaTiO3 thin film on Si(001) using a SrTiO3 buffer layer
ments, together with a slit on the diffracted beam. To have a sufficiently high for non-volatile memory application. Microelectron. Eng. 88, 1232–1235 (2011).
signal-to-noise ratio, grazing-incidence in-plane measurements were performed 8. Marchiori, C. et al. Thermal stability of the SrTiO3/(Ba,Sr)O stacks epitaxially
without monochromator, and using Soller slits on the detector side. By neglecting a grown on Si. Appl. Phys. Lett. 88, 072913 (2006).
possible orthorhombic distortion that would be within the error bars of the XRD 9. Park, J. et al. Creation of a two-dimensional electron gas at an oxide interface
data, the lattice parameters of the tetragonal unit cell were determined to be on silicon. Nat. Commun. 1, 94 (2010).
a ¼ 3.997±0.005 Å and c ¼ 4.032±0.005 Å using the out-of-plane diffraction 10. Miller, D. A. B. Device requirements for optical interconnects to silicon chips.
peaks and by deconvoluting the peaks measured in grazing-incidence in-plane Proc. IEEE 87, 1166–1185 (2009).
geometry. 11. Simply silicon (Editorial). Nat. Photon 4, 491 (2010).

NATURE COMMUNICATIONS | 4:1671 | DOI: 10.1038/ncomms2695 | www.nature.com/naturecommunications 5

& 2013 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms2695

12. Michel, J. et al. High-performance Ge-on-Si photodetectors. Nat. Photon 4, 30. Reitze, D. H. et al. Electro-optic properties of single crystalline ferroelectric thin
527–534 (2010). films. Appl. Phys. Lett. 63, 596–598 (1993).
13. Reed, G. T. et al. Silicon optical modulators. Nat. Photon 4, 518–526 (2010). 31. Rabiei, P. et al. Optical and electro-optical properties of submicrometer lithium
14. Wooten, E. et al. A review of lithium niobate modulators for fiber-optic niobate slab waveguides prepared by crystal ion slicing and wafer bonding.
communications systems. IEEE J. Sel. Topics Quantum Electron 6, 69–82 Appl. Phys. Lett. 85, 4603–4605 (2004).
(2000). 32. Akazawa, H. et al. Electro-optic properties of c-axis oriented LiNbO3 films
15. Guarino, A. et al. Electro-optically tunable microring resonators in lithium grown on Si(100) substrate. Mater. Sci. Eng. B 120, 50–54 (2005).
niobate. Nat. Photon 1, 407–410 (2007). 33. Lee, Y. S. et al. Hybrid Si-LiNbO3 microring electro-optically tunable resonators
16. Chen, L. et al. Compact electric field sensors based on indirect bonding of for active photonic devices. Opt. Lett. 36, 1119–1121 (2011).
lithium niobate to silicon microrings. Opt. Express 20, 4032–4038 (2012). 34. Lu, Y. et al. Enhanced dielectric and electro-optic effects in relaxor oxide
17. Jacobsen, R. S. et al. Strained silicon as a new electro-optic material. Nature heterostructured superlattices. J. Phys. D: Appl. Phys 37, 2432 (2004).
441, 199–202 (2006). 35. Swanson, H. E. et al. Natl. Bur. Stand. (U.S.) 3, 45 (1954).
18. Chmielak, B. et al. Pockels effect based fully integrated, strained silicon electro-
optic modulator. Opt. Express 19, 17212–17219 (2011).
19. Yariv, A. & Yeh, P. Optical Waves in Crystals (John Wiley & Sons, 1984). Acknowledgements
20. Petraru, A. et al. Ferroelectric BaTiO3 thin-film optical waveguide modulators. We gratefully acknowledge H. Siegwart for technical support, U. Drechsler for sample
Appl. Phys. Lett. 81, 1375–1377 (2002). preparation, and F. Mohn for useful discussions.
21. Wessels, B. W. Ferroelectric epitaxial thin films for integrated optics. Annu.
Rev. Mater. Res. 37, 659–679 (2007). Author contributions
22. Bernasconi, P. et al. Temperature dependence and dispersion of electro-optic S.A. prepared the samples with support from J.F. and D.C. D.C. performed the litho-
and elasto-optic effect in perovskite crystals. J. Appl. Phys. 78, 2651–2658 graphy steps. S.A., C.M. and M.S. carried out the XRD, and M.D.R. and R.E. performed
(1995). the TEM analysis. S.A. designed and performed the EO characterization, supported by
23. Jang, J. W. et al. Thickness dependence of room temperature permittivity of T.S., C.R. and A.C. Simulations and data analysis were done by S.A. J.F., B.J.O., and A.C.
polycrystalline BaTiO3 thin films by radio-frequency magnetron sputtering. provided conceptual advice. All authors discussed the results and commented on the
J. Appl. Phys. 81, 6322–6327 (1997). manuscript, which was written by S.A.
24. Vaithyanathan, V. et al. c-axis oriented epitaxial BaTiO3 films on (001) Si.
J. Appl. Phys. 100, 024108–024109 (2006).
25. Abel, S. et al. Electro-optical properties of barium titanate films epitaxially Additional information
grown on silicon. Proc. SPIE 8263, 82630Y (2012). Supplementary Information accompanies this paper at http://www.nature.com/
26. Sakayori, K.-i. et al. Curie temperature of BaTiO3. Jpn J. Appl. Phys. 34, naturecommunications
5443–5445 (1995).
27. Rabe, K. M., Ahn, C. H. & Triscone, J.-M. Physics of Ferroelectrics (Springer- Competing financial interests: The authors declare no competing financial interests.
Verlag, 2007). Reprints and permission information is available online at http://npg.nature.com/
28. Cazzanelli, M. et al. Second-harmonic generation in silicon waveguides strained reprintsandpermissions/
by silicon nitride. Nat. Mater. 11, 148–154 (2012).
29. Liu, Z. et al. Dynamic response of polydomain ferroelectric barium titanate How to cite this article: Abel, S. et al. A strong electro-optically active lead-free ferro-
epitaxial thin films and its field dependence. J. Appl. Phys. 104, 064115 (2008). electric integrated on silicon. Nat. Commun. 4:1671 doi:10.1038/ncomms2695 (2013).

6 NATURE COMMUNICATIONS | 4:1671 | DOI: 10.1038/ncomms2695 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.