Functional plasmonic nano-circuits with low
insertion and propagation losses
Arian Kriesch,
∗
,
†
,
‡
Stanley P. Burgos,
‡
Daniel Ploss,
†
Hannes Pfeifer,
†
Harry A. Atwater,
‡
and Ulf Peschel
†
Institute of Optics, Information and Photonics, Erlangen Graduate School in Advanced Optical
Technologies, Friedrich-Alexander-University Erlangen-Nuremberg (FAU) and Max Planck
Institute for the Science of Light (MPL), 91058 Erlangen, Germany, and Kavli Nanoscience
Institute, California Institute of Technology, Pasadena, California 91125, USA
E-mail: arian.kriesch@mpl.mpg.de
Abstract
We experimentally demonstrate plasmonic nano-circuits operating as sub-diffraction directional
couplers optically excited with high efficiency from free-space using optical Yagi-Uda style antennas at
λ
0
=
1550 nm. The optical Yagi-Uda style antennas are designed to feed channel plasmon waveguides
with high efficiency (45% in coupling, 60% total emission), narrow angular directivity (
<
40
◦
) and low
insertion loss. SPP channel waveguides exhibit propagation lengths as large as 34 μm with adiabatically
tuned confinement, and are integrated with ultra-compact (5
×
10 μm
2
), highly dispersive directional
couplers, which enable 30 dB discrimination over
∆
λ
=
200 nm with only 0
.
3 dB device loss.
Surface plasmon polariton (SPP) waveguides are
uniquely advantaged by their high confinement, al-
lowing for subwavelength integration. This is a
requirement for integrating optics with a footprint
size that is comparable with electronic circuits -
thus enabling plasmonic-electronic hybrid integra-
tion, a path that has been repeatedly highlighted as
a future key application of plasmonics.
1–4
However, high confinement in plasmonics usu-
ally increases loss due to the larger field over-
laps with the metal. The second major obstacle
to deep subwavelength plasmonics is high inser-
tion loss due to the limited modal field overlap
of less-confined waveguide schemes like Si inte-
grated photonics
5,6
or optical fibers,
7
thus intrinsi-
cally limiting the performance of hybrid dielectric-
plasmonic circuits.
8
Here, we illustrate how in-circuit-loss can be
?
To whom correspondence should be addressed
†
FAU and MPL
‡
Caltech
mitigated by restricting strong optical confinement
only to components where it is absolutely essen-
tial (Fig. 1a), and how insertion-loss can be ad-
dressed by coupling light into plasmonic nano-
circuits via impedance-matched optimized Yagi-
Uda
9
style nano-antennas (Fig. 1b,c). Using this
platform, we experimentally demonstrate optical
directional couplers
10,11
integrated on a microme-
ter scale that show unusually strong spectral dis-
persion, a key prerequisite for integrated wave-
length division multiplexers.
To implement these device concepts, we use SPP
channel waveguides,
12,13
which offer maximum
confinement
14
in a narrow rectangular gap etched
a few hundreds nanometers into a metal film. We
note that this waveguide geometry does not suffer
from optical mode cutoff when scaled down.
By filling the air gap (
n
=
1) in the plasmonic
waveguide with the substrate material silica (
n
≈
1
.
45) we make the modal field distribution more
symmetric (Fig. 1d), therefore eliminating an up-
1
arXiv:1308.3261v1 [physics.optics] 14 Aug 2013
per modal cut-off that otherwise prohibits larger
waveguide channel widths (
>
80 nm for usual Au
channel waveguides), therefore limiting the prop-
agation length to below 10 μm.
Wherever possible, the connections from one
functional plasmonic unit to the next must be
bridged with low loss plasmonic waveguides. As
the waveguide mode is basically maintained for
different gap widths, easy to fabricate adiabatic
waveguide tapers can form the transition from
highly confining to low loss sections (Fig. 1e).
We experimentally demonstrate that the inves-
tigated circuits achieve a propagation length of
L
1
/
e
=
34 μm, while they are still subwavelength
with a wave-guide gap width of 300 nm and that
they have an effective refractive index of
n
eff
≈
1
.
54 at
λ
0
=
1550 nm with low dispersion.
To reduce insertion loss, each presented nano-
plasmonic circuit utilizes at least two connected
Yagi-Uda antennas to enhance coupling efficiency
from a focused laser beam and to achieve nar-
row directionality (Fig. 1c). We measured the an-
tenna and waveguide properties spectrally and de-
rive their fundamental properties: efficiency (45 %
in coupling, 60 % total emission), directionality
(
<
40
◦
) and spectral dispersion.
For our experiments, embedded SPP nano-
circuits (Figs. 1a and 6b) were fabricated in a
two-step process. A 220 nm thin Au film on a
160
−
μm-thin, polished quartz glass substrate was
patterned by lift-off with a 100 kV electron beam
lithography system. The designed minimum fea-
ture size at the antenna gap (Fig. 1f) was 80 nm.
Subsequently, flowable spin-on glass (Filmtron-
ics 400F) was spin-coated and baked to form a di-
electric cladding layer. Scanning electron micro-
scope images of device cross-sections, cut with
a focused ion beam (FIB) at different positions
show that the cladding layer fills all gaps, covers
the whole structure without cavities (Fig. 1g-i) and
forms a smooth surface that is
≈
320 nm higher
than the underlying structures.
The fabricated circuits consistently deviate less
than 15
◦
from perfectly rectangular sidewall an-
gles. To investigate the dependence of the optical
properties of each circuit component on design pa-
rameters, arrays of circuits were fabricated, while
systematically varying waveguide length and bend
radii (Fig. 1a).
400 nm
400 nm
400 nm
a
c
d
e
f
h
i
1 μm
20 μm
1 μm
b
Au
Au
SiO
2
SiO
2
Au
Au
SiO
2
SiO
2
contrast C
contrast C
SiO
2
SiO
2
contrast C
g
Figure 1: (a) SEM of an array of multicom-
ponent plasmonic nano-circuits. (b, c) Optical
loaded Yagi antennas are optimized to couple a
highly focused, linearly polarized laser beam into
waveguides (d) that are covered with a cladding
layer of spin-on-glass. (e) Adiabatic tapers enable
low-loss transitions between low-loss and high-
confinement circuit sections. (f) The circuits are
fabricated of Au with e-beam lithography. Cross-
section cuts with a focused ion beam device (FIB),
(g) through the 300 nm wide waveguide, (h) the
antenna connector and (i) the antenna with a min-
imum feature size of only 80 nm display the sand-
wich of a substrate layer of silica, the 220 nm thick
metal structures and the 320 nm thick cladding
layer of spin-on-glass, which completely fills the
metal gaps.
2
Spectral transmission measurements of each cir-
cuit were taken on an optical far-field excitation
and imaging setup
15
with a 4 W supercontinuum
light source, spectrally filtered to
λ
=
1200
−
1850 nm at 5 nm FWHM with an acousto opti-
cal tunable filter (AOTF) (for sketch and details
see supplementary information). A collimated and
linearly polarized laser beam was focused by an
objective (
NA
=
0
.
9 dry respectively 1
.
3 oil im-
mersion) to a diffraction-limited spot (
<
2 μm).
The sample was positioned with 5 nm resolution
on a stabilized piezo stage (PI) to couple maxi-
mum power into the optical antennas (Fig. 1b).
The whole circuit was simultaneously imaged
with 150
×
or 300
×
magnification through the ex-
citation objective, a polarization filter and NIR
imaging optics on an infrared InGaAs CCD cam-
era (Xenics XS, 320
×
256 pixels). With this setup,
the linearly polarized emission from output anten-
nas, turned by 90
◦
with respect to the input antenna
(Fig. 1a), was imaged with enhanced dynamic
range by suppressing the back-reflection from the
incident beam with a ratio of up to 10
,
000.
Absolute circuit transmission values were deter-
mined by normalizing with the spectral reflectiv-
ity of a silica-air interface on the same sample (see
supplementary information for details). With addi-
tional Fourier imaging, the polarization dependent
angular emission pattern of the antennas was ana-
lyzed.
For any application, including probing the cir-
cuit properties presented here, light has to be
coupled into and out of the guided mode of the
plasmonic waveguide with highest possible power
conversion efficiency.
2
Plasmonic antennas are
best suited to fulfill this task for exciting channel
plasmons from freespace.
16,17
The concept of optical antennas has already
been successfully transferred from well-proven ra-
diofrequency technology to the optical and near-
infrared with a resulting antenna size in the μm
range.
18–20
Here we apply the macroscopic de-
sign concept of Yagi-Uda antennas
9
in the near
infrared
21–25
and connect them to feed the plas-
monic circuits.
For the investigated antennas, the feed element,
an approximately 1 μm long, 100 nm wide, split Au
rod, is centrally illuminated with a focused laser
beam (
<
2 μm) that is polarized parallel to the
a
Waveguide
T
wg
(λ,L)
Yagi antenna in
T
ant
(λ),A
ant
(λ)
Spectral total system transmission T
tot
(λ)
R
ant
(λ)
Yagi antenna out
T
ant
(λ),A
ant
(λ)
R
ant
(λ)
b
Wavelength in
μ
m
Min. antenna efficiency
T
ant
(%)
1.2
1.4
1.6
1.8
0
10
20
30
40
50
60
+σ
-σ
+σ
-σ
+σ
-σ
from air
from substrate
total efficiency
Figure 2: (a) The plasmonic circuits are analyzed
as a sequence of black boxes, each with a char-
acteristic, spectral transmission
T
(
λ
)
, absorption
loss
A
(
λ
)
and reflection coefficient
R
(
λ
)
towards
the waveguide.
(b) Experimentally determined
lower bound for the spectral efficiency of the Yagi
antennas
T
ant
(
λ
)
reaching a spectral peak value of
60 % of intensity to couple the bound mode inside
the waveguide out into air (15 %) and silica (45 %),
respectively from a focused beam into the waveg-
uide. The antenna efficiency was determined by
fitting measured data with a Lorentzian line shape
and separating loss effects. The shaded areas indi-
cate
±
σ
error bands.
3
θ
20°
30°
40°
50°
60°
φ
20°
30°
40°
50°
60°
θ
φ
Measured angular emitted int. into substrate
20°
30°
40°
50°
60°
0
0.5
1
θ
FDTD, far field transformed |E|
2
into substrate
0
0.5
1
FDTD, far field transformed |E|
2
into air
0
0.5
1
Measured angular emitted int. into air
a to air
b
c
d
to substrate
e
f
20°
30°
40°
50°
60°
θ
0
0.5
1
φ
θ
φ
φ
θ
φ
Figure 3: A narrow and strongly linear polarized angular emission directionality of the Yagi antenna at
λ
=
1550 nm (a,b,c) into air and (d,e,f) into the silica substrate was measured experimentally by Fourier
plane imaging of the antenna emission during excitation with (b) an objective of
NA
=
0
.
9 and (e) an
immersion objective of
NA
=
1
.
3. The maximum azimuthal emission cone of 30
◦
is in accordance with
the 3D FDTD simulation of the same structure into (c) air and (f) silica.
antenna arms (Fig. 1b,c). This element is driven
at its resonance frequency and feeds the 300 nm
wide SPP channel waveguide. An additional non-
resonant Yagi-reflector element is placed in a dis-
tance of 600 nm to the feed element to construc-
tively reflect the electromagnetic field back into
the waveguide, thus enhancing the in-coupling ef-
ficiency.
Our whole design was developed using an itera-
tive particle-swarm-optimization algorithm based
on full 3D FDTD simulations while taking into
account the limitations of fabrication and real op-
tical material parameters as determined with el-
lipsometry. As it was found that in-coupling ef-
ficiency increases with decreasing size of the an-
tenna gap21, we limited the antenna gap dimen-
sion to 80 nm based on fabrication limitations.
A taper was added to connect the antenna with
a 300-nm-wide low-loss waveguide to minimize
impedance mismatch and back reflections.
To simplify the experimental analysis of our
circuits we attributed lumped properties to each
element (Fig. 2a).
1
The antennas were assigned
a characteristic spectral transmission
T
ant
(
λ
) =
P
wg
(
λ
)
/
P
freespace
(
λ
)
that represents the ratio of
power, which is converted into the waveguide
mode for a specific wavelength, a spectral absorp-
tion
A
ant
(
λ
)
due to Ohmic losses and a reflectiv-
ity
R
ant
(
λ
)
at the antenna-waveguide connection
caused by residual impedance mismatch1,17.
For the waveguides we assigned a length-
dependent transmission
T
wg
(
λ
,
L
) =
1
e
−
L
/
L
0
which is quantified in terms of the propagation
length
L
0
of their supported SPPs. Hence,
L
0
is
separated by multiple measurements over a sys-
tematic variation of the waveguide length
L
. In
applying this model to our experiments, we fur-
thermore assume that in- and out-coupling an-
tennas had similar, inversion-invariant properties.
This is a reasonable approximation as we inject
and extract light using the same objective. Hence,
the total system transmission is given by
T
tot
(
λ
) =
T
ant1
(
λ
)
T
wg
(
λ
)
T
ant2
(
λ
)
=
T
ant
(
λ
)
2
T
wg
(
λ
)
.
We experimentally measured the spectral system
4
transmission
T
tot
(
λ
)
in the range of
λ
=
1200
−
1850 nm in steps of 5 nm, coupling in and out
of the circuit from air or through the silica sub-
strate. An ensemble of 20 (5
×
5) different plas-
monic nano-circuits was characterized (Fig. 1a),
repeated on 4 different samples with the same fab-
rication settings. Each parameter variation ensem-
ble contained circuits with different total waveg-
uide lengths (
L
=
12
.
28
−
42
.
28 μm) and different
radii (
R
=
1
−
4 μm) for the 90
◦
bend that turns the
waveguide mode polarization for the emitting an-
tenna.
A systematic analysis of the bend loss over ra-
dius demonstrated that bends with
R
=
3
,
4 μm of-
fer negligible additional loss compared to the lin-
ear waveguide propagation loss, which is compre-
hensible since the modal effective wavelength dif-
ference for those radii is negligible compared to
straight waveguides. Hence, we focused our anal-
ysis to circuits with a 3 μm bend radius.
Based on this finding, we fitted the transmissions
of the circuits with an iterative two-dimensional
least-square fit over the independent variables,
wavelength and waveguide length. The antenna
transmission was modeled assuming a Lorentzian
spectral broadening response,
T
ant
=
T
max
ant
1
+((
λ
−
λ
0
)
/
γ
)
2
.
Under the assumption that dispersive effects of
the in- and outcoupling are solely caused by the
antenna resonance, we find that the peak wave-
length
λ
0
and linewidth
γ
are independent of the
excitation, whereas the maximum antenna effi-
ciency
T
max
ant
is higher for excitation from and
emission into the substrate than into air due to
the refractive index asymmetry of the structure,
n
substrate
≈
1
.
44
>
n
air
=
1.
From this fit the spectral propagation length
L
0
(
λ
)
(Fig. 5a, blue curve) and the spectral an-
tenna transmission
T
ant
(
λ
)
were determined over
all measurements and circuits (few circuits with
obvious fabrication errors were excluded from the
evaluation). The fit error (standard deviation
σ
),
therefore includes statistical errors of both the
measurement and fabrication (Fig. 2b). The de-
termined antenna efficiency is a lower limit, as we
might not have eliminated all possible sources of
losses in the circuit.
The antenna transmission peaks at
λ
max
=
1530 nm with a spectral width of
γ
=
244 nm, cor-
responding to a
FWHM
=
488 nm. The spectral
peak efficiency of the antennas is
T
max
ant,air
=
15
±
1%
from air and
T
max
ant,substr
=
45
±
2% from the sub-
strate with minimum statistical error around the
peak and increasing error towards the measure-
ment limits of
λ
=
1200 nm and
λ
=
1850 nm,
where the absolute circuit transmission and the
InGaAs camera efficiency decrease.
Keeping the initial assumption that the antenna
properties obey optical inversion symmetry, the
total antenna efficiency, giving the total emission
into air and silica, can be spectrally summated to
be
T
max
ant,tot
=
60
±
3%, the highest power transmis-
sion for a waveguide loaded optical antenna, re-
ported to date.
2 μm
2 μm
SEM
1 μm
3D FDTD of antenna: |E|
2
reflector
waveguide
bend
a
b
c
1
10
-8
10
-6
10
-4
10
-2
NSOM intensity
0
0.2
0.4
0.6
0.8
1
Figure 4: (a) SEM of a Yagi antenna that was
illuminated with a highly focused beam through
the substrate and (b) scanned with an aperture
near field optical microscope (NSOM) through the
300 nm thin layer of cladding SiO
2
. Strong en-
hancement of the spatial near-field distribution is
visible around the antenna and the Yagi reflector.
The reflector suppresses emission of the antenna
away from the waveguide and couples the electro-
magnetic wave into the SPP channel waveguide
with only slight emission towards the cladding.
(c) Electric field distribution (
|
E
|
2
) in the structure
plane of the optimized antenna geometry from a
full 3D FDTD simulation in the same configura-
tion as the measurement.
Validity of the applied transmission analysis is
based on the hypothesis that the objective collects
all light that is emitted from the Yagi antennas
and that a good overlap of the focused exciting
beam and the angular emission directionality is
achieved.
5