ARTICLE

Received 22 Apr 2013 | Accepted 5 Sep 2013 | Published 5 Oct 2013

DOI: 10.1038/ncomms3561

Broadband asymmetric waveguiding of light

without polarization limitations
Yadong Xu1,*, Chendong Gu1,*, Bo Hou1,*, Yun Lai1, Jensen Li2 & Huanyang Chen1

Optical diodes are fundamental elements for optical computing and information processing.
Attempts to realize such non-reciprocal propagation of light by breaking the time-reversal
symmetry include using indirect interband photonic transitions, the magneto-optical effect,
optical nonlinearity or photonic crystals. Alternatively, asymmetric reciprocal transmission of
light has been proposed in photonic metamaterial structures for either circularly or linearly
polarized waves. Here we employ the recent concept of gradient index metamaterials to
demonstrate a waveguide with asymmetric propagation of light, independent of polarization.
The device blocks both transverse electric and magnetic polarized modes in one direction but
transmits them in the other for a broadband spectrum. Unlike previous works using chiral
properties of metamaterials, our device is based on the principle of momentum symmetry
breaking at interfaces with phase discontinuities. Experiments in the microwave region verify
our ndings, which may pave the way to feasible passive optical diodes.

1 School of Physical Science and Technology, Soochow University, Suzhou 215006, China. 2 School of Physics and Astronomy, University of Birmingham,
Birmingham B15 2TT, UK. * These authors contributed equally to this work. Correspondence and requests for materials should be addressed to H.C.
(email: chy@suda.edu.cn) or to J.L. (email: j.li@bham.ac.uk).

NATURE COMMUNICATIONS | 4:2561 | DOI: 10.1038/ncomms3561 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms3561

lectrical diodes with p-n junctions are fundamental

elements for integrated circuits. Similarly, devices with
optical diode effect are fundamental elements for information processing in optical integration systems. Great efforts have
been made to realize such optical diodes with non-reciprocal
propagation of light by breaking the time-reversal symmetry,
such as indirect interband photonic transition1,2, magneto-optic
effect37, optical nonlinearity810 or photonic crystals11. In
addition, numerous reciprocal structures are designed as
alternatives to achieve asymmetric propagation of light by
employing metamaterials1220 for either circularly polarized
waves or linearly polarized waves.
Recently, the concept of phase discontinuity21,22 and
metasurfaces has been used to manipulate the propagation of
light and change the conventional laws of reection and
refraction. The basic idea is that abrupt phase shifts are
introduced at the interface between two media, which are a
function of position. Such a conguration will break the
momentum symmetry and provide an additional tangential
momentum to match the momentum conservation, thereby
changing the propagation of light at the interface. This allows for
the design of optical components with ultrathin thickness to
achieve functionalities that go beyond traditional approaches23,24.
In fact, the strong interaction between light and metasurfaces has
also provided new ways to control polarization so that
metasurface lenses can function differently for different
polarizations, including dipolar lenses and polarizationdependent directional couplers2429. Recent approaches exploit
the polarization of incident light to generate asymmetric
propagation of light28,30. However, it will be more useful to
obtain asymmetric propagation independent of polarization.
Here we utilize the metasurface concept and employ gradient
index metamaterials23 (GIMs) to design a simple waveguide
structure to realize asymmetric propagations of light. We rst
combine two metasurfaces into a waveguide structure and
heuristically explain the asymmetric propagation from the view
of geometric optics. We also nd that if we sacrice the thickness,
the metasurfaces will become GIMs. We numerically nd that
such a device can work in a broad band of spectrum and is not
limited to different polarizations of light. By several steps of
reduction, we fabricate a prototype for microwave frequencies.
The measurements verify asymmetric propagation and agree with
the simulations very well.
Results
Concept of asymmetric propagation for a point source. We
start our conguration with the schematic diagram in Fig. 1, that
is, a waveguide structure with three layers of media. In the middle
is air of width w, which is sandwiched by two slabs of GIMs of
width d. The outer boundaries are the perfect electric conductor
(PEC) walls. The index prole of the GIMs are both,
ex mx 1

kx  x0
;
2k0 d

where x0 is the starting point of the waveguide structure, k0 is the

wave vector in free space and k is a crucial momentum parameter
for designing the novel properties of waveguide23. First, let us
look at what happens if we put a line source inside the waveguide.
For a transverse electric (TE) mode, a line current source is
applied. An asymmetric propagation is visually exhibited
in Fig. 1b,c. In Fig. 1b, we show that without the GIMs, the
electromagnetic (EM) wave radiates in both directions, whereas
with the GIMs attached, the EM wave mainly propagates along
the x direction, as shown in Fig. 1c. To show the effect more
clearly, we also plot the power ow of Poynting vectors (time2

averaged, S c/(8p)Re[E  H*]) accordingly. We nd that the

power ows in both directions if the GIMs are absent, whereas for
the current structure with GIMs, the power owing along the  x
direction will be suppressed at a position and come back to join
the power owing along the x direction. In the simulations, the
parameters are set as follows: w is 22.5 mm; k is 0.2k0; the length
of the waveguide L is 240 mm, the starting point of the waveguide
x0 is  120 mm; the width of GIMs d is 1.5 mm the working
wavelength l is 37.5 mm (the working frequency is 8 GHz); and
the range of index prole is from 1 to 17. Similar phenomenon
happens for a transverse magnetic (TM) mode, as shown in
Fig. 1d,e. The point source is a numerical setup with Hz 1 (A/m)
at a tiny circular boundary (see the black circles in Fig. 1d,e).
Figure 1d is the case without GIMs, whereas Fig. 1e is that with
GIMs. Moreover, the power propagates only to the x direction.
By carefully inspecting the eld patterns of Fig. 1c,e, it is clearly
seen that the EM wave radiating from the source propagates
along the  x direction for a distance, before it is blocked at some
position and comes back in the x direction (a U-turn
behaviour). For the EM wave radiating in the x direction, it
will be split into two surface-like waves propagating along the
interfaces between GIMs and air, which is also well demonstrated
by the plotted power ow of Poynting vectors.
Another interesting nding is the interference between the EM
wave coming back from the  x direction after the U-turn and
the EM wave originally radiating along the x direction. Hence,
one may observe that at some frequencies, where the EM wave
from the two directions neutralizes each other, almost all the
energy is localized inside the device and transmission at the right
port of the waveguide is nearly zero, whereas at some frequencies,
they can give rise to constructive interference. In order to
demonstrate such an interference effect, we calculate the power
transmission at both ports of the waveguide. The frequency
dependences are shown in Supplementary Fig. S1a,b. Here all the
chosen parameters of the structure are the same as those in Fig. 1.
From the transmission curves for both TE and TM modes, it is
clear that wave propagating along the  x direction is suppressed
and only a little energy leaks out at the left port (see the red line),
whereas for the right port, the power transmission oscillates for
different frequencies (see the green line), which is stemmed from
the interference effect between two componentsone from the
waves radiating along the x direction and the other from those
radiating along the  x direction with U-turn behaviours. At
some particular frequencies, for example at 9.74 GHz for TE
mode or 10 GHz for TM mode, the power transmission reaches
almost zero. The above two components cancel out each other
completely and almost all the energy is localized inside the
waveguide. To better visualize the effect, we plot the amplitude of
the electric eld for TE mode at 9.74 GHz in Supplementary Fig.
S1c, and the amplitude of the magnetic eld for TM mode at
10 GHz in Supplementary Fig. S1d. The amplitudes at both ports
for the two modes are very weak, and most of the energy is
localized inside the structure.
It is noted that the above working frequency is not below the
cutoff frequency of TE modes (which is fc c/(2(w 2d)) 5.88
GHz for the waveguide structure without GIMs and 2 GHz for
the waveguide structure with GIMs replaced by a dielectric of
refractive index 17). Therefore, the reason why the waves stop at
some position and come back to the x direction is not because
of the stop band of the waveguide structure. In fact, we know that
for TM modes, there is no cutoff frequency for the same
structure, which again shows that the above asymmetric
propagation comes from unknown reasons yet to be further
explored.
Now, let us try to understand the interesting physics behind
this concept. Recently, the concept of phase discontinuity is

NATURE COMMUNICATIONS | 4:2561 | DOI: 10.1038/ncomms3561 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms3561

PEC
Meta-surface

Air
y

Point source

y (mm)

Meta-surface

10

104
3.4

Source

10

y (mm)

120

y (mm)

40

0
x (mm)

10

10
120

40

80

120
104
3.4

Source

3.4
80

40

0
x (mm)

10

40

80

120

Source

10
120

y (mm)

3.4
80

2
80

40

0
x (mm)

10

40

80

120

Source

10
120

2
80

40

0
x (mm)

40

80

120

Figure 1 | Concept of asymmetric propagation of light inside a waveguide. (a) Schematic diagram of a one-dimensional waveguide structure with
two layers of GIMs of width d (the gradually changing blue regions). The core medium in the waveguide is air. An excited point source is located inside the
waveguide. The waveguide is bounded by perfect electric conductor walls. (b) The simulated Ez eld distribution for a TE polarized line source (denoted
by a red point) inside the waveguide without the GIMs. (c) The simulated Ez eld distribution for a TE source when the GIMs are attached, showing
the asymmetric propagation. Ez in (b) and (c) is in units of V m  1. (d) The simulated Hz eld distribution for a TM polarized line source (denoted by a tiny
circle) inside the waveguide without the GIMs. (e) The simulated Hz eld distribution for a TM source when the GIMs are attached, showing the
asymmetric propagation. Hz in (d) and (e) is in units of A m  1. The black arrows represent the power ow of time-averaged Poynting vectors.

introduced to control the propagations of EM wave using

metasurface consisting of sub-wavelength metal antennas or
GIMs2123. If dool and doow, we can regard the two slabs of
GIMs here as metasurfaces23. In the above cases, l 20d, w 15d
The index prole of equation (1) provides an approximate linear
function of phase shifts: j(x) j0 k(x  x0), where j0 is the
initial phase. The EM wave propagates through the GIMs and is
reected by the PEC walls and nally propagates across the
interface, causing a gradient phase shift along the x direction. Such
a gradient phase shift along the interface leads to an additional
linear momentum,
jjj k rjx djx=dx;

for each part of the EM wave across the upper or lower

metasurface. Here the direction of j is set along the x direction.
First, we would like to visualize the physics for the asymmetric
propagation of light within the metasurface waveguide using a
geometric optics approach (Fig. 1a). The EM waves (now rays)
coming from the point source rst radiate in both directions. For
the x direction, after the rays impinge the upper PEC wall, they
will not propagate according to the conventional reection law

r
kin
x kx (as shown by red dashed line) but propagate according to
r
the generalized reection law kin
x kx  k (as shown by red solid
line) because the additional momentum should be taken into
account to maintain momentum conservation21. Here we dene
the x direction to be positive for wave vectors. Thus, we can see
that the reected angle becomes larger than the incident angle.
After the reected rays propagate and impinge the lower PEC wall,
the reected angle becomes even larger. Consequently, we can
predict that after several times of reection, the rays (or waves)
propagating along the x direction will be converted into surface
waves propagating along the interfaces between metasurfaces and
air when krx kin
x Nk4k0 (N is positive integer presenting the
reection time). However, it is completely different for rays
propagating along the  x direction. When they impinge the
upper PEC wall, the reected rays will not propagate according to
r
the conventional reection law kin
x kx (as shown by black
dashed line) but will propagate according to the generalized
r
reection law kin
x kx k (as shown by black solid line). In this
way, the reected angle becomes smaller than the incident angle.
After several times of reection, the reected angle becomes
smaller and smaller. Hence, as long as N is large enough, the sign

NATURE COMMUNICATIONS | 4:2561 | DOI: 10.1038/ncomms3561 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms3561

of  krx  kin
x Nk will be changed from  to . It means
that the rays (or waves) propagating along the  x direction come
back. Thus, following the same physics of rays (or waves)
propagating along the x direction, the rays (or waves) coming
back are nally converted into surface waves, thereby interfering
with those originally propagating along the x direction.
Therefore, the total effect of EM wave inside the waveguide with
metasurfaces is that it can only propagate along one direction.
Physics of asymmetric propagation using dispersion analysis.
The above explanation provides a simple physics picture behind this
asymmetric transmission phenomenon. However, in order to
explain the underlying physics in a quantitative manner, we would
next examine the physics using the wave optics approach. The
asymmetric propagation behaviour of the waveguide can be best
described by its waveguide modes and the associated dispersion
relation. We numerically calculate the power transmission at one
port (the output port) when the EM wave is incident from the other
port (the input port) in the x direction. For TE mode, the rst mode
is coupled into the waveguide, whereas for TM mode, the zeroth
mode is coupled into the waveguide. In Fig. 2a, the black curve
shows the power transmission at different frequencies for the
waveguide without GIMs. When the GIMs are added, high transmission (B80% of above one, denoted by the green curve) can be
achieved if the input energy of the EM wave from the left port is the
same. The difference from unity is mainly stemmed from the
mismatched impedances at the interfaces between the waveguide
and the wave coupler instead of material loss. However, for the
incident EM wave from the right port, it is clearly seen that the
calculated transmission (shown by the red curve) is very low for a
broadband of frequencies from 7 to 13 GHz, which demonstrates
signicantly asymmetric transmission with difference over 20 dB in
the two directions. For the same device, the asymmetric transmission also happens for TM mode, as shown in Fig. 2b. For the
incident EM wave from right, the calculated transmission is also
nearly zero (denoted by the red curve), whereas for the incident EM
wave from left, the power transmission (denoted by the dashed
green curve) almost reaches that of the waveguide without GIMs
(denoted by the black curve). The reason for this is that for TM
mode there is no cutoff frequency, and the zeroth mode supported
in the waveguide is mainly in the x direction; therefore, the power of
the zeroth mode can be perfectly coupled into the left port of the
waveguide, leading to high transmission as the case without GIMs.
To show the asymmetric transmission more vividly, the eld
patterns for TE mode are calculated and plotted in Fig. 2c,d at the
frequency 8 GHz. Figure 2c is the eld pattern for the wave
incident from left, whereas Fig. 2d is that for wave from right. For
TM mode, similar effect happens (see Supplementary Fig. S4 and
Supplementary Note S2).
In fact, the band structures of the proposed device reveal the
above asymmetric propagation effect more accurately than that
from geometric optics. The geometric optics only gives a heuristic
description. We notice that the band structures of the waveguide
with GIMs are difcult to gure out. However, if we replace the
GIMs with a kind of dielectric with a constant refractive index,
the dispersion relations of the waveguide can be calculated
analytically (for details, see Supplementary Note 1). Therefore, we
calculate ve dispersion relations for ve kinds of dielectrics
attached to the waveguide and plot them in Fig. 2e. Their
refractive indices are related to ve different positions of the
GIMs, which are 1, 8, 11, 14 and 17. In particular, n 1 and
n 17 are related to the left and right ports of the waveguide,
respectively; n 11 is related to the approximate position (see the
vertical blue dashed line in Fig. 2d) where the incident EM wave
(of frequency 8 GHz) from right is blocked.
4

Now we come to explain the phenomena using the above

dispersion relations. In Fig. 2c, when the EM wave is incident
from the left port and reaches the interface between the wave
coupler and the waveguide, it will be coupled as the rst mode
(m 1) as shown in the rst plot in Fig. 2e. After that, it will be
gradually converted into surface waves because the branch of the
rst mode goes below the light line when the refractive indices of
the dielectrics increase (see in plots 25 in Fig. 2e). For the
incident wave from the right port, when it reaches the interface
between the wave coupler and the waveguide, it will rst be
coupled as the third mode (m 3), as shown in the fth plot in
Fig. 2e. Afterwards, the EM wave propagates and keeps the third
mode unchanged until it is blocked at the position where the
refractive index is about 11. When we look at the dispersion
relation at this position, as shown in the third plot in Fig. 2e, we
nd that the working frequency is just related to the bottom of
the third mode. Therefore, the EM wave is either changed into the
rst/second mode and continues to propagate in the x direction
(as shown in the rst and second plots in Fig. 2e, there is no real
solution for the fourth mode) or is reected and propagates in the
x direction as the same mode (the third mode). If we examine
the electrical eld pattern carefully, we will nd that most of the
waves are reected. Only a little energy is changed into the rst/
second mode and it leaks out to the x direction. For example, we
plot the same eld pattern but with a colour bar of a much
smaller range in Supplementary Fig. S3b, where we nd only a
little energy leaking out, mainly as the rst mode (m 1).
However, compared with the amplitude of the incident eld, the
leaked wave can be almost neglected. For instance, when we plot
the electric eld along the middle plane (the horizontal dashed
line in Supplementary Fig. S3a) of the waveguide in
Supplementary Fig. S3c, the oscillations of the leaked wave at
the left part is much smaller than those of the incident wave at the
right part of the waveguide structure (about 1/30 if we magnify
the part near the stopping position as shown in Supplementary
Fig. S3d). In this way, the EM wave from the right port is blocked
at the position where n 11 and shows an U-turn behaviour
later.
For details, we plot the waveguide modes at the ve positions
to demonstrate the evolution of modes in Fig. 2f. Comparing the
simulations in Fig. 2c,d and the mode proles in Fig. 2f, we can
clearly observe that, when EM wave is incident from the left, it is
coupled to the rst mode A1. When it reaches the position
where the index is 8, it is still the rst mode B1. Then it will be
gradually converted into surface wave with the rst mode, as
shown in C1. The second mode C2, however, is not excited out
here. After this position, the modes are unchanged, keeping the
rst modes with surface waves, as shown in Fig. 2c. We do not
show the mode proles D1 and E1 at positions with indices 14
and 17, as we cannot see the cross points of the rst modes from
the current dispersions (the wave vectors are too large here). For
the right incident EM wave, however, the modes are all third
modes, as shown by E3 and D3. Therefore, the evolution of the
waveguide modes for EM wave coupled from left and from right
is now very clear to usthat is, A1B1C1D1E1 and E3D3,
respectively. The results are consistent with the eld pattern in
Fig. 2c,d.
For TM mode, similar physics can also be found from the
dispersion relations, resulting in the asymmetric transmission of
light (see Supplementary Note 2).
In fact, the only necessary condition for asymmetry propagation is that we have two slabs of GIMs as the cladding layers of
the waveguide, whose refractive index increases along the x
direction and the change happens at least in a distance larger than
a wavelength. This allows us to make a more general and robust
version. If we replace the metasurfaces with GIMs in the above

NATURE COMMUNICATIONS | 4:2561 | DOI: 10.1038/ncomms3561 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms3561

20

20
Transmission (dB)

Transmission (dB)

0
20
40
60
80
100
7

TE mode
8

9 10 11 12
Frequency (GHz)

0
20
40
60

13

TM mode

80
100

9 10 11 12
Frequency (GHz)

13

10
y (mm)

10
y (mm)

10
120

0
x (mm)

120

0
10
120

0
x (mm)

m=1

120

m=3

10

10

m=3

10

m=3

10

10

m=3

Frequency (GHz)

m=2
8

8
B1

A1

6 m=2

m=1
C2

4 8

4
m=1

0.5

 =  = 11

1.5

 =  = 14

0 0.5 1 1.5 2
Wave vector  (2/a)

B1

==8

3.1

==8

3.1

1.5
D3

1.5
E3

0
 =  = 11

3.1

1.5

C2

 =  = 11

1.5

C1

0 0.5 1 1.5 2

B2

0
==1

 =  = 17

0 0.5 1 1.5 2

1.5

1.5
A1

m=1

1.57

==8

==1
0 0.5 1 1.5 2

6
m=2

m=1

C1
2

E3

D3
6

B2

8
m=2

0
 =  = 14

3.1

 =  = 17
1.5

1.5

Figure 2 | Asymmetric propagation for EM wave incident from left and right in the x direction. (a,b) The calculated power transmission for TE and
TM modes. The black curves are for the waveguide system without the GIMs. The green curves are for the system with the GIMs and with an input
from the left port. The red curves are for the system with the GIMs and with an input from the right port. The power transmission is calculated by
integrating the Poynting vector along the cross-section of the output port. The width of the waveguide couplers set beside is the same as the width of the
core medium of the waveguide, 22.5 mm. (c,d) The simulated Ez eld distribution for the EM wave incident from left and right at 8 GHz, respectively.
Ez in c and d is in units of Vm  1. The dashed blue line denotes the position where the wave stops. The arrows in c and d show the propagation directions
and the U-turn behaviours of the waves. (e) The dispersion relations for TE mode but with the GIMs replaced by ve kinds of dielectrics, whose refractive
indices are related to ve different positions of the GIMs. From left to right, the plots correspond to refractive indices 1, 8, 11, 14 and 17. The black
dashed lines are the light lines. The red dashed lines indicate the working frequency of (c) at which different waveguide modes are indicated by the arrows
and numbered by capital letters with a number relating to the bands. The inset in the middle panel is a magnied plot for a band region indicated
by a dashed circle. (f) The mode proles of the electric eld Ez for the corresponding waveguide modes as shown in (e) at the working frequency. The unit
of Ez is V m  1. The unit cell is l 25.5 mm.
NATURE COMMUNICATIONS | 4:2561 | DOI: 10.1038/ncomms3561 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms3561

heuristic geometric model, we will nd that the phase shift is no

longer a linear function of x, as the EM wave inside the GIMs is
not a plane wave along the y direction and its wave vector is
changed gradually as propagating. In other words, the refractive
index prole should not be necessarily a linear function of x but
just a monotonously increasing function of x.
Microwave implementation of asymmetric propagation. The
proposed device here can work in principle from microwave
frequencies all the way to optical frequencies as long as the
materials with required refractive index prole can be fabricated.
It is noted that in the above discussions, the materials with the
special refractive index prole is a bit challenging to realize using
the current metamaterial techniques. In addition, materials with
high permeability at optical frequencies would violate the
LandauLifshitz permeability argument31. Fortunately, we could
design a reduced version alternatively by increasing the width of
the metasurfaces, and setting m(x) 1, e(x) (1 k(x  x0)/
2k0d)2. We numerically nd that the asymmetric transmission
still happens for the reduced version (see Supplementary Note 3).
Figure 3ac is a picture of the fabricated sample, in which the
parameters of the reduced device are set as follows: w 22.5 mm,

36
I

28.8
54
II
III
Top view

93.6 (mm)
IV

Source

Source

Samples

7.2 mm

Aluminum

Unit cell

d 7.2 mm, L 212.4 mm, k 0.2k0 and x0 0, with the permittivity ranging from 1 to 16 and m 1. The GIM regions are
actually discretized into four parts, each with different substrate
materials, whose e is 2.2, 3.5, 7.5 and 16, respectively, and corresponding to the regions I, II, III and IV in Fig. 3a. By drilling
holes with different sizes, we can obtain the above required
refractive index prole along the x direction (see Supplementary
Note 4).
First, we scanned the electric eld patterns for two cases (the
simulated patterns are shown together). One is for a TE mode
coupled from the left port (see Fig. 3b), and the eld patterns are
plotted in Fig. 3d (simulation) and 3e (measurement). The wave
propagates to the x direction and is smoothly transferred into
surface waves. The other is for a TE mode coupled from the right
port (see Fig. 3c), and the eld patterns are plotted in Fig. 3f
(simulation) and 3g (measurement). The EM wave stops at some
position after propagating to the  x direction and comes back
again to the x direction before being converted into surface
waves. Both the above two cases are obtained at the frequency
8 GHz and show the asymmetric propagation of EM waves very
clearly as proposed in the above theoretical sections. During the
measurements, we have to add the wave-absorbing materials at
the right port to eliminate the scattering caused by the

18.45

x (mm)

18.45
212.4

x (mm)
18.45

20

30

40
18.45
212.4

Exp.
x (mm)

50

Simu.
8

18.45
y (mm)

0
0
Transmission (dB)

y (mm)

10
18.45

y (mm)

18.45
212.4

y (mm)

3.6 mm

18.45
212.4

10

11

Frequency (GHz)

x (mm)

Figure 3 | Experimental measurements of asymmetric propagation for TE modes. (a) The fabricated sample (top view). The unit cell of the fabricated
metamaterial is magnied as well. (b,c), The congurations for the TE modes to be coupled from left and right, respectively. (d,e), The simulated and
scanned electric eld patterns (|Ez|) at 8 GHz corresponding to the congurations in (b). (f,g) The simulated and scanned electric eld patterns
(|Ez|) at 8 GHz corresponding to the congurations in (c). |Ez| in dg are dimensionless with each normalized to its maximum value. The arrows in dg show
the propagation directions and the U-turn behaviours of the waves. (h) The power transmission at different frequencies for different congurations.
The solid lines are numerical results, whereas the data points are measured results. The curve in green and the green data points are for wave
incidence from the left, whereas those in red are for wave incidence from the right. The curve in black and the black data points are for the conguration
without GIMs.
6

NATURE COMMUNICATIONS | 4:2561 | DOI: 10.1038/ncomms3561 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms3561

impedance mismatching at the interfaces of the GIM regions

and air. Then, we measured the power transmission for the
above two TE modes incident and coupled in the x direction
at different frequencies, as shown in Fig. 3h. The transmission
curve for the wave coupled from the left is denoted by the green
triangular experiment data points. The transmission is close to
the waveguide structure without GIMs (the black plus-sign
experiment data points). The transmission curve for the wave
coupled from the right is denoted by the red star experiment
data points, which is reduced to about  15 dB from 7 to
11 GHz. We also plot the related transmission curves from
numerical simulations (denoted by solid lines in the same
colours correspondingly) for comparison. We note that it is
difcult to measure the power transmission directly using a
point detector. Instead, we measure |Ez|2 (similar to those in
Fig. 3dg) and integrate it along the cross-section of the output
port. Such a value is proportional to the energy. By normalizing
it to the input power, we can obtain the power transmission
curves for different congurations. When we scanned the eld
patterns, a 1-mm air gap between the samples and the upper
aluminium plate is left for the detector to insert. Such a gap
results in some energy leaking during the measurement. That is
why the elds in the GIM regions are enhanced, causing some
inconsistencies with the simulated ones. However, the asymmetry propagation is still well demonstrated from the eld
patterns. For the measurements of the transmission curves, we
close the gap and measure the electric elds at the ports,
therefore showing higher accuracy and more consistency with
the simulated results. The asymmetry propagation happens for a
very broadband of spectrums.
Discussion
We have proposed and fabricated a simple waveguide device with
asymmetric propagation effect. The structure is simple to
fabricate as only dielectrics are needed and works well for a
broadband of frequencies. Most importantly, the device shows
very good performance of asymmetric transmission for both TE
and TM modes, which can be extended to three-dimensional
photonic structure (see Supplementary Note 5). Although only
the experiment at the microwave region is shown in this Article,
we expect our device can be realized at THz or even optical
frequencies in future, considering the recent progress of
invisibility cloaks3234. However, we note that the asymmetric
transmission assisted with mode conversion demonstrated here is
different from the asymmetric transmission of the same mode
along a waveguide arising from non-reciprocity.
Methods
Theory and simulation. The eld patterns in Figs 1be, 2c,d,f and 3d,f are
obtained using the nite element solver COMSOL Multiphysics. The power
transmission in Fig. 2a,b (and a part in Fig. 3h) is also numerically calculated using
COMSOL Multiphysics. The power transmission is calculated by integrating the
Poynting vectors along the cross-section near the output port of the waveguide, and
normalized to the incident power. In all calculations, the scattering boundaries are
set for both sides of the waveguide.
Sample fabrication. The materials in regions I, II, III and IV are fabricated by
drilling different sizes of holes in four pieces of different dielectric plates35, whose
permittivities are 2.2, 3.5, 7.5 and 16, respectively. The detailed sizes of the holes in
each part are listed in Supplementary Table S1.
Experimental setup. The functionalities of the above waveguide structure can be
characterized by testing the sample of nite height inside a parallel-plate waveguide
system36. The system consists of two at conducting plates spaced 6 mm apart,
whereas inside the gap the TE mode is dominant. The lower plate rests on the
translational stage and is movable along two orthogonal directions. Waveguide
structure is located at the lower plate. The probing coaxial detects the electric eld
distribution inside the gap between the sample and the upper plate via a small hole

in the upper plate. In order to detect the E-eld on the GIMs, a tiny hole is drilled
in the upper plate of the parallel-plate waveguide system and inserted into the hole
of the coaxial, which has an inner conductor (o1 mm) naked and extruded into
the air gap between the sample and the upper plate. The probe (the inner
conductor of the coaxial) is directing the z direction to measure the Ez component.
In the measurement, we scan the region with the spatial resolution 1 mm  2 mm
and record the signal received by the probing coaxial at every spatial point with a
microwave network analyser. Two machined aluminium blocks are used as the
PEC boundaries and are xed on the lower plate together with the sample, as
shown in Fig. 3a. In our experiments, both the aluminium blocks and the sample
are 5 mm high. To excite a TE wave in the source, a X-band coaxial-to-waveguide
adapter is employed. Such an adapter is the standard microwave waveguide
component, with one end being coaxial interface and the other end being rectangular waveguide interface for launching TE modes (http://www.atmmicrowave.com/wave.html). In our experiments, the adapter is xed to the edge of the lower
plate of the parallel-plate waveguide system and its top cover is removed to
facilitate the movement of the lower plate with respect to the upper plate, which is
big enough to cover the adapter and the GIMs during the movement.

References
1. Yu, Z. F. & Fan, S. H. Complete optical isolation created by indirect interband
photonic transitions. Nat. Photon. 3, 9195 (2009).
2. Lira, H., Yu, Z. F., Fan, S. H. & Lipson, M. Electrically driven nonreciprocity
induced by interband photonic transition on a silicon chip. Phys. Rev. Lett. 109,
033901 (2012).
3. Bi, L. et al. On-chip optical isolation in monolithically integrated nonreciprocal optical resonators. Nat. Photon. 5, 758762 (2011).
4. Espinola, R. L., Izuhara, T., Tsai, M. C., Osgood, R. M. & Dotsch, J. H.
Magneto-optical nonreciprocal phase shift in garnet/silicon-on-insulator
waveguides. Opt. Lett. 29, 941943 (2004).
5. Levy, M. Nanomagnetic route to bias-magnet-free, on-chip Faraday rotators.
J. Opt. Soc. Am. B 22, 254260 (2005).
6. Zaman, T. R., Guo, X. & Ram, R. J. Faraday rotation in an InP waveguide. Appl.
Phys. Lett. 90, 023514 (2007).
7. Dotsch, H. et al. Applications of magneto-optical waveguides in integrated
optics: review. J. Opt. Soc. Am. B 22, 240253 (2005).
8. Soljaic, M., Luo, C. Y., Joannopoulos, J. D. & Fan, S. H. Nonlinear photonic
microdevices for optical integration. Opt. Lett. 28, 637639 (2003).
9. Gallo, K., Assanto, G., Parameswaran, K. R. & Fejer, M. M. All-optical diode in
a periodically poled lithium niobate waveguide. Appl. Phys. Lett. 79, 314316
(2001).
10. Fan, L. et al. An all-silicon passive optical diode. Science 335, 447450 (2012).
11. Haldane, F. D. M. & Raghu, S. Possible realization of directional optical
waveguides in photonic crystals with broken time-reversal symmetry. Phys. Rev.
Lett. 100, 013904 (2008).
12. Fedotov, V. A. et al. Asymmetric propagation of electromagnetic waves through
a planar chiral structure. Phys. Rev. Lett. 97, 167401 (2006).
13. Fedotov, V. A., Schwanecke, A. S., Zheludev, N. I., Khardikov, V. V. &
Prosvirnin, S. L. Asymmetric transmission of light and enantiomerically
sensitive plasmon resonance in planar chiral nanostructures. Nano Lett. 7,
19961999 (2007).
14. Schwanecke, A. S. et al. Nanostructured metal lm with asymmetric optical
transmission. Nano Lett. 8, 29402943 (2008).
15. Drezet, A., Genet, C., Laluet, J. Y. & Ebbesen, T. W. Optical chirality without
optical activity: how surface plasmons give a twist to light. Opt. Exp. 16,
1255912570 (2008).
16. Zhukovsky, S. V., Novitsky, A. V. & Galynsky, V. M. Elliptical dichroism:
operating principle of planar chiral metamaterials. Opt. Lett. 34, 19881990
(2009).
17. Singh, R. et al. Terahertz metamaterial with asymmetric transmission. Phys.
Rev. B 80, 153104 (2009).
18. Plum, E., Fedotov, V. A. & Zheludev, N. I. Planar metamaterial with
transmission and reection that depend on the direction of incidence. Appl.
Phys. Lett. 94, 131901 (2009).
19. Menzel, A. et al. Asymmetric transmission of linearly polarized light at optical
metamaterials. Phys. Rev. Lett. 104, 253902 (2010).
20. Mutlu, M., Akosman, A. E., Serebryannikov, A. E. & Ozbay, E. Diodelike
asymmetric transmission of linearly polarizedwaves using magnetoelectric
coupling and electromagnetic wave tunneling. Phys. Rev. Lett. 108, 213905
(2012).
21. Yu, N. et al. Light propagation with phase discontinuities: generalized laws of
reection and refraction. Science 334, 333337 (2011).
22. Ni, X., Emani, N. K., Kildishev, A. ,V., Boltasseva, A. & Shalaev, V. M.
Broadband light bending with plasmonic nanoantennas. Science 335, 427
(2012).
23. Sun, S. L. et al. Gradient-index meta-surfaces as a bridge linking propagating
waves and surface waves. Nat. Mater. 11, 426431 (2012).

NATURE COMMUNICATIONS | 4:2561 | DOI: 10.1038/ncomms3561 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms3561

24. Francesco, A. et al. Aberration-free ultrathin at lenses and axicons at telecom

wavelengths based on plasmonic metasurfaces. Nano Lett. 12, 49324936
(2012).
25. Chen, X. et al. Dual-polarity plasmonic metalens for visible light. Nat.
Commun. 3, 1198 (2012).
26. Kildishev, A. V. et al. Planar photonics with metasurfaces. Science 339,
12891289 (2013).
27. Yin, X. et al. Photonic spin hall effect at metasurfaces. Science 339, 14051407
(2013).
28. Lin, J. et al. Polarization-controlled tunable directional coupling of surface
plasmon polaritons. Science 340, 331334 (2013).
29. Huang, L. et al. Helicity dependent directional surface plasmon polariton
excitation using a metasurface with interfacial phase discontinuity. Light Sci.
Appl. 2, e70 (2013).
30. Balthasar Mueller, J. P. & Capasso, F. Strong asymmetric surface plasmon
polariton emission by a dipole emitter near a metal surface. Preprint at http://
arxiv.org/abs/1304.2403 (2013).
31. Merlin, R. Metamaterial and the Landau-Lifshitz permeability argument: large
permittivity begets high-frequency magnetism. Proc. Natl Acad. Sci. USA 106,
16931698 (2009).
32. Schurig, D. et al. Metamaterial electromagnetic cloak at microwave frequencies.
Science 314, 977980 (2006).
33. Valentine, J., Li, J., Zentgraf, T., Bartal, G. & Zhang, X. An optical cloak made of
dielectrics. Nat. Mater. 8, 568571 (2009).
34. Ergin, T., Stenger, N., Brenner, P., Pendry, J. B. & Wegener, M. Threedimensional invisibility cloak at optical wavelengths. Science 328, 337339
(2010).
35. Mei, Z. L., Bai, J. & Cui, T. J. Gradient index metamaterials realized by drilling
hole arrays. J. Phys. D: Appl. Phys. 43, 055404 (2010).
36. Justice, B. J. et al. Spatial mapping of the internal and external electromagnetic elds of negative index metamaterials. Opt. Express 14, 86948705
(2006).

Acknowledgements
This work is supported by the National Science Foundation of China for Excellent Young
Scientists (grant no. 61322504), the Foundation for the Author of National Excellent
Doctoral Dissertation of China (grant no. 201217), the National Natural Science Foundation of China (grant nos 11004147, 11104196 and 11104198), the Natural Science Foundation of Jiangsu Province (grant nos. BK2010211 and BK2011277), the State Key Program
for Basic Research of China (grant no. 2012CB921501), the Priority Academic Program
Development (PAPD) of Jiangsu Higher Education Institutions and the Program for New
Century Excellent Talents in University (NCET). J.L. would like to thank the support from
the GRO program of Samsung Advanced Institute of Technology for the work and the
student visit. We would like to thank Profs. T. Tyc, Z. Hang and W.X. Lu for their helpful
discussions. H.C. would like to thank Winsley Yangs effort to polish the English.

Author contributions
Y.X., J.L. and H.C. conceived the idea. Y.X. and J.L. performed the initial modelling. Y.X.
and Y.L. performed the theoretical calculations and the numerical simulations. C.G. and
B.H. fabricated the samples and performed the experimental measurements. J.L. supervised the initial stage. B.H. supervised the experimental part. H.C. supervised the whole
project. Y.X. and H.C. wrote the manuscript, all the authors reviewed the manuscript.

Additional information
Supplementary Information accompanies this paper at http://www.nature.com/
naturecommunications
Competing nancial interests: The authors declare no competing nancial interests.
Reprints and permission information is available online at http://npg.nature.com/
reprintsandpermissions/
How to cite this article: Xu, Y. et al. Broadband asymmetric waveguiding of light
without polarization limitations. Nat. Commun. 4:2561 doi: 10.1038/ncomms3561
(2013).

NATURE COMMUNICATIONS | 4:2561 | DOI: 10.1038/ncomms3561 | www.nature.com/naturecommunications

& 2013 Macmillan Publishers Limited. All rights reserved.