On-chip silicon switchable polarization beam splitter

Abstract

We propose and experimentally demonstrate an on-chip switchable polarization beam splitter (PBS) using silicon waveguides. To the best of our knowledge, it is the first demonstration of an on-chip PBS that is not only able to split polarization beams but can be tuned to allow these beams to switch the output paths. The design of the switchable PBS is based on a directional coupler. Measurements show extinction ratios of >12 dB in both the initial state and the switched state, which is realized by heating the device up to 57°C. By adding switching ability to an on-chip PBS, this work is expected to benefit quantum technology, communications, microwave photonics, etc.

Full text

Letter Vol. 47,No. 4 /15 February 2022 / Optics Letters 961
On-chip silicon switchable polarization beam splitter
Shaojie Yin,1,2,3 Huaqing Qiu,4Zhibin Wang,3Daoxin Dai,1,2,5 AND Xiaowei Guan1,2,4,∗
1Jiaxing Key Laboratory of Photonic Sensing & Intelligent Imaging, Jiaxing 314000, China
2Intelligent Optics & Photonics Research Center, Jiaxing Research Institute Zhejiang University, Jiaxing 314000, China
3School of Electrical Engineering, Yanshan University, 066004 Qinhuangdao, China
4DTU Fotonik, Department of Photonics Engineering, Technical University of Denmark, Ørsteds Plads, 2800 Kgs. Lyngby, Denmark
5State Key Laboratory for Modern Optical Instrumentation, College of Optical Science and Engineering, International Research Center for
Advanced Photonics, Zhejiang University, Hangzhou 310058, China
*Corresponding author: xiaoweiphotonics@gmail.com
Received 16 December 2021; revised 12 January 2022; accepted 12 January 2022; posted 13 January 2022; published 11 February 2022
We propose and experimentally demonstrate an on-chip
switchable polarization beam splitter (PBS) using silicon
waveguides. To the best of our knowledge, it is the first
demonstration of an on-chip PBS that is not only able to
split polarization beams but can be tuned to allow these
beams to switch the output paths. The design of the switch-
able PBS is based on a directional coupler. Measurements
show extinction ratios of >12 dB in both the initial state and
the switched state, which is realized by heating the device up
to 57°C. By adding switching ability to an on-chip PBS, this
work is expected to benefit quantum technology, communi-
cations, microwave photonics, etc. © 2022 Optica Publishing
Group
https://doi.org/10.1364/OL.451486
Polarization handling is a long-standing topic in optical systems
and usually involves devices such as polarization beam split-
ters (PBSs) [1,2], polarizers [3,4], polarization rotators [5,6],
and polarization splitter and rotators (PSRs) [7,8]. With their
capability to split orthogonally polarized light into different
paths, PBSs have been found to be indispensable in conven-
tional applications such as microwave photonics [9,10], optical
communications [11,12], and optical sensors [13] as well as
emerging applications such as quantum information process-
ing [14–16] and optical computing [17]. While mature discrete
PBSs based on bulky crystals, film-covered lenses, or optical
fibers have been widely used in free-space optics or fiber optics,
on-chip PBSs are still under intensive research and are attracting
increasing attention towards the creation of a system on a small
and low-cost chip. In particular, numerous PBSs fabricated on
a silicon photonic platform have been demonstrated that use,
for example, symmetrical or asymmetric directional couplers
[18,19], a Mach–Zehnder interferometer (MZI) [20], microring
resonators [21], a multi-mode interferometer (MMI) [22], and
grating couplers [23].
However, these silicon photonic PBSs are limited by the same
characteristic: once fabricated, they only allow the polarization
beams to output along fixed paths. On the other hand, tun-
able PBSs have been theoretically proposed or experimentally
demonstrated by filling dual hollow-core fibers with magnet-
ically sensitive material [24], putting a tapered, fiber-based
PBS into an opto-fluidic ring resonator [25], or using a size-
switchable microsphere [26]. Unfortunately, none of these
devices were on a chip and hence they could not leverage
the manufacturing scalability of silicon photonics technology.
Moreover, they were not switchable, i.e., the two polarization
beams could not completely swap their output paths. Thus, such
a switchable function requires a proof of concept, as it could
be useful for enhancing the flexibility of, e.g., optical routers,
multi-parameter optical sensors, switchable microwave photonic
devices, or quantum information possessors.
In this work, we report the design, fabrication, and measure-
ment of a switchable silicon photonic PBS, which we believe
is the first demonstration of an on-chip PBS possessing such a
function. By carefully designing the length of a silicon photonic
directional coupler (DC), the two orthogonally polarized beams
are coupled back and forth an odd and even number of times,
respectively. After heating up the DC, the two beams are again
coupled multiple times, but one less time than before the heat-
ing, i.e., an even and odd number of times, respectively. This
causes not only splitting but also switching of the two polar-
ization beams. Measurements show extinction ratios of >12 dB
for both output paths before the heating and after the heating as
well.
Figure 1(a) shows a schematic of the configuration of the
proposed silicon photonic DC-based switchable PBS. The
straight-waveguide DC has a length of L, and the 90°-bent
waveguides have the same radius, R. The inset shows a cross
section of the DC; the structural parameters are the wave-
guide height h, the waveguide width w, and the distance
between the two parallel straight waveguides wgap. Using a
finite-difference method (FDM, Mode Solutions from Ansys
Inc.), we calculate the modes of the transverse-electric (TE)
and transverse-magnetic (TM) polarizations at a wavelength of
1550 nm, as shown by the electric-field profiles in Fig. 1(b).
Both the fundamental (even) modes, i.e., TE0and TM0, and the
first higher-order (odd) modes, i.e., TE1and TM1, are exhib-
ited. Here, the refractive indices (RIs) of the silicon core and
the silica cladding are nSi =3.478 and nSiO2 =1.444 at room tem-
perature (300 K), and the structure is h=250 nm, w=400 nm,
and wgap =150 nm.
We then extract the propagation constants of the foregoing
modes and calculate the coupling length Lc, defined as the length
0146-9592/22/040961-04 Journal ©2022 Optica Publishing Group

962 Vol. 47,No. 4 / 15 February 2022/ Optics Letters Letter
Fig. 1. (a) Schematic configuration of the proposed switchable
PBS based on a silicon waveguide DC. Inset shows a cross section
of the parallel coupling region. (b) Simulated mode profiles (elec-
tric field) at 1550 nm in the DC with h=250 nm, w=400 nm, and
wgap =150 nm.
for which the light power is completely transferred from one
waveguide to the other in the DC. Lcis usually different for TE
and TM polarizations since they have different mode profiles, as
already shown in Fig. 1(b). Meanwhile, Lccan be theoretically
calculated with the formulas
Lc(TE)=π
|βe,TE −βo,TE|,(1)
Lc(TM)=π
|βe,TM −βo,TM|,(2)
where Lc(TE)and Lc(TM)are the coupling lengths of the TE and
TM polarization beams, and βe,TE,βo,TE ,βe,TM, and βo,TM are
the propagation constants of the even and odd modes of the two
polarizations, respectively. The propagation constants and hence
the coupling lengths are affected by both the waveguide dimen-
sions and the material properties, which depend on parameters
such as the temperature. Therefore, we first study the influences
of w,wgap, and the heating temperature Ton the coupling lengths.
Figures 2(a) and 2(b) separately show the dependences of
Lc(TE)and Lc(TM)on wgap for different wvalues. It can easily be
found that both Lc(TE)and Lc(TM)tremendously increase as wgap
or wincreases. Furthermore, the physical length of the DC in
the proposed switchable PBS depends on not only the absolute
Fig. 2. Calculated coupling lengths of different polarization
beams and their relativity for a silicon DC as a function of the
waveguide dimensions. Curves from bottom to top, where wis
equal to 300, 350, 400, 450, and 500 nm. (a) Coupling length of
TE, Lc(TE). (b) Coupling length of TM, Lc(TM). (c) Lc(TE)/Lc(TM). (d)
Temperature dependence of Lc(TE)/Lc(TM). Here, h=250 nm is fixed
and the wavelength is 1550 nm.
values of Lc(TE)and Lc(TM)but also the relative value of them and
the variation of this relativity with temperature. In principle, we
hope that the coupling length of one polarization beam is far
larger than that of the other, and that the coupling length should
be much more temperature sensitive. Thus, one can easily find
a physical length that fulfills the design principle and allows
the two polarization beams to couple between the two parallel
waveguides an even/odd number of times before heating and an
odd/even number of times after heating. In order to reach this
goal, we define the ratio K=Lc(TE)/Lc(TM)and calculate its tem-
perature dependence, i.e., dK/dT. Then, the variations of Kand
dK/dTwith respect to the waveguide dimensions are calculated;
these are shown in Figs. 2(c) and 2(d), respectively. Note that
we ignore the influence of the thermal expansions of silicon and
silica on the waveguide dimensions since these expansions are
extremely small (<0.019% with a maximal temperature change
of 60 K [27]). Clearly, both a larger waveguide width and a larger
waveguide distance will produce larger values of Kand dK/dT,
which meets the design requirements but inevitably compro-
mises the device compactness due to the long coupling lengths.
As a balance, we finally takew=400 nm and wgap =150 nm as the
starting point. With these dimensions, the coupling lengths for
the two polarization beams are calculated to be Lc(TE)=11.33 µm
and Lc(TM)=4.238 µm.
The coupling lengths calculated analytically using Eqs. (1)
and (2) are first verified using a variational FDTD solver (Ansys
Inc.) and then a 3D FDTD solver (Ansys Inc.). The former
method gives Lc(TE)=8.91 µm and Lc(TM)=5.09 µm, showing
discrepancies of 21.4% and 20.1% from the analytical results,
respectively, while the latter method gives Lc(TE)=11.18 µm and
Lc(TM)=4.2368 µm, which disagree with the analytical results
by only 1.3% and 0.028%, respectively. Since the variational
FDTD method produces less accurate results as it ignores the
coupling in the waveguide height direction (y), we use the 3D
FDTD method to determine the Lc(TE)and Lc(TM), in spite of it
being much more time consuming.
In order to achieve the splitting and switching functions, the
total physical length Lof the DC should fulfill the following
before switching:
L=m×Lc(TE)=(m+p) × Lc(TM),(3)
and the following after switching:
L=n×L′
c(TE)=(n+q) × L′
c(TM).(4)
Here, mand nare integers with different parities, pand qare odd
numbers, L′c(TE)and L′c(TM)are the coupling lengths of the TE
and TM polarization beams after switching (e.g., by heating up
the DC). We calculate L′c(TE)and L′c(TM)at different temperatures
and choose a moderate temperature of 360 K for the design, at
Fig. 3. The change in the physical length of the DC, L, with
respect to the coupling times Nfor the two polarizations before and
after heating (lines from left to right, TE at 360, 300, 360 300 K).
The dashed line indicates the best choice of L, 521.2 µm.

Letter Vol. 47,No. 4 / 15 February 2022 /Optics Letters 963
Fig. 4. Bending loss of a silicon 90°-bent waveguide and the
coupling ratio between a pair of silicon 90°-bent waveguides with
a distance of 150 nm for the (a) TE polarization and the (b) TM
polarization.
which L′c(TE)=11.588 µm and L′c(TM)=4.27 µm. Figure 3shows
the physical length Lof the DC as a function of the coupling
times Nfor the two polarization beams before and after heating.
We finally find that L=521.2 µm (indicated by the dashed line
in Fig. 3) produces the best approximation to the required inte-
ger properties for m,n,p, and q, where m=46, n=45, p=77,
and q=77. This corresponds to 46 couplings for TE polariza-
tion before heating, 123 couplings for TM polarization before
heating, 45 couplings for TE polarization after heating, and 122
couplings for TM polarization after heating. Aside from the DC
part, the bent waveguides are also necessary to couple the light
into the DC and decouple the light out of it. Generally, sharply
bent waveguides are preferred due to the resulting device com-
pactness. However, if the bend radius is too small, the bending
loss of the connectors is non-negligible and will degrade the
performance of the PBS. Meanwhile, there is also power cou-
pling between the two adjacent bent waveguides, which should
be taken into consideration when designing the coupler length
in the PBS.
Figures 4(a) and 4(b) show the calculated bending loss of a
90°-bent silicon waveguide and the coupling ratio between a
pair of these waveguides as a function of the bend radius for
TE polarization and TM polarization, respectively. Here, the
waveguide dimensions are the same as that for the DC, i.e.,
w=400 nm and h=250nm, and the wavelength is 1550 nm. It
is observed that the bending loss dramatically increases when
the radius is smaller than 5 µm, especially for TM polarization
due to weaker confinement of light. While a larger radius can
mitigate this loss, it not only enlarges the device footprint but
it adds much more extra coupling, which will complicate the
design of the length of the straight region in the PBS. Thus, we
choose a radius of 5 µm for all the bent waveguides.
Taking the extra coupling of the two pairs of bent waveg-
uides into consideration and using the 3D FDTD method, we
Fig. 5. Light propagation in the designed switchable PBS with
w=400 nm, wgap =150 nm, R=5µm, and L=516.8 µm. (a) TE
input before heating. (b) TM input before heating. (c) TE input
after heating. (d) TM input after heating. Here, the wavelength is
1550 nm.
Fig. 6. SEM image of the fabricated switchable silicon photonic
PBS. Insets show enlarged views of the PBS.
optimize the length of the DC and obtain L=516.8 µm for the
switchable PBS. Figures 5(a) and 5(b) show the simulated light
propagation in the designed PBS for TE (Ex) and TM (Ey) polar-
ization beams input at a wavelength of 1550 nm, respectively.
The splitting function, where the TE beam outputs from the
cross port after coupling 47 times and the TM beam outputs
from the through port (thru port) after coupling 124 times, is
clearly seen. Note that the coupling times for both beams is one
more than that of the DC with only the straight region presented
in Fig. 3, since the bent waveguides also contribute to the cou-
pling. After increasing the temperature by 60 K, the TE and TM
polarization beams couple 46 and 123 times, respectively. This
results in the achievement of the switching function: the TE
and TM polarization beams are output from the thru and cross
ports, which is the opposite to the case before heating, as shown
in Figs. 5(c) and 5(d), respectively. The losses/extinction ratios
(ERs) corresponding to Figs. 5(a) to 5(d) are 0.045 dB/24.9 dB,
0.16 dB/22.5 dB, 0.058 dB/17.8 dB, and 0.1 dB/27.3 dB, respec-
tively. As guidance for fabrication, PBSs with wand wgap
variations of ±5 nm have additionally been simulated, and ER
degradations of >10 dB were seen.
The proposed switchable PBSs were fabricated on a silicon-
on-insulator (SOI) wafer with a 250 nm top silicon layer and
a 3 µm silica insulator layer. The patterns of the PBSs were
first defined on the resist by electron beam lithography and
then transferred to the silicon using reactive ion etching. After
residual resist stripping, the devices were clad with 1 µm silica.
Inverse tapers were also fabricated to assist the coupling between
the waveguides and the fibers. Figure 6shows a scanning electron
microscope (SEM) image of part of the fabricated PBS, and the
upper and lower insets display close-up images of the straight
coupling region and the bent waveguide region, respectively.
Not only were PBSs with the designed dimensions (i.e.,
w=400 nm, wgap =150 nm, and L=516.8 µm) fabricated, but
PBSs with slightly different dimensions to those were also fab-
ricated and measured. This is because the width varies along
the waveguide due to the roughness of the sidewalls, and this
variation is usually ±10 nm. The waveguide gap wgap varies in
a similar manner. Meanwhile, the silicon layer is not exactly
uniform; it has a nominal thickness of 250 nm but is actu-
ally 240 ±5 nm, partially due to natural oxidation. These sorts
of nonuniformities shifted the best measured results obtained
from a fabricated PBS with w=390 nm, wgap =160 nm, and
L=519.8 µm. The starting temperature was set at 28°C – slightly
higher than the room temperature of 22°C – in order to fully
achieve the splitting function at 1550nm, as shown by Figs. 7(a)
and 7(b), which exhibit the measured transmission spectra of the
thru and cross ports and their sum when the TE and TM polar-
ization beams are coupled in the PBS, respectively. Note that the
transmissions were normalized to a straight silicon waveguide
on the same sample. The PBS was characterized and found to
achieve ERs of 17.3dB and 18 dB, with losses of 1.6 dB and

964 Vol. 47,No. 4 / 15 February 2022/ Optics Letters Letter
Fig. 7. Measured and normalized transmission spectra of the fab-
ricated switchable PBS. (a) TE input at 28°C. (b) TM input at 28°C.
(c) TE input at 85°C. (d) TM input at 85°C.
0.18 dB, for the TE and TM inputs at 1550nm, respectively.
The switching function was fully achieved by heating the PBS
by 57°C to 85°C using a feedback temperature control system
comprising a temperature sensor inside the sample holder and a
semiconductor temperature cooler (TEC) beneath the holder.
Figures 7(c) and 7(d) exhibit the corresponding measured
transmission spectra. Now, most of the light at 1550 nm outputs
from the thru port instead of being coupled out from the cross
port for the TE polarization beam, while light primarily outputs
from the cross port for the TM polarization beam, in contrast to
both the TE case at this higher temperature and the TM case at
the previous lower temperature. The losses/ERs are now meas-
ured as 0.59 dB/12.6 dB and 0.6 dB/19.7 dB at 1550 nm for the
input TE and TM polarization beams, respectively, comprehen-
sively verifying the switching function. The slight degradation
in the measurement results compared to the simulations can be
attributed to the roughness of the fabricated waveguides and the
minor impurity of a polarization mode in the silicon waveguides
during the measurements.
In summary, we have proposed and experimentally verified
the concept of a switchable polarization beam splitter. This was,
to the best of our knowledge, the first demonstration of such a
system. The design principle was given. The fabricated switch-
able PBS allows two different polarization beams to output from
different ports before heating but then exchanges the ports upon
heating, simultaneously keeping large extinction ratios (>12 dB)
and low losses (<2 dB). On the one hand, the utilization of sil-
icon photonic waveguides takes advantage of the compactness
of a chip-based device and the large thermal turnability of a
silicon waveguide; on the other hand, it imposes challenges in
relation to fabrication tolerance and agreement with the design.
Nevertheless, this issue may be resolved by, e.g., using silicon
nitride waveguides, which are less sensitive to fabrication imper-
fections, and the proposed design procedure should still work.
All in all, we believe that our proposed concept of a switchable
PBS, and especially its realization on a chip, will create new
opportunities for more flexible signal processing, information
sensing, or quantum signal handling.
Funding. The Local Science and Technology Development Fund Projects
Guided by the Central Government, China (206Z1703G); Villum Fonden
(00023316).
Acknowledgment. The authors thank DTU Nanolab for the support of
fabrication facilities and technologies.
Disclosures. The authors declare no conflicts of interest.
Data availability. Data underlying the results presented in this paper are
not publicly available at this time but may be obtained from the authors upon
reasonable request.
REFERENCES
1. T. Barwicz, M. R. Watts, M. A. Popovi´
c, P. T. Rakich, L. Socci, F. X.
Kärtner, E. P. Ippen, and H. I. Smith, Nat. Photonics 1, 57 (2007).
2. D. Pérez-Galacho, R. Halir, A. Ortega-Moñux, C. Alonso-Ramos, R.
Zhang, P. Runge, K. Janiak, H. G. Bach, A. G. Steffan, and Í Molina-
Fernández, Opt. Express 21, 14146 (2013).
3. M. Tan, X. Xu, J. Wu, T. G. Nguyen, S. T. Chu, B. E. Little, A. Mitchell,
R. Morandotti, and D. J. Moss, J. Semicond. 42, 041305 (2021).
4. X. Guan, P. Chen, S. Chen, P. Xu, Y. Shi, and D. Dai, Opt. Lett. 39,
4514 (2014).
5. H. Guan, Y. Ma, R. Shi, A. Novack, J. Tao, Q. Fang, A. E.-J. Lim, G.-Q.
Lo, T. Baehr-Jones, and M. Hochberg, Opt. Lett. 39, 4703 (2014).
6. A. E. Elfiqi, R. Kobayashi, R. Tanomura, T. Tanemura, and Y. Nakano,
IEEE Photonics Technol. Lett. 32, 663 (2020).
7. D. Dai and H. Wu, Opt. Lett. 41, 2346 (2016).
8. D. Guo, K. Hou, W. Tang, and T. Chu, J. Semicond. 40, 100401
(2019).
9. X. Chen, W. Li, and J. Yao, IEEE Photonics Technol. Lett. 25, 1373
(2013).
10. Y. Zhang and S. Pan, IEEE J. Quantum Electron. 54, 1 (2018).
11. D. Perez-Galacho, R. Zhang, A. Ortega-Monux, R. Halir, C. Alonso-
Ramos, P. Runge, K. Janiak, G. Zhou, H.-G. Bach, A. G. Steffan, and
I. Molina-Fernandez, J. Lightwave Technol. 32, 361 (2014).
12. H. Wang, T. Wu, H. Wang, X. Peng, and H. Guo, Phys. Rev. A 101,
063408 (2020).
13. S. Yin, X. Wang, Z. Wang, S. Xiao, and X. Guan, Sensors 20, 5870
(2020).
14. P. Kok, W. J. Munro, K. Nemoto, T. C. Ralph, J. P. Dowling, and G. J.
Milburn, Rev. Mod. Phys. 79, 135 (2007).
15. L.-T. Feng, M. Zhang, Z.-Y. Zhou, M. Li, X. Xiong, L. Yu, B.-S. Shi, G.-
P. Guo, D.-X. Dai, X.-F. Ren, and G.-C. Guo, Nat. Commun. 7, 11985
(2016).
16. Y. Chen, X.-T. He, Y.-J. Cheng, H.-Y. Qiu, L.-T. Feng, M. Zhang, D.-
X. Dai, G.-C. Guo, J.-W. Dong, and X.-F. Ren, Phys. Rev. Lett. 126,
230503 (2021).
17. J.-W. Pan, Z.-B. Chen, C.-Y. Lu, H. Weinfurter, A. Zeilinger, and M.
˙
Zukowski, Rev. Mod. Phys. 84, 777 (2012).
18. X. Liu, D. Liu, and D. Dai, Opt. Express 29, 2720 (2021).
19. C. Li and D. Dai, J. Lightwave Technol. 36, 2129 (2018).
20. Z. Lin, K. Chen, Q. Huang, and S. He, IEEE Photonics J. 13, 1 (2021).
21. T. Yu, J. Huang, N. Liu, J. Yang, Q. Liao, and X. Jiang, Appl. Opt. 49,
2168 (2010).
22. L. Xu, Y. Wang, E. El-Fiky, D. Mao, A. Kumar, Z. Xing, M. G. Saber,
M. Jacques, and D. V. Plant, J. Lightwave Technol. 37, 1231 (2019).
23. C. Li, M. Zhang, J. E. Bowers, and D. Dai, Opt. Lett. 45, 2259
(2020).
24. J. Wang, L. Pei, S. Weng, L. Wu, L. Huang, T. Ning, and J. Li, IEEE
Photonics J. 9, 1 (2017).
25. S. Zhu, Y. Liu, L. Shi, X. Xu, S. Yuan, N. Liu, and X. Zhang, Opt.
Express 24, 17511 (2016).
26. H. Zhou, R. Ma, S. Zhu, H. Chen, G. Zhang, L. Shi, and X. Zhang,
Opt. Express 28, 32847 (2020).
27. H. Nejat Pishkenari, E. Mohagheghian, and A. Rasouli, Phys. Lett. A
380, 4039 (2016).