Optics Communications 245 (2005) 137–144

www.elsevier.com/locate/optcom

Fabrication and analysis of 2 · 2 thermo-optic SOIwaveguide switch with low power consumption and

fast response by anisotropy chemical etching

Jingwei Liu *, Jinzhong Yu, Shaowu Chen, Jinsong Xia

State Key Laboratory on Integrated Optoelectronics, Chinese Academy of Sciences, Institute of Semiconductors, Hai-Dian,

P.O. Box 912, Beijing 100083, PR China

Received 2 May 2004; received in revised form 21 September 2004; accepted 4 October 2004

Abstract

A low power consumption 2 · 2 thermo-optic switch with fast response was fabricated on silicon-on-insulator by

anisotropy chemical etching. Blocking trenches were etched on both sides of the phase-shifting arms to shorten device

length and reduce power consumption. Thin top cladding layer was grown to reduce power consumption and switching

time. The device showed good characteristics, including a low switching power of 145 mW and a fast switching speed of

8 ± 1 ls, respectively. Two-dimensional finite element method was applied to simulate temperature field in the phase-

shifting arm instead of conventional one-dimensional method. According to the simulated result, a new two-

dimensional index distribution of phase-shifting arm was determined. Consequently finite-difference beam propagation

method was employed to simulate the light propagation in the switch, and calculate the power consumption as well as

the switching speed. The experimental results were in good agreement with the theoretical estimations.

� 2004 Elsevier B.V. All rights reserved.

PACS: 42.82.Ds; 4282.Et; 42.82.�mKeywords: Thermo-optic switch; Silicon-on-insulator; Anisotropy chemical etching; Finite-element method; Finite-difference beam

propagation; Switching time; Power consumption

0030-4018/$ - see front matter � 2004 Elsevier B.V. All rights reserv

doi:10.1016/j.optcom.2004.10.004

* Corresponding author. Tel.: +861082304511; fax:

+861082305052.

E-mail address: jwliu@red.semi.ac.cn (J. Liu).

1. Introduction

Optical switch is one of the key devices for opti-

cal communication systems. It plays an important

role in the applications including optical cross con-

nection (OXC) and optical add-drop multiplexing

ed.

Fig. 1. Schematic structure of the 2 · 2 TO MZ-MMI switch.

138 J. Liu et al. / Optics Communications 245 (2005) 137–144

(OADM) systems [1]. Planar waveguides technolo-

gies, based on silica-on-silicon [2–4] and silicon-

on-insulator, are very popularly employed for

optical switching application, and some switches

with small port count have been fabricated [5–8].A 2 · 2 MZ optical waveguide switch often

consists of two paired couplers (i.e. multi-mode-

interference (MMI) couplers or directional inter-

ference couplers) joined by two phase-shifting

arms. The couplers act as 3 dB splitter and comb-

iner. With silicon, there are two possible choices

for phase-shifting: thermo-optical (TO) effect and

free carriers (FC) dispersion effect. The FC effectrequires high free carrier injection to reduce index

significantly [9]. Unfortunately, such high current

density contributes to rising of temperature that

will increase the refractive index. Therefore, the

effect of carrier injection will be weakened seri-

ously. Additionally, the free carriers inevitably

induce considerable absorption loss. Therefore, a

good choice for phase-shifting is TO effect sincesilicon has high thermo-optical coefficient and

thermal conductivity.

Fischer et al. [6] have fabricated 2 · 2 thermo-

optical switch, which showed an 85 mW switching

power, but no results about the switching speed

were reported. In 2003, House et al. [5] realized

high speed 2 · 2 thermo-optical switch with a

10-ls switching time, but the device showed a highswitching power of 440 mW. In order to get

smoother interface, we fabricated 2 · 2 thermo-

optic switch recently by anisotropy chemical

etching in stead of conventional dry-etching [8].

However, in the reported device, switching speed

is rather low (about 60 ls) [8].Furthermore, it is important to picture temper-

ature field in the phase-shifting arm accurately. Aone-dimensional (1D) method can be used to esti-

mate the temperature simply [10]. But the SOI

waveguide often has a large area rib-shape

cross-section in order to achieve a high coupling

efficiency. On the other hand, the waveguide will

have a trapezoidal cross-section instead of a regu-

lar rectangular one if the wet etching method is

adopted. Then, the 1D method is not suitablefor describing the temperature field and determin-

ing switching speed as well as power consumption

accurately.

In this paper, a 2 · 2 thermo-optic switch with

MMI couplers was fabricated on SOI wafer. A

bonding and back-etching SOI (BE-SOI) wafer

with a 5-lm top silicon layer and a 1-lm buried sil-

icon oxide was used here. Anisotropy chemicaletching was used similar with [8]. Blocking

trenches were etched near the phase-shifting arms

to shorten the device length and reduce the power

consumption. A top cladding layer is thin enough

to reduce power consumption and switching time.

The device achieved good characteristics including

the low switching power of 145 mW and the fast

switching speed measured to be about 8 ± 1 ls.These values are much better than our former

results [8].

Two-dimensional finite element method (2D-

FEM) has been chosen for temperature field simu-

lation to overcome the drawbacks of the previous

1D method. According to this temperature field, a

new 2D index distribution of the phase-shifting

arm was obtained, and then, finite-difference beampropagation method (FD-BPM) was used to simu-

late its light propagation. Meanwhile, the power

consumption and the switching speed were

determined.

2. Design

Fig. 1 shows a structure of the proposed 2 · 2

TO-MZ switch with MMI splitter and combiner.

It consists of four input/output ports with S-bend

connectors, two MMIs, two straight phase-shifting

arms, two heaters, and some blocking trenches.

Considering single-mode condition for SOI rib

waveguide with trapezoidal cross-section [11], the

J. Liu et al. / Optics Communications 245 (2005) 137–144 139

width of single-mode waveguides is 5 lm, and the

etching depth is designed to be 1.5 lm. The MZ

arms and the heaters are 4500 and 3380 lm long,

respectively. S-bends are used to enlarge the dis-

tance of neighbor ports to 127 lm. Radius of theS-bend is set at 50,000 lm to ensure low bend radi-

ation loss. In general, self-images are bigger than

the original fields in MMIs. In order to reduce en-

ergy loss and obtain large fabrication tolerance,

tapers are used at the end of the MMIs to connect

the following single-mode waveguides.

The MMIs are connected by two straight phase-

shifting arms instead of conventional S-bends toshorten the device length. To avoid mode cou-

pling, 5 lm wide blocking trenches are introduced

on both sides of each arm, as shown in Fig. 1. An-

other important advantage of using trenches is to

isolate heat transfer between the two arms. Such

transfer will reduce switching power greatly, which

was proved by our results of simulations and

experiments.MMI coupler is based on the self-imaging effect

of multi-mode waveguide [12]. There are three

kinds of interference in MMI, which are general,

paired, and symmetric interference. In this paper,

the general interference is utilized.

3. Analysis and simulation

A cross-section of the phase-shifting arm is

shown in Fig. 2. The arm is structured as follows:

A buried silicon oxide layer was grown by thermal

Fig. 2. Cross-section of the phase-shifting arm.

oxidation on a silicon substrate. Subsequently, a

silicon layer was bonded on the buried oxide layer.

The waveguides and the MMIs were formed on

this bonded silicon layer. Dry thermal oxidation

was applied to form a 0.12-lm top cladding layer.As heaters, aluminum film with a thickness of 0.5

lm were evaporated and patterned. For getting

smoother interface, the waveguides aligned to

Æ110æ-crystal direction were etched by KOH. Side-

wall angle theta of the trapezoidal cross-section is

54.74�, which arises from chemical etching [13].

Both steady and transient temperature fields in

the heated arm are analyzed by 2D-FEM basedon the cross-section shown in Fig. 2. The 2D mod-

el is sufficient for simulation because the arm�slength is much greater than its width. What we

cared is not absolute temperature, but temperature

difference. Such difference is almost independent

from environment temperature. Therefore, it is

reasonable to assume the boundaries of the section

are Direchlet ones and the environment tempera-ture is 0 K. With such assumptions, the calculated

temperature is just equal to the temperature differ-

ence that is requisite for the simulation. Table 1

summarizes data of material used for calculation.

When an electrical current is applied to one of

the heaters, the optical phase change in the heated

arm can be expressed as

DU ¼ onoT

LDT2pk; ð1Þ

where (on/oT) (= 1.86·10�4 K�1) is the thermo-op-

tic coefficient of silicon. DT is the temperature

change over the heater length L. The temperature

difference which is needed to induce a p shift in thearm can be given by

DT ¼ k2Lon=oT

: ð2Þ

Table 1

Material properties

Material Density

(kg/m3)

Specific heat

(J/gK)

Thermal conductivity

(W/mK)

Si 2330 752 140

SiO2 2210 89 1.38

Al 2700 864 240

Air 1.29 1007 0.026

Fig. 4. The cross-output versus time.

Fig. 5. Distribution of index difference in the heating arm.

140 J. Liu et al. / Optics Communications 245 (2005) 137–144

The cross-output intensity I of the switch as a

function of temperature difference can be ex-

pressed simply as [4]

I ¼ 1� sin2 p � on=oT � DT � Lh

k

� �: ð3Þ

According to the formulas and structure

parameters, the temperature difference of 1.2 K isrequired to realize a switching operation from a

cross state to a bar state. The steady temperature

distribution at a heating power of 140 mW is illus-

trated in Fig. 3. It demonstrates that the tempera-

ture difference of cross-section center area is

around 1.2 K and the trenches isolate the thermal

diffusion effectively. A transient temperature field

was also analyzed to determine the center temper-ature as a function of time. Then according to (3),

we could curve the response of the optical intensity

against time in a heating-and-cooling period as

shown in Fig. 4. Considering that the upper and

lower traces correspond to the heating and cooling

steps respectively. Therefore, the ‘‘fall’’ time and

‘‘rise’’ time of optical intensity denote the rise time

and fall time of the switch. In Fig. 4, the two val-ues are estimated to be about 8 and 7 ls,respectively.

Then FD-BPM is employed to simulate the

light propagation in the arms and the MMIs. Un-

like conventional process, index distribution of

phase-shift arm is no longer uniform. A 2D distri-

bution of the index change shown in Fig. 5 is pic-

tured according to the temperature distribution

Fig. 3. Temperature distribution in the phase-shifting arm

when steady-state thermal analysis is applied.

shown as Fig. 3. Simulation results of BPM are

shown in Fig. 6(a)–(d). From these results, we

can determine the optimal length of the MMIs.

The top width of the MMIs is 30 lm. The freespace wavelength is 1550 nm. Fig. 6(a) and (b)

show that the optimal length of the first MMI is

5220 lm for two images. Fig. 6(c) shows one good

image at the end of the second MMI with the same

length of 5220 lm when the switch works at cross

state. When one of the arms is heated at heat

power of 140 mW, the switch will work at bar

state, and as shown in Fig. 6(d), the image appearsat the opposite side, which proves the validity of

the thermal analysis. Small peaks lie in the middle

area of the waveguide (see Fig. 6(b)–(d)), which

will induce additional energy loss and degrade

the extinction ratio. This problem can be partly

Fig. 6. Simulated self-imaging in the MMI: (a) input field at the entrance of the first MMI; (b) two images at the end of the first MMI;

(c) one image at the end of the second MMI while the switch works at the cross state and (d) one image at the end of the second MMI

while the switch works at the bar state and one of arms is heated. The length and the width of MMI are 5220 and 30 lm, respectively.

J. Liu et al. / Optics Communications 245 (2005) 137–144 141

solved by widening the waveguide and choosing

paired interference.

Fig. 7. Output near-field profile of the cross state (a) and the

bar state (b).

4. Experiment and results

The fabrication process is shown as: first, a 0.15

lm thick SiO2 mask film was grown thermally,then the waveguides patterns whose direction were

paralleled to the Æ110æ orientation of the wafer

were defined with photolithography technology.

The rib waveguides were etched to a depth of 1.5

lm by saturated KOH solution. Then another

0.15 lm thick SiO2 film, acted as the mask of the

second chemical etching for the trenches, was

grown by thermal oxidation. After patterning themask by photolithography, the wafer was etched

to the buried SiO2layer by KOH solution. Then,

a 0.12-lm thick SiO2 film that acted as the top

cladding layer was grown thermally. Metal heater

was evaporated and patterned on the two phase-

shifting arms. Finally, the end facets of the wave-

guides were polished carefully.

The characteristics of the device were measuredby coupling laser from a single-mode fiber directly

to the device itself. Output near-fields were moni-

tored by a CCD system. Fig. 7(a) and (b) show the

output near-fields when the switch worked at cross

and bar states, respectively. In Fig. 8, solid and

dashed curves show the basic switching character-

istics measured for two orthogonal polarization

states at 1550 nm. As denoted, the switching

1530 1535 1540 1545 1550 1555 1560 1565 1570-18.0

-17.5

-17.0

-16.5

-16.0

-15.5

-15.0

-14.5

-14.0

-13.5

-13.0

-12.5

-12.0

Inse

rtio

n Lo

ss (

dB)

Wavelength (nm)

Ref 0 DegreeRef 45 DegreesRef -45 DegreesRef 90 Degrees

Fig. 9. Wavelength dependence of insertion loss on transmis-

sion maximum of switch.

0 20 40 60 80 100 120 140 160 180 200 220 240-20

-18

-16

-14

-12

-10

-8

-6

-4

-2

0

2

14.5 dB

145 mw

143 mw

Out

put P

ower

(dB

,a.u

.)

Electrical Power (mw)

Fig. 8. The output power of the two output ports versus

heating power.

142 J. Liu et al. / Optics Communications 245 (2005) 137–144

powers are 143 and 145 mW, respectively, lessthan the value of 235 mW reported in [8]. We have

fabricated a similar switch without trench and the

thickness of top cladding was about 0.8 lm. In

this device, the switching power consumption

was 420 mW [14]. These prove that the blocking

trenches and the thin top cladding layer are effi-

cient to reduce switching power. Due to limit of

technique, the two arms could not be made thesame, so initial powers about 20 and 46 mW,

respectively, are needed for the best extinction

ratio.

The extinction ratios are 18.5 and 15.1 dB,

respectively. However, the polarization-dependent

offset between the two switch transmission mini-

mas limits the extinction ratio to 14.5 dB. Rea-

sons for such low extinction ratios may bederived from the thickness unevenness of the top

layer and fabrication error. The thickness fluctua-

tion is about ±0.5 lm which is severe comparing

to the total thickness of 5 lm. For the same

SOI wafer, wider MMI will produce the self-im-

ages with higher quality because of more modes

being excited, but the size of device will be

enlarged at the same time.The insertion loss was measured by a fiber–

chip–fiber system. End facets of fibers did not

touch the chip directly, so there were air gaps be-

tween the chip and the fibers. Therefore the system

was a fiber–air–chip–air–fiber one in fact. The sys-

tem reflections included two fiber–air reflections

and two air–chip reflections.

The measured insertion loss of a straight wave-

guide, which is 3 cm long and 5 m wide, is �10 dB

at 1550 nm. The total reflection loss and coupling

loss, for both input and output, were calculated of

�3.5 and �3.4 dB, respectively. The propagationloss of the straight waveguide could be estimated

to be �1 dB/cm. The calculated results are ideal

compared with the experiment ones. So the propa-

gation loss can only be less than �1 dB/cm. The

insertion loss of the 2 · 2 switch is �14 dB. Com-

pared with the value of the straight waveguide, a

surplus loss of �4 dB should derive from energy

radiation in the S-bend connectors and the imper-fect interference because of design and fabrication

errors in MMIs.

Interferometric devices are susceptible to

changes in insertion loss and extinction ratios

as functions of wavelength. Fig. 9 shows the

wavelength dependence of insertion loss when

different polarization incident lights were used.

It is seen that a ripple of ±0.8 dB about an aver-age value of �14 dB loss at the C band (1530–

1570 nm). Across this band, the extinction ratio

limited by polarization-dependent offset changes

within a range of 13.9–14.5 dB according meas-

ured results. However, when the wavelength

increases to 1580 nm or more, the insertion loss

and the extinction ratio will deteriorate

significantly.

J. Liu et al. / Optics Communications 245 (2005) 137–144 143

The polarization dependent losses (PDL) were

also measured at different wavelengths. In the

worst case across the whole C band, the PDL

was 1.56 dB, and the best value of 0.56 dB was

attributed at 1550 nm.There are two reasons for which switching char-

acteristics are susceptible to the polarization of

incident light: (1) asymmetry in the layout of the

metal tracking to the device. This will cause a

non-uniform stress field across the arms [5]; (2) dif-

ference of thermal expansion coefficient between

silicon and silica. This difference will cause a resid-

ual stress during heating process, such as thermaloxidation and metal evaporating.

Because of the high heat conductivity of silicon,

SOI thermo-optic switch has a much faster re-

sponse speed, in comparison with silica or polymer

waveguide switches whose switch time was 180 lsand 5 ms [4,15], respectively. Fig. 10 shows the re-

sponse of cross-output against time when a square

wave pulse was applied to one of the heaters. Therise time and fall time were measured to be about

8 ± 1 and 5 ± 1 ls, respectively. These speeds are

much faster than the former results [8]. The meas-

ured results are in good agreement with theoretical

estimations. As shown in Fig. 2, the generated heat

by heater will be conducted to cladding layer first.

The heat conductivity of silicon is about 100 times

greater than SiO2s. Therefore, reducing the thick-ness of this SiO2 layer, the speed of heat conduct-

ing will be increased regardless the arm is heated

or cooling. The lower switching power and the

Fig. 10. Switching characteristic for the switch.

shorter switching time prove that a thinner top

cladding layer is advantageous to both power con-

sumption and speed.

The improved researches by FEM implied that

if the thickness of buried SiO2 increased, the powerconsumption would reduce obviously, but the

speed would also slower remarkably. Simulation

results show that, for the same structure, doubling

the buried layer to 2 lm would lower the power to

about 87 mW, but meanwhile, it would increase

the switching time to about 50 ls.

5. Conclusion

We have fabricated low power 2 · 2 thermo-op-

tic switch with fast response on SOI by anisotropy

chemical etching. Blocking trenches were etched

on both sides of the phase-shifting arms. A thin

top cladding SiO2 film was thermally grown. The

device obtained superior characteristics includinga low switching power of 145 mW and a fast

switching speed measured to be about 8 ± 1 ls.2D-FEM was applied to simulate the temperature

field of the phase-shifting arm. Then, FD-BPM

was used to simulate the light propagation in the

switch. Good agreement between the results of

measurement and simulation was achieved.

Acknowledgement

This work was supported in part by the Na-

tional Science Foundation of China (Grant No.

69896260) and The Ministry of Science and Tech-

nology ‘‘973’’ plan (No. G2000-03-66) and ‘‘863’’

plan (No. 2002AA312060).

References

[1] G.I. Papadimitriou, J. Lightwave Technol. 21 (2003) 384.

[2] T. Goh, A. Himeno, M. Okuno, H. Takahashi, K. Hattori,

J. Lightwave Technol. 17 (1999) 1192.

[3] R. Kasahara, M. Yanagisawa, A. Sugita, T. Goh, M.

Yasa, A. Himero, S. Matsui, IEEE Photon. Technol. Lett.

11 (1999) 1132.

[4] Q. Lai, W. Hunziker, H. Melchior, IEEE Photon. Technol.

Lett. 10 (1998) 681.

144 J. Liu et al. / Optics Communications 245 (2005) 137–144

[5] A. House, R. Whitemen, L. Kling, S. Day, A. Knighs, D.

Hogan, F. Hopper, M. Asghari, OFC 2 (2003) 449.

[6] U. Fischer, T. Zinke, K. Peternann, IEEE Int. SOI Conf.

(1995) 141.

[7] G.V. Treyz, Electron. Lett. 27 (1991) 118.

[8] J. Xia, J. Yu, Opt. Commum. 232 (2004) 223.

[9] R.A. Soref, B.R. Bennett, IEEE J. Quantum Electron. 23

(1987) 123.

[10] U. Fischer, T. Zinke, B. Schuppert, Electron. Lett. 30

(1994) 406.

[11] J. Xia, J. Yu, Opt. Commum. 26 (2004) 253.

[12] L.B. Soldano, C.M. Pennings, J. Lightwave Technol. 13

(1995) 615.

[13] H. Seidel, L. Cseprege, A. Heuberger, J. Electrochem. Soc.

137 (1990) 3612.

[14] X.L. Wang, J.W. Liu, Q.F. Yan, S.W. Chen, J.Z. Yu,

Chinese Opt. Lett. 1 (2003) 527.

[15] M.H. Lee, Y.H. Min, S. Park, IEEE Photon. Technol.

Lett. 14 (2002) 615.