Thermal-Optic Switch by Total Internal Reflection of ... - NTU

348

IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 12, NO. 2, MARCH/APRIL 2007

Thermal-Optic Switch by Total Internal Reflection of Micromachined Silicon Prism T. Zhong, X. M. Zhang, Member, IEEE, A. Q. Liu, Member, IEEE, J. Li, C. Lu, Member, IEEE, and D. Y. Tang

Abstract—This paper reports the conceptual design and experimental demonstration of an optical switch that utilizes the thermooptic effect (TOE) and total internal reflection (TIR) of a micromachined silicon prism. The key idea is to change the refractive index of the prism via TOE to switch the light from the transmission state to the TIR state. The structure is fabricated by microelectromechanical systems (MEMS) technology on a silicon-oninsulator wafer. It requires a temperature change of 69 K to switch from the transmission to the reflection states, which measure isolations of 15.6 and 40.1 dB, respectively. This design is advantageous over the other waveguide switches and photonic crystal devices in the aspects of the absence of beam splitting, tremendously enhanced sensitivity of switching to small change in refractive index, high compactness, and potentially fast and low power switching. Index Terms—Microelectromechanical systems (MEMS), optical switch, thermo-optic effect (TOE), total internal reflection (TIR).

I. INTRODUCTION CTIVE optical switching plays a vital role in turning optical network into reality. Over the years, researchers have never ceased to look for a compact, reliable, low-power, and fast-speed solution for routing optical signals. A simple and effective idea is to physically alter the propagation of light beam by the mechanical translation or rotation of a reflector (mirror or grating), marking the most general mechanism in the optical microelectromechanical systems (MEMS) [1]–[3]. Such systems have successfully miniaturized and integrated active light switching components onto a single chip; however, they do not promise an auspicious future in practical optical communications due to the limitations of mechanical stability and milliseconds-scale response time. An improved idea is to switch the light based on some ultrafast mechanisms (such as cross-phase modulation, Kerr effect, etc.) rather than mechanical movement, while the MEMS serves for fine adjustment of the working conditions [4], [5]. In this light, a focus has been shifted to the new working principles based on light–matter interaction and tuning of the material optical properties. Since silicon has, for decades, played an unchallenged role in the

A

Manuscript received September 18, 2006; revised xxxx. This work was supported in part by the Agency for Science, Technology and Research (A*STAR) under Grant 042 108 0095. The work of X. M. Zhang was supported by the Singapore Millennium Foundation (SMF). T. Zhong, X. M. Zhang, A. Q. Liu, and D. Y. Tang are with the Division of Microelectronics, School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798 (e-mail: [email protected]; [email protected]; [email protected]). J. Li is with the Institute of Microelectronics, Singapore 117685 (e-mail: [email protected]). C. Lu is with the Department of Electronic and Information Engineering, Hong Kong Polytechnic University, Kowloon, Hong Kong (e-mail: enluchao@ ntu.edu.sg). Digital Object Identifier 10.1109/JSTQE.2007.893111

integrated circuit industry with its highly matured fabrication technology, a recent trend shows a rise in utilizing the photonic potential of silicon material [6]–[10]. With the more established theories in the electrooptic effect [11] and thermo-optic effect (TOE) [12] of silicon material, fast modulation of the silicon dielectric constant can be achieved by heating, applying electric field, or applying high-power lasers [4], [5], [13]–[15]. At present stage, thermo-optical approach seems to win an edge over the others due to its easy engineering implementation and relatively large thermo-optic coefficient [16]. The Mach–Zehnder interferometer (MZI) is perhaps the most extensively studied thermo-optic switch so far. Such a configuration has been demonstrated to be feasible to obtain high-speed, low-power, and substantial miniaturization [17]. However, the continuous phase-shifting nature of MZI limits the extinction ratio to approximately 20 dB and, thus, compromises its application as an optical switch that usually requires abrupt physical changes for high sensitivity. Recognizing these nonidealities inherent in the MZI design, an effort has been put in searching for physics of more distinctive optical transitions, especially in the emerging area of photonic crystal (PhC). It has been proposed that by crossing the edge of the photonics band gap, a sudden change of light propagation characteristic in the PhC can be anticipated [14]. Thus, modulation of light becomes realizable by tuning the position of the cutoff frequency of the PhC. Unfortunately, a recent study shows that it requires a temperature increase as high as 625 ◦ C in order to obtain an effective switching of the probe signal [18]. In this paper, we report the thermo-optic switching based on total internal reflection (TIR) in a micromachined silicon prism. The switching is realized by the abrupt and highly sensitive light transition between the transmission and the TIR near the critical angle. In comparison to the MZI designs, such prism prototype possesses a one-to-two switching characteristic and allows direct interface with optic fibers in a compact structure. Furthermore, optical switching can be achieved with low temperature change. In the following structure of the paper, a conceptual illustration of the working principle will be elaborated in Section II, followed by thorough optical analyses in Section III to account for Gaussian beam divergence, Fabry–Pe´rot (FP) effect, and various losses factors. Finally, the fabrication and experimental results will be presented along with some discussions. II. CONCEPT AND DESIGN A. Working Principle The principle of the thermal-optic switch is illustrated in Fig. 1, in which light switching is realized by the TIR at a

1077-260X/$25.00 © 2007 IEEE

ZHONG et al.: THERMAL-OPTIC SWITCH BY TIR OF MICROMACHINED SILICON PRISM

349

a useful model based on thermal expansion and temperature variation of the excitonic band gap was proposed to physically account for the silicon TOE. By assuming a single dominant electronic oscillator with a constant oscillator strength, the thermo-optic coefficient can be expressed as [12] n2 − 1 1 2 dEg dn = (1) −3kex − dT 2n Eg dT 1 − (E/Eg )2

Fig. 1. Working principle and configuration of the optical switch realized by the abrupt transition of TIR at the second silicon–air interface. (a) Transmission state at room temperature, while a mirror is used to change the output light direction for easy coupling. (b) TIR state upon heating.

silicon–air interface (i.e., the second surface of the prism). A micromachined silicon prism is used as the host for light refraction and reflection. The input light is designed to be normally incident on the first surface of the prism. In the initial state, the input light is chosen to make a subcritical angle θ (slightly less than the critical angle) at the second surface so that the light is transmitted into output 1 (e.g., transmission state), as shown in Fig. 1(a). Thereafter, the temperature of the prism is raised by heating the prism uniformly and, thus, the prism refractive index n will be increased owing to the TOE of silicon material, which, in turn, decreases the critical angle θc as determined by θc = arcsin (1/n). As a result, the critical angle becomes lower than the incident angle, and the incident light is totally reflected into output 2 (e.g., TIR state), as shown in Fig. 1(b). In this way, it realizes the light switching function. According to the geometry, the prism is an isosceles triangle with the acute angle equal to α. In the ideal case of normal incidence at the first surface of the prism, the incident angle θ at the second surface is also equal to α. In a practical implementation, as the transmitted light is quite close to the second surface, it has no space to place the fiber for output coupling. For this reason, a mirror can be placed next to the prim to steer the transmitted light, as demonstrated in Fig. 1(a). B. TOE of Single-Crystalline Silicon Material In this design, modulation of the prism refractive index by changing temperature plays a key part for optical switching. Thus far, the TOE of single-crystalline silicon has been both experimentally measured and theoretically modeled [12], [16], [19]–[21]. It was found experimentally that the thermo-optic coefficient of single-crystalline silicon is independent of sample doping and crystal orientation from room temperature to 550 K at the wavelength around 1523 nm [16]. This feature brings in a convenience that one switch design does not need to change even if the fabrication uses silicon wafers of different crystalline orientations and different doping levels. With reasonable agreements on the various thermo-optic coefficient measurements,

where Eg is the excitonic band gap at a critical point where the aforementioned oscillator resides and kex is the silicon thermal expansion coefficient. The expression of kex over the range of 120–1500 K can be readily known [22] as kex (T ) = 3.725 × 10−6 {1 − exp[−5.88 × 10−3 (T − 124)]} + 5.548 × 10−10 T , whereas the parameter Eg (T ) = 4.03 − 3.417T 2 /(T + 439) is quoted from the previous experimental fitting [23]. With an initial value n = 3.42 for silicon at room temperature, the thermo-optic coefficient at wavelength 1550 nm over the temperature range of 293–600 K can be well fit by a quadratic polynomial dn = −1.615 × 10−10 T 2 + 3.156 × 10−7 T + 8.919 × 10−5 . dT (2) With this equation, the relationship between light switching and temperature can be established. For conceptual illustration, it is assumed that the incident light is a perfectly collimated beam and the prism–air interfaces are perfectly smooth. If the initial incident angle is smaller than the critical angle, the refractive index change ∆n required to divert the refracted light to the TIR state can be expressed as 1 − n0 (3) sin α where n0 is the refractive index of the prism at room temperature. It is readily seen that the angle α should preferably approximate the room temperature critical angle θc0 as α = θc0 − ∆α in order to achieve a sensitive switching. If n0 = 3.42, θc0 is calculated as 17.0◦ . According to (2) and (3), the required temperature increase to achieve light switching under various incident angles is plotted in Fig. 2. At the critical angle, the incident light is in the critical state for light switching. With the increase of the deviation of incident angle from the critical angle, it requires a higher refractive index change and correspondingly a large temperature change. For example, at a deviation of 0.05 ◦ , the refractive index is required to increase by 0.010, which corresponds to a temperature change of 56 K. Linear relationship is maintained over the observation region. It is clearly demonstrated that the refractive index change and the corresponding temperature change are increased to quite high values with any significant deviation of incident angle away from the room-temperature critical angle. At ∆α = 0.2◦ , the values are ∆n = 0.04 while ∆T = 320 K, which is too high for practical uses. Such undesired deviation would significantly compromise the effectiveness and sensitivity of the light switching. For this reason, maintaining a very small angle deviation is crucial to the realization of such optical switch. Fortunately, the MEMS technology is good at accurate positioning and can solve this problem easily. ∆n =

350


TABLE I DESIGNED PARAMETERS OF THE SILICON PRISM THERMO-OPTICAL SWITCH

Fig. 2. Required changes of refractive index and temperature for realizing the light switching in response to the deviation of incident light from the critical angle. n.u.: no unit.

Fig. 3.

Tuning mechanism of incident angle using MEMS actuators.

C. Design of Optical Switch In the design of the optical switch, there is another reason that requires the MEMS angular adjustment. Although a configuration can be well designed according to the desired incident angle, it remains a challenge to fabricate it precisely. It is not uncommon that the incident angle is deviated from the design by a fraction of 1 ◦ due to the fabrication uncertainty. Besides, the silicon wafer may have a refractive index different from the designed value due to the doping level and environmental temperature, etc. Therefore, it is necessary to make use of MEMS structure for angular adjustment. In this way, it tolerates the uncertain factors and also builds up the optimal condition for the switch (i.e., adjust the initial angle to be close to the critical state). For this purpose, the optical switch design consists of a pair of MEMS actuators in addition to the key element of prism, as schemed in Fig. 3. The actuators are connected to the corners of the prism through suspension beams, while all are arranged in symmetry. The actuators in Fig. 3 are formed by electrostatic parallel plates, each has a central plate sandwiched between two pads. The central plate can be displaced laterally toward a pad when a potential difference is applied between them. Bidirectional displacement of the central plate can be obtained by simply choosing which pad to add to the potential. When the two actuators introduce the same amount of displacement in the opposite direction, as shown by the dashed lines in Fig. 3, the prism will be rotated by δα without dislocating the central point O. This rotation could be counterclockwise or clockwise if the actuators are bidirectional. Although the incidence condition at

the first surface will also be affected in the presence of angular adjustment, such effect should be negligible since the direction of light is almost unaltered when light is incident at a small angle from air to a much higher index silicon medium. As typical MEMS actuators can provide a lateral displacement on the order of a few microns at an accuracy of submicron, a proper design could allow to tune the incident angle over a fraction of a degree at a resolution of a few seconds. The parameters of this design are listed in Table I. The prism angle is set at the room temperature critical angle as calculated previously. The maximum tuning angle achievable by the actuators is calculated as |∆αmax | = s/(a + l cos α), where the definitions and values of the variables are also listed Table I. III. THEORETICAL ANALYSES OF OPTICAL PERFORMANCE In the following sections, a complete analysis of the light switching characteristics will be presented to account for the influence of Gaussian beam divergence, FP cavity effect, polarization effect, and various optical losses associated with the prism design. The divergence of beam requires a safe angle to obtain high isolation (i.e., low crosstalk) between the output ports; the FP cavity effect determines the best combination of the parameters; the polarization affects the achievable specifications; and the loss factors causes insertion losses. The specifications are listed in Tables I and II. A. Gaussian Beam Divergence The analysis so far assumes a perfectly collimated input light. In reality, however, the incident light from the single mode optic fiber is a Gaussian beam, which exhibits beam divergence along its propagation in the air gap and inside the prism, as illustrated in Fig. 4(a). Therefore, the rays in the input light, when striking the second surface of the prism, will see a spread of angles of incidence. Such beam divergence would affect the choice of


TABLE II OPTICAL LOSSES AT TWO SWITCHING STATES

351

in the other words, large crosstalk) between the two outputs. For a high-isolation optical switching from the transmission to the TIR state, the incident light should initially be less than the critical angle by a certain value (i.e., safe angle) to make sure the majority of rays (in other words, the optical power) is transmitted and, then, the temperature change should be higher enough to let the majority of rays be reflected. For a simple estimation of the safe angle and the corresponding additional amount of refractive index change caused by the Gaussian beam divergence, it is assumed that the prism is high enough in the depth direction and the rays at the second surface have approximately the same curvature and beam waist. For practical uses, it requires an isolation of >40 dB. That is, in the transmission state, the reflected power is less than 10−4 of the incident power, and in the TIR state, the refracted power is also less than 10−4 . The isolation of 40 dB is, thus, set as the target for choosing the working conditions of the prism switch. According to the ABCD law of Gaussian optics [24], the equivalent traveling distance of the incident Gaussian beam from the input fiber facet to the second surface of the prism is Z = Zp /n + Za [refer to Fig. 4(a)]; therefore, the beam radius ω and the curvature radius R of phase wavefront can be expressed as ω = ω0 1 + (Z/zR )2 and R = Z[1 + (zR /Z)2 ], where ω0 is the waist radius of Gaussian beam from the input fiber, zR is the Rayleigh range given by zR = πω02 /λ, and λ is the wavelength. The field distribution of the incident light at the second surface can be expressed as 2 x + y2 U (x, y) = u0 exp − (4) ω2 where u0 = 2/πω 2 is the normalization factor, x is along the surface of the prism, and y is in the vertical direction. The divergent angle ∆θ in the prism horizontal plane is related to x by ∆θ = x/nR. Therefore, the power distribution over the divergent angle is also a Gaussian distribution, as shown in the inset of Fig. 4(b). If the prism can steer all the angles ≤ ∆θ to one output (for example, the transmitted output 1), the rays with divergent angle >∆θ will be reflected to output 2. This part of power P (∆θ) is considered as leaked optical power, which can be expressed as +∞ +∞ dx dy |U (x, y)|2 P (∆θ) = nR ∆θ

Fig. 4. Influence of Gaussian beam divergence on the choice of incident angle. (a) Spreading of incident angle due to Gaussian beam divergence. (b) Leaked optical power beyond a certain divergent angle.

the incident angle and would broaden the range of refractive index and the corresponding temperature change over which a significant modulation of output power could be achieved. As illustrated by an extreme case in Fig. 4(a), in which the primary axis of the incident light is right at the critical TIR angle in the second surface, the upper half of the light is beyond the critical angle and, thus, will be reflected; while the lower half is refracted (e.g., at point B). This causes a low isolation (or

1 1 − Erf = 2

−∞

√

2nR ∆θ ω

(5)

where x is the error function defined by Erf (x) = √ Erf (2/ π) 0 exp(−t2 ) dt. Here, the leaked power actually represented the crosstalk between the two outputs, in other words, the isolation of one output to the other. The calculated relationship between the leaked power (in decibels) and the divergent angle is plotted in Fig. 4(b). The parameters are λ = 1.55 µm, Za = 10 µm, and ω0 = 15 µm, which are the same as those in the experiment. According to Fig. 1 and Table I, Zp = l cos α sin α = 33.34 µm. Therefore,

352


In addition, r2 and t2 are used for the reflection and transmission coefficients at the second surface of the prism. For different polarizations, the expressions of t and t are the same as given by t = 2/(1 + n) and t = 2n/(1 + n), but r, r , r2 , and t2 are different [27]. For the s-polarization (TE polarization, i.e., the electric field is perpendicular to the incident plane), it has r = −r = (1 − n)/(1 + n), r2 = (n cos α − cos φ)/(n cos α + cos φ), and t2 = 2n cos α/(n cos α + cos φ). For p-polarization (TM polarization, i.e., the electric field lies in the incident plane), it becomes r = −r = (n − 1)/(1 + n), r2 = (cos α − n cos φ)/(cos α + n cos φ), and t2 = 2n cos α/(cos α + n cos φ). Here, the expressions are for normal incidence at the first surface. The powers P1 and P2 of two outputs can be calculated by adding up the complex amplitudes of all the light resulting from multiple reflections, yielding P1 =

Fig. 5. Influence of FP cavity effect on the output powers due to the interference of multiple reflections at the air–silicon interfaces.

at the second surface of the prism, ω = 15.01 µm and R = 11039 µm. As observed in Fig. 4, most of the power will be leaked at ∆θ = −0.01◦ . With the increase of ∆θ, the leaked power quickly drops. At ∆θ = 0.05◦ , it is already only −52.5 dB. Therefore, a safe angle of 0.05◦ is large enough to guarantee the targeted isolation of 40 dB. The data are listed in Table I. Here, the incident Gaussian beam should have large radius so as to reduce the divergence. If the normally cleaved Corning SMF-28 single-mode fiber (SMF) is used [25], which has ω0 = 5.4 µm, a safe angle of 0.05◦ can only obtain an isolation of approximately 2.5 dB, too low for practical uses. Such beam size can be readily achieved by the use of lensed fibers [26] or thermal-diffusion expanded core fibers. B. FP Effect and Polarization Effect Light passing through any air–silicon interface will generally undergo a partial reflection. When there are two or more interfaces present in the path of the beam, multiple reflections will be generated, thereby giving rise to interference. A special case will be met for normal incidence and when the switch works in the TIR state, in which the first and third surfaces of the prism form a FP cavity via the reflection at the second surface. When the refractive index and the corresponding optical path length are tuned, the output light intensity is anticipated to vary significantly. To take into account such FP cavity effect, the powers at two outputs need to be formulated based on the Fresnel’s equations [27]. The analytical model of the FP cavity effect is illustrated in Fig. 5, which also defines various terminologies to be used in the formulation. Here, r and t represent the amplitude reflection coefficient and transmission coefficient for light incident from air into silicon, respectively, and r and t refer to the counterparts when light is incident from silicon into air.

P2 =

cos φ t2 t22 4 4 2 2 1 + r2 r − 2r2 r cos δ n cos α 1+

r22 t2 t2 − 2r22 r2 cos δ

r24 r4

(6) (7)

where φ refers to the refraction angle at the second surface and δ denotes the phase difference between any two consecutively reflected lights as given by δ = 2πnl sin 2α/λ, where l and α are the geometry of the prism defined in Fig. 1 and Table I, and n is the refractive index, respectively. The term cos φ/n cos α in (6) is to take into account the cross-sectional area, which is necessary for converting the flux to optical power [27]. It is noted that (6) and (7) are derived for normal incidence at the first surface, the expressions will be more complicated for nonnormal incidence but can be derived in a similar way. The influence of the polarization on the switching function is studied in Fig. 6, in which the transmitted optical power at output 1 and the reflected power at output 2 are plotted against the refractive index change for the s-polarization and the p-polarization cases. The parameters for calculation are normal incidence (i.e., ∆α = 0), λ = 1550 nm, n0 = 3.42, and α = θc0 = 17.0◦ . The choice of parameters is to reach the critical TIR state for ∆n = 0. For the s-polarization in Fig. 6(a), the reflection is always greater than −1.8 dB while the transmission is smaller than −8.8 dB. Here, the powers are normalized relative to the incident power. When ∆n approaches 0 from the negative side, the transmission is turned off but the reflection does not change obviously. This is not preferable for optical switching since otherwise the transmission suffers large insertion loss while the reflection does not exhibit a sharp transition. The ripple on the curves comes from the FP cavity effect due to the change of effective cavity length in response to the refractive index variation. For p-polarization as shown in Fig. 6(b), the reflection and transmission at ∆n = −0.02 are −9.8 and −2.2 dB, respectively. With the increase of ∆n, the reflection is gradually increased while the transmission is reduced. Over the critical TIR state, the reflection goes up to 0 dB while the transmission suddenly drops to a very low value (< −50 dB). Such properties are very useful for the optical switch to obtain very low insertion loss, abrupt power change, and high isolation


Fig. 6. Optical power of different outputs for different polarization states. (a) s-polarization. (b) p-polarization. n.u.: no unit.

between the two outputs. For this reason, the discussion and experiment henceforward are restricted to the p-polarized light. In Fig. 6, the reflected and transmitted power may not add up to 1 since there exist another two outputs. One is the back reflection on the incident side while the other is the refracted light out of the second surface due to the reflection from the third surface (symmetric to the position of output 1). The presence of FP cavity effect causes the complexity of power fluctuation with respect to the changes of wavelength and refractive index. However, it brings in an important advantage that the insertion loss can be reduced to 0 by proper combination of the parameters, as indicated in Fig. 6(b). Without this FP cavity effect, the insertion loss cannot be 0 since the back reflection at the first surface always wastes approximately 30% incident power. It can also be observed from Fig. 6(b) that a refractive index change of 0.010 (from −0.005 to +0.005) is large enough for realizing the optical switching function. This value is the same as that required by the Gaussian beam divergence.

353

Fig. 7. Contours of output power as functions of incident angle and refractive index change. (a) Transmission at output 1. (b) Reflection at output 2. n.u.: no unit.

The aforementioned study is for normal incidence. In real practice, however, perfect normal incidence is least likely ensured. Through the angular adjustment by the MEMS actuators, the amount of deviation ∆α might take a very small but a nonzero value (expectedly within ±0.1◦ ). As the switching characteristics are also functions of ∆α, it is important to take into account such incident angle deviation. The output power is contoured in Fig. 7 with respect to two variables—the refractive index change and the incident angle deviation. The bright part represents high output power while the dark part is for low power. For the transmitted power of output 1 as shown in Fig. 7(a), the high-power region and the low-power region are separated by a diagonal straight line, which represents the critical TIR state. The dark half above the diagonal line corresponds to the TIR state and, thus, output 1 has nearly no power. At a given angle deviation, it requires a minimum amount of refractive index change to switch from high output to low output

354


(i.e., from the transmission state through the critical TIR state to finally the TIR state). For example, if ∆α = −0.05◦ , it should have ∆n ≥ 0.010 (taking ∆n = 0.013 to have a safe margin). For the reflected power of output 2 as shown in Fig. 7(b), highpower region is only a small stripe, i.e., high power output can only be reached by proper combination of the incident angle deviation and the refractive index change. For example, when ∆α = −0.05◦ , as given by the requirement of Gaussian beam divergence, it should have ∆n = 0.013 (correspondingly, ∆T = 69 K) to switch the light from the initial low-power position to the brightest part. Otherwise, it will end up with low output power. The fluctuation in the right-top half is also due to the FP cavity effect. The value of 0.013 for the nonnormal incidence is a bit larger than the value of 0.010, as required by the beam divergence and FP cavity effect in the normal incidence. In summary, to obtain an effective optical switching from output 1 to output 2 (i.e., transmission to reflection), it would require the incident angle to be 0.05◦ below the critical TIR angle at room temperature and, then, an increase of refractive index from 3.420 to 3.433 by heating the prism by 69 K. The parameters are listed in Table I. C. Sources of Optical Losses In this optical switch, there are various loss mechanisms associated with each output. For output 1 in the transmission state, the loss is caused by the Fresnel reflections, fiber coupling loss, and the mirror scattering loss of the second surface. For output 2 in the TIR state, the optical loss comes from the FP cavity effect, fiber coupling loss due to the Gaussian beam divergence, and scattering loss due to surface roughness at the second surface. In this section, the various optical loss factors will be evaluated for an estimation of the device performance as listed in Table II. 1) Loss of Fresnel Reflection: Fresnel reflections are universally present when an optical beam traverses through boundaries of two or more homogeneous media with different refractive indices. In this switch, such losses mainly occur at the surfaces of the prism by single and multiple reflections, as already analyzed in studying the FP cavity effect. The subsequent amount of losses can be extracted directly from Fig. 6. For the p-polarized light as shown in Fig. 6(b), output 1 (transmission output) has an insertion loss of approximately 2.2 dB when the incident light is not close to the critical TIR angle, while output 2 (reflection output) can have no loss at its peal value but is always < 2 dB thereafter. The interface of fiber facet and free space also contributes to the reflection loss, with a value of 0.2 dB per interface. So, in total, there is 2.6 dB Fresnel reflection loss associated with output 1, whereas for output 2, it ranges between 0.4 and 2.4 dB. 2) Loss of Fiber Coupling: In this design, the input beam is injected from a SMF while outputs are collected by the SMF of the same type. The coupling losses can be associated with three kinds of fiber misalignments, namely, longitudinal separation, lateral shift, and angular misalignment. In this optical switch, lateral shift and angular misalignment can be well reduced thanks to the high-resolution lithography and etching

technology. Thus, the insertion loss for each output is solely determined by the optical path length traveled by the input beam. The light from the input fiber to any output should pass through certain lengths in air and inside the prism. The longitudinal separation ∆Z can be expressed as ∆Z = Zair + Zprism /n, where Zair and Zprism are the physical distances of the light path in air and prism, respectively. For output 1 at room temperature, it has n = 3.42, Zprism = 30.3 µm, and Zair = 140 µm, which includes 10 µm from input fiber to the first surface plus 130 µm from the second surface via the mirror to the output fiber (refer to Fig. 1(a)). As a result, it has the separation ∆Z1 = 148.9 µm. For output 2 at the TIR state, it has Zprism = 60.7 µm, Zair = 20 µm (10 µm from the input to the first surface and 10 µm from the third surface to the output fiber) and n ≈ 3.433; hence, the separation ∆Z2 = 37.7 µm. According to the formulae presented by Yuan and Riza [28], the total coupling losses are as small as 0.1 dB and nearly 0 for output 1 and output 2, respectively. In practical application, the lateral and angular alignment cannot be ideal. Assuming it has a lateral shift of 2 µm and an angular misalignment of 0.5 ◦ , the coupling losses rise to 0.6 and 0.4 dB for output 1 and output 2, respectively. It should also be noticed that after the refraction at the second surface, the beam transmitted toward output 1 does not possess a symmetric Gaussian distribution. Thus, the method used here might underestimate the coupling loss of output 1. 3) Loss of Prism Surface Scattering: As the prism is fabricated using deep reactive ion etching (DRIE) process, the prism surfaces are not as smooth as ideal mirrors and, thus, would introduce scattering loss. Such loss is especially large in the second surface since the transmission state has an incident angle very close to the critical TIR angle and the refractive light grazes over the second surface. Therefore, the transmitted power is very sensitive to the surface roughness. The roughness of a surface is described by its root mean square (rms) deviation and the correlation distance. In the etched surface, two types of roughness are present. The first is the high-frequency component of the surface roughness with a correlation length of 0.5 µm. The other type is called bumpiness with a correlation length of 2.5 µm [29]. The rms roughness of the second surface is about 60 nm, and could be as fine as 20 nm in the upper part of the prism after the DRIE process. Such roughness makes the reflection and transmission powers deviate from those of the perfect surface at the TIR condition due to enhanced backscattering and dispersed transmission [30]. By the extinction theory [31], the losses with respect to surface roughness of 60 nm are estimated to be 10.1 and 1.9 dB for output 1 and output 2, respectively. The scattering at the second surface also causes a higher crosstalk since the scattering power provides a strong background. 4) Loss of Mirror Scattering: output 1 has to experience an additional loss in comparison to output 2. It comes from the scattering of the mirror, which is used to direct output 1 into the output fiber, as illustrated in Fig. 1(a). This configuration facilitates the positioning of output 1 fiber but at the expense of the scattering loss. In real device, the mirror is fabricated by deep etching and metal coating; thus, it yields certain surface roughness that serves as light scatterers. Such scattering depends on the surface roughness, incident wavelength, and incident


355

Fig. 9. Closeups of the optical switch. (a) Micromachined silicon prism. (b) Bidirectional electrostatic actuator for fine angular adjustment of the prism.

Fig. 8.

SEMs of the overview of a packaged optical switch device.

angle, as given by [32]

2 4πσ cos θi η = 1 − exp − λ

(8)

where η is the scattering percentage, σ is the rms roughness of the mirror surface, and θi is the incident angle. In the designed structure, the transmitted beam is assumed to impinge on the mirror at an average of 45 ◦ . Therefore, for a surface roughness of 60 nm, the mirror scattering loss, particularly for output 1, is calculated as 0.5 dB. As a summary, the respective losses analyzed earlier are listed in Table II. IV. EXPERIMENT AND DISCUSSION A. Fabrication and Integration The scanning electron micrograph (SEM) of a packaged optical switch is shown in Fig. 8. The device consists of a silicon prism, two bidirectional actuators, a high-reflection mirror (to change the direction of the transmitted light for easy positioning of the fiber of output 1), and four optic fibers (three for input and outputs, the additional one is for alignment monitoring). The dimensions of the prism are listed in Table I. The device has an overall footprint of 5 mm × 4 mm. The fabrication employs DRIE process [33] on silicon-on-insulator (SOI) wafers. The etched silicon layer has a 75 µm depth. Shadow mask is used to coat the mirror surface while keeping the prism untouched. The closeups of the prism and one of the actuators are shown in Fig. 9. As the prism is quite wide for dry release or wet release, an etching process called backside open has to be used to remove the part of the substrate that is right under the prism [34]. As a result, the prism is hung in free space, as can be seen from Fig. 9(a). The removal of the substrate eliminates the possible stiction problem due to dust particles and the substrate’s static charge, which happens easily in the conventional SOI devices that remain in the substrate and have the movable parts only about 2 µm above the substrate. As observed from the sidewalls of the prism in Fig. 9(a), the upper part (from top to about 40 µm

depth) has quite smooth surface while the lower part is rougher. The light spot striking onto the prism from input fiber should be aligned to the smooth range of depth by properly burying the fiber into the etched grooves. To have enough space to place the fibers for input and output 2, the fiber ends have to have a large distance (here 210 µm) from the prism surfaces. To reduce the coupling loss, lensed fibers that have the Gaussian beam waist (radius 15 µm) at 200 µm away from the fiber end are used for the input and outputs. The bidirectional actuator is actually a parallel plate, as shown in Fig. 9(b). The central part is a beam connected to the prism and is generally electrically grounded. It will be attracted to displace when a potential is applied at any one of the two pads. The maximum displacement in any direction is 5 µm for a driving potential of 6.8 V. The stopper is to avoid the contact, thus, the short circuit of the central beam and the pad at too high voltage. A serpentine beam acts as a soft link to connect the actuator to the anchor, which allows a certain amount of extension (or compression) in the axial direction when the beam subject to fabrication stress and temperature change. This feature helps maintain the prism position during the heating and cooling processes. The actuation relationship between the adjustment angle and the driving voltage is as shown in Fig. 10. With the increase of the voltage from 0 to 6.4 V, the rotational angle rises up to 0.15 ◦ continuously. The angle can be adjusted within the range of ±0.3 ◦ at an accuracy of approximately 0.02 ◦ /V. B. Experimental Results This optical switch is experimentally characterized using a laser light source at a power of 0 dBm at wavelength of 1550 nm (Ando AQ4321D). The p-polarization state of the input beam is ensured by a polarization controller. Two optical power meters (Newport 2832-C) are employed to detect the powers at the two outputs. A thermo-electric cooler (TEC) is stuck beneath the optical switch by a thermal conductive tape for precise temperature control. This TEC provides temperature modulation over the range from 5 to 150 ◦ C by pumping in a dc current with proper intensity and polarity. By accurate calibration against the injected current intensity, the actual temperature of the TEC and, thus, the switch can be extracted from the reading of current. For visualization, a microscope with an infrared camera (Hamamatsu C5332) mounted is used to monitor the power distribution by capturing the scattered light.

356


Fig. 11. Infrared snapshots of the different switching states. (a) Initial transmission state at room temperature. (b) TIR state after heating up by 69 K. Fig. 10. Adjustment angle of the prism with respect to the actuation voltage applied on the electrostatic parallel-plate actuator.

The optical switching can be realized by choosing either the TIR state or the transmission state at room temperature as the initial condition. The former would require a decrease in the temperature so as to switch the light from the TIR to the transmission state. However, such cooling is difficult to obtain large temperature decrease in open environment. It is also subject to condensation of water vapors on the prism surfaces, especially in a high humidity environment. To avoid such problems, the device is initially set at the transmission state and, then, it is heated to achieve the TIR state. The initial orientation of the prism is optimized with the aid of bidirectional rotational actuators. In experiment, the prism is heated up to about 92 ◦ C, then dc voltage is applied to the actuators to finely rotate the prism in the direction of increasing incident angle until the first peak is observed in output 2 power. This corresponds to the peak power due to the FP cavity effect, as discussed earlier in Fig. 6(b). After this adjustment, the dc voltage is maintained but the temperature is returned to room temperature. As mentioned earlier, such adjustment makes it convenient to build up the initial condition regardless of the fabrication error and uncertain factors. The choice of temperature 92 ◦ C (i.e., ∆T = 69 K for room temperature, 23 ◦ C in the experiment) is to match the designed working condition, as listed in Table I, and to make sure the safe angle approximates to 0.05 ◦ . The switching is observed by the infrared camera, as illustrated by the snapshots in Fig. 11. The scattering at the interfaces of fiber/air and prism/air indicates qualitatively the light paths, power distribution, and surface roughness. At room temperature, the incident light is transmitted to output 1, as shown in Fig. 11(a). It has strong scattering at the second surface of the prism and the mirror, indicating quite rough surface quality. At the same time, a residue reflection can be observed at the third surface to output 2. Upon heating, the final state in Fig. 11(b) shows almost no power present at output 1 while the light at output 2 is significantly intensified. The experiment was repeated several times and yielded precisely the same results provided

Fig. 12. Measured changes of the transmitted and reflected powers with respect to the increase of temperature. The abrupt transition is smoothened due to the divergence of Gaussian beam and the scattering on the second surface of the prism.

that sufficient time is given for the switch to be cooled down to the initial condition after every heating. The real characterization is carried out by slowly raising the temperature of the prism from room temperature. The power meter readings for both outputs are recorded at varying TEC currents. Through the earlier calibration of TEC, an experiment result in the form of optical powers against temperature change can be plotted as shown in Fig. 12. At room temperature (i.e., transmission state), the transmitted power at output 1 is −17.3 dB (normalized relative to the input power) while the reflected power at output 2 is −32.9 dB. With the increase of temperature, the power of output 1 gradually drops while that of output 2 rises up. As the prism is then heated up by 69 K, it comes to an obvious switching point. The reflected power of output 2 rises up to −6.7 dB while that of output 1 decreased dramatically to −46.8 dB. Further increase of temperature results in lower power at output 1; however, it also reduces the power at output 2 due to the FP cavity effect as shown in


Fig. 6(b). According to Fig. 12, the transmission state and the TIR state have isolations of 15.6 and 40.1 dB, respectively, and insertion losses of 17.3 and 6.7 dB, respectively. The TIR state has better specifications. The measured insertion losses agree reasonably with the analysis of optical losses 13.8 and 4.7 dB. The excess losses might be attributed to the fiber shift induced by the thermal expansion of the MEMS substrate. For output 1, the distorted Gaussian profile in the transverse direction also contributes more loss as pointed out in previous analysis. It is observed in Fig. 12 that the power transition at output 1 exhibits a more smoothened, instead of an abrupt, change as predicted by the calculation in Fig. 6(b). This might largely be due to scattering by the roughness of the second surface of the prism, as the literature has confirmed that with the presence of surface roughness, the distribution of the evanescent components of the scattered transmitted field is broadened and, hence, results in a smooth transition of reflectivity at a dielectric–vacuum interface [31]. The power of output 2 also experiences a slower change than that analyzed in Fig. 6(b). This could be mainly due to the divergence of the incident Gaussian beam, which causes the intermediate state that the power distributed on some angle region is transmitted while the other is reflected. Based on the experimental experience, it is found that the surface roughness is the main cause of the specification degradation in the aspects of high insertion loss, smoothened power transition, and large scattering loss and high crosstalk in the TIR state. Further improvement will focus on the fabrication process for better sidewall smoothness. As it is slow and inconvenient to heat the whole prism, localized heating is under development by patterning electrical heating wires right on top of the prism or by coating electrical conductive transparent films (such as indium–tin oxide) on the prism sidewalls. V. CONCLUSION A silicon prism optical switch has been designed and experimentally demonstrated. The TOE of silicon material is utilized to change the prism refractive index so that the light can be routed into two different output directions, either transmitted or reflected. The application of TIR mechanism serves to enhance the sensitivity of switching to small refractive index change. As a result, an optical switching with isolations of 15.6 and 40.1 dB for the two switching states has been demonstrated with only a temperature increase of 69 K in the fabricated device. The reasonable agreement between the measurement and analysis validates the plausibility of the design. A couple of optimizations can be readily available in the ensuing devices to minimize the optical loss and potentially enables a fast and low-power light modulation. REFERENCES [1] L. Y. Lin and E. L. Goldstein, “Opportunities and challenges for MEMS in lightwave communications,” IEEE. J. Sel. Topics Quantum Electron., vol. 8, no. 1, pp. 163–172, Jan./Feb. 2002. [2] C. Marxer and N. F. de Rooij, “Micro-optomechanical 2 × 2 switch for single-mode fibers based on a plasma-etched silicon mirror and electrostatic actuation,” J. Lightw. Technol., vol. 17, no. 1, pp. 2–6, Jan. 1999. [3] J. Li, Q. X. Zhang, and A. Q. Liu, “Advanced fibre optical switches using deep RIE (DRIE) fabrication,” Sens. Actuators A, vol. 102, pp. 286–295, 2003.

357

[4] V. R. Almeida, C. A. Barrios, R. R. Panepucci, and M. Lipson, “All-optical control of light on a silicon chip,” Nature, vol. 431, pp. 1081–1084, 2004. [5] M. J. Potasek and Y. Yang, “Multiterabit-per-second all-optical switching in a nonlinear directional coupler,” IEEE J. Sel. Topics Quantum Electron., vol. 8, no. 3, pp. 714–721, May/Jun. 2002. [6] G. T. Reed, “Optical age of silicon,” Nature, vol. 427, pp. 595–596, 2004. [7] A. Liu, R. Jones, L. Liao, D. Samara-Rubio, D. Rubin, O. Cohen, R. Nicolaescu, and M. Paniccia, “A high-speed silicon optical modulator based on a metal–oxide–semiconductor capacitor,” Nature, vol. 427, pp. 615–618, 2004. [8] C. E. Png, G. T. Reed, R. M. H. Atta, G. J. Ensell, and A. G. R. Evans, “Development of small silicon modulators in SOI,” Proc. SPIE, vol. 4997, pp. 190–197, 2003. [9] C. K. Tang and G. T. Reed, “Highly efficient optical phase modulator in SOI waveguides,” Electron. Lett., vol. 31, pp. 451–452, 1995. [10] P. Dainesi, A. K¨ung, M. Chabloz, A. Lagos, Ph. Fluckiger, A. Ionescu, P. Fazan, M. Declerq, Ph. Renaud, and Ph. Robert, “CMOS compatible fully integrated Mach–Zehnder interferometer in SOI technology,” IEEE Photon. Technol. Lett., vol. 12, no. 6, pp. 660–662, Jun. 2000. [11] R. A. Soref and B. R. Bennett, “Electrooptical effects in silicon,” IEEE J. Quantum Electron., vol. 23, no. 1, pp. 123–129, Jan. 1987. [12] F. G. D. Corte, M. E. Montefusco, L. Moretti, I. Rendina, and G. Cocorullo, “Temperature dependence analysis of the thermo-optic effect in silicon by single and double oscillator models,” J. Appl. Phys., vol. 88, no. 12, pp. 7115–7119, 2000. [13] X. H. Ma and G. S. Kuo, “Optical switching technology comparison: Optical MEMS vs. other technologies,” IEEE Commun. Mag., vol. 41, no. 11, pp. S16–S23, Nov. 2003. [14] H. W. Tan, H. M. van Driel, S. L. Schweizer, and R. B. Wehrspohn, “Nonlinear optical tuning of a two-dimensional silicon photonic crystal,” Phys. Rev. B, vol. 70, no. 20, pp. 205110-1–205110-5, 2004. [15] S. H. G. Teo, A. Q. Liu, J. B. Zhang, and M. H. Hong, “Induced free carriers modulation of photonic crystal optical intersection via localized optical absorption effect,” Appl. Phys. Lett., vol. 89, pp.091910-1–0991910-3, 2006. [16] G. Cocorulla, F. G. D. Corte, and I. Rendina, “Temperature dependence of the thermo-optic coefficient in crystalline silicon between room temperature and 550 K at the wavelength of 1523 nm,” Appl. Phys. Lett., vol. 74, no. 22, pp. 3338–3340, 1999. [17] M. Harjanne, M. Kapulainen, T. Aalto, and P. Heimala, “Sub-µs switching time in silicon-on-insulator Mach–Zehnder thermooptic switch,” IEEE Photon. Technol. Lett., vol. 16, no. 9, pp. 2039–2041, Sep. 2004. [18] M. T. Tinker and J. B. Lee, “Thermo-optic photonic crystal light modulator,” Appl. Phys. Lett., vol. 86, no. 22, pp. 221111-1–221111-3, 2005. [19] E. Jellison, Jr. and H. H. Burke, “The temperature dependence of the refractive index of silicon at elevated temperatures at several laser wavelengths,” J. Appl. Phys., vol. 60, no. 2, pp. 841–843, 1986. [20] H. H. Li, “Refractive index of silicon and germanium and its wavelength and temperature derivatives,” J. Phys. Chem. Ref. Data, vol. 9, no. 3, pp. 561–658, 1980. [21] M. Bertolotti, V. Bogdanov, A. Ferrari, A. Jascow, N. Nazorova, A. Pikhtin, and L. Schirone, “Temperature dependence of the refractive index in semiconductors,” J. Opt. Soc. Amer. B, vol. 7, no. 6, pp. 918–922, 1990. [22] Y. Okada and Y. Tokumaru, “Precise determination of lattice parameter and thermal expansion coefficient of silicon between 300 and 1500 K,” J. Appl. Phys., vol. 56, no. 2, pp. 314–320, 1984. [23] G. Ghosh, “Temperature dispersion of refractive indices in crystalline and amorphous silicon,” Appl. Phys. Lett., vol. 66, no. 26, pp. 3570–3572, 1995. [24] N. Hodgson and H. Weber, Optical Resonators—Fundamentals, Advanced Concepts and Applications. Berlin, Germany: Springer, 1996. [25] Corning SMF-28 Optical Fiber Product Information, Corning, Inc.. Corning, NY, Apr. 2001, pp. 1–4. [26] Corning OptiFocus Collimating Lensed Fiber Product Information, Corning, Inc.. Corning, NY, Aug. 2006. [27] M. Born and E. Wolf, Principles of Optics, 6th ed. Oxford, U.K.: Pergamon Univ. Press, 1999. [28] S. Yuan and N. A. Riza, “General formula for coupling-loss characterization of single-mode fiber collimators by use of gradient-index rod lenses,” Appl. Opt., vol. 38, no. 15, pp. 3214–3222, 1999. [29] S. M. Lee, W. C. Chew, M. Moghaddam, M. A. Nasir, S. L. Chuang, R. W. Herrick, and C. L. Balestra, “Modeling of rough-surface effects in an optical turning mirror using the finite-difference time-domain method,” J. Lightw. Technol., vol. 9, no. 11, pp. 1471–1480, Nov. 1991.

358


[30] M. Nieto-Vesperinas and J. A. Sanchez-Gil, “Light scattering from a random rough interface with total internal reflection,” J. Opt. Soc. Amer. A, vol. 9, no. 9, pp. 424–436, 1992. [31] J. A. Sanchez-Gil and M. Nieto-Vesperinas, “Light scattering from random rough dielectric surfaces,” J. Opt. Soc. Amer. A, vol. 8, no. 8, pp. 1270– 1286, 1991. [32] M. Nieto-Vesperinas, Scattering and Diffraction in Physical Optics, 2nd ed. Hackensack, NJ: World Scientific, 2006. [33] J. Li, Q. X. Zhang, A. Q. Liu, W. L. Goh, and J. Ahn, “Technique for preventing stiction and notching effect on silicon-on-insulator microstructure,” J. Vac. Sci. Technol. B, vol. 21, no. 6, pp. 2530–2539, 2003. [34] P. Kurzawski, T. Salo, H. Baltes, and A. Hierlemann, “Towards a versatile DRIE: Silicon pit structures combined with electrochemical etch stop,” IEEE J. Microelectromech. Syst., vol. 15, no. 4, pp. 840–848, Aug. 2006.

T. Zhong is currently an undergraduate student in the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, with the major subjects being photonics and microelectronics. He has worked on the design and characterization of optical MEMS under the university undergraduate research program and received the top award.

X. M. Zhang (S’03–M’05) received the B.Eng. degree in precision mechanical engineering from the University of Science and Technology of China, Hefei, Anhui, China, in 1994, the M.Eng. degree in optical instrumentation from Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Science, Beijing, China, in 1997, the Postgraduate degree in MEMS from the Department of Mechanical Engineering, National University of Singapore (NUS), Singapore, in 2002, and the Ph.D. degree from the School of Electrical and Electronic Engineering, Nanyang Technological University (NTU), Singapore, in 2006. Since 2000, he has been a Research Associate at the School of Electrical and Electronic Engineering, NTU, where he is currently a Singapore Millennium Foundation Postdoctoral Fellow. His current research interests include optical MEMS, optical communications, micro-optics, photonic MEMS, and microsystems.

A. Q. Liu (M’03) received the B.Eng. degree in mechanical engineering from Xi’an Jiaotong University, Xi’an, China, in 1982, the M.Sc. degree in applied physics from Beijing University of Posts and Telecommunications, Beijing, China, in 1988, and the Ph.D. degree in applied mechanics from the National University of Singapore, Kent Ridge, Singapore, in 1994. Currently, he is an Associate Professor at the Division of Microelectronics, School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore. He is also an Associate Editor of the IEEE SENSORS JOURNAL. His current research interests include MEMS design, simulation, and fabrication processes.

J. Li received the B.Eng. and M.Eng. degrees in electrical and electronic engineering from Xi’an Jiaotong University, Xi’an, China, in 1995 and 1998, respectively, and the Ph.D. degree from the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, in 2006. Currently, she is a Senior Research Engineer with the Institute of Microelectronics, Singapore. Her current research interests include optical MEMS, optical communications, and optical module package.

C. Lu (M’91–S’06) received the B.S. degree from Tsinghua University, Beijing, China, in 1985, and the M.S. and Ph.D. degrees from the University of Manchester, Manchester, U.K., in 1987 and 1990, respectively, all in electrical engineering. Since 2006, he has been a Professor at the Hong Kong Polytechnic University, Hong Kong, China. His current research interests include optical communication systems and networks. He is the author or coauthor of more than 100 journal and conference papers.

D. Y. Tang was born in Hubei, China, on July 2, 1963. He received the B.Sc. degree in physics from Wuhan University, Wuhan, China, in 1983, the M.Sc. degree in laser physics from Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Science, Beijing, China, in 1986, and the Ph.D. degree in physics from Hannover University, Hannover, Germany, in 1993. From 1993 to 1994, he was a Scientific Employee at the PhysikalischTechnische Budesanstalt, Braunschweig, Germany. From 1994 to 1997, he was a Postdoctoral Research Fellow, and from 1997 to 1999, an Australian Research Council (ARC) Postdoctoral Research Fellow at the University of Queensland, Brisbane, Australia. He is currently an Associate Professor in the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore. Dr. Tang is a member of the Optical Society of America and the International Society for Optical Engineering.