JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 24, NO. 12, DECEMBER 2006
`
`4433
`
`Optical MEMS for Lightwave Communication
`
`Ming C. Wu, Fellow, IEEE, Olav Solgaard, Member, IEEE, and Joseph E. Ford
`
`Invited Paper
`
`Abstract—The intensive investment in optical microelectro-
`mechanical systems (MEMS) in the last decade has led to many
`successful components that satisfy the requirements of lightwave
`communication networks. In this paper, we review the current
`state of the art of MEMS devices and subsystems for lightwave
`communication applications. Depending on the design, these com-
`ponents can either be broadband (wavelength independent) or
`wavelength selective. Broadband devices include optical switches,
`crossconnects, optical attenuators, and data modulators, while
`wavelength-selective components encompass wavelength add/drop
`multiplexers, wavelength-selective switches and crossconnects,
`spectral equalizers, dispersion compensators, spectrometers, and
`tunable lasers. Integration of MEMS and planar lightwave cir-
`cuits, microresonators, and photonic crystals could lead to further
`reduction in size and cost.
`
`Index Terms—Microelectromechanical devices, optical fiber
`communication, optical signal processing, optical switches.
`
`I. INTRODUCTION
`
`N EARLY three decades ago, Petersen published a paper
`
`on the micromechanical spatial light modulator (SLM)
`array [1] and another on the silicon torsion mirror [2]. Thirty
`years later, this has become a thriving field known as optical
`microelectromechanical systems (MEMS), sometimes also
`called microoptoelectromechanical systems, with several con-
`ferences dedicated to the field. It is a key enabling technology
`for the “dynamic” processing of optical signals. The first mar-
`ket driver of optical MEMS was display [3], [4]. The digital
`micromirror devices developed by Texas Instruments Incorpo-
`rated are one of the most successful MEMS products. They
`are now widely used in portable projectors, large-screen TVs,
`and digital cinemas [3]. The applications of optical MEMS in
`telecommunications started in the 1990s [5], [6]. Early efforts
`
`Manuscript received July 7, 2006; revised October 2, 2006. This work was
`supported in part by the U.S. Defense Advanced Research Project Agency
`(DARPA)/Army Research Office under Grant W911NF-05-1-0359 and DARPA
`under Grant MDA972-02-1-0020.
`M. C. Wu is with the Berkeley Sensor and Actuator Center (BSAC) and
`Electrical Engineering and Computer Sciences Department, University of
`California, Berkeley, CA 94720 USA (e-mail: wu@eecs.berkeley.edu).
`O. Solgaard is with the E. L. Ginzton Laboratory, Stanford University,
`Stanford, CA 94305 USA (e-mail: solgaard@stanford.edu).
`J. E. Ford is with the Department of Electrical and Computer Engineering,
`University of California, San Diego, CA 92093-0407 USA (e-mail: jeford@
`ucsd.edu).
`Color versions of Figs. 3, 5, 10–12, 14, 15, 17, 18, 20, 22, and 25–28 are
`available online at http://ieeexplore.ieee.org.
`Digital Object Identifier 10.1109/JLT.2006.886405
`
`have focused on the development of optical MEMS devices and
`fabrication technologies [7]–[10]. The telecom boom in the late
`1990s and early 2000s has accelerated maturation of the tech-
`nology. A wide range of optical MEMS components were taken
`from laboratories to reliable products that meet Telcordia qual-
`ifications. Although not all commercialization endeavors were
`successful due to the market downturn, the technology devel-
`oped is available for new applications in communications and
`other areas [11].
`In this paper, we will review the recent developments in
`optical MEMS for communication applications. With the rapid
`expansion of the field and proliferation of literature, it is not
`possible to cover all developments in the last decade. Instead,
`we will focus on a selected set of applications and discuss the
`design tradeoffs in MEMS devices and systems. Topics selected
`in this paper include optical switches, filters, dispersion com-
`pensators, spectral equalizers, spectrometers, tunable lasers,
`and other dense-wavelength-division-multiplexing (DWDM)
`devices such as wavelength add/drop multiplexers (WADMs),
`wavelength-selective
`switches
`(WSSs),
`and wavelength-
`selective crossconnects (WSXC). Most of the practical com-
`ponents reported were based on free-space optics. There are
`increasing interests in extending the benefits of optical MEMS
`to guided-wave optics or even nanoscopic photonic structures.
`This new trend will be discussed at the end of this paper.
`Various types of optical switches are needed in telecommuni-
`cation networks. Small 1 × N and N × N switches are useful
`for protection, while optical crossconnect (OXC) offers fast
`provisioning and network management at the wavelength level.
`Nodes in ring networks employ WADMs. As the networks
`evolve toward mesh configuration, WSSs and WSXC become
`important. Dispersion compensators and spectral equalizers are
`essential for improving the link performance as the data rates
`approach 40 Gb/s. Spectral filters and tunable lasers increase
`the flexibility of DWDM nodes.
`This paper is organized as follows: Section II discusses
`broadband (wavelength-independent) devices, including data
`modulators, variable optical attenuators (VOAs), and two-
`dimensional (2-D) and three-dimensional (3-D) MEMS optical
`switches. Section III describes wavelength-selective MEMS,
`including spectral equalizers, WADMs, WSSs, WSXCs, filters,
`dispersion compensators, transform spectrometers, and tunable
`lasers. Section IV focuses on the integration of MEMS and
`planar lightwave circuits (PLC). Section V introduces new de-
`vice concepts based on MEMS-actuated microresonators and
`photonic crystals, and Section VI concludes this paper.
`Capella 2025
`Fujitsu v. Capella
`IPR2015-00727
`
`0733-8724/$20.00 © 2006 IEEE
`
`

`
`4434
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 24, NO. 12, DECEMBER 2006
`
`Fig. 1. MEMS etalon modulator used for digital data modulation at over
`1 Mb/s. The circular optical aperture is 22 μm in diameter.
`
`Fig. 2. Package configuration for a MEMS data transceiver.
`
`II. WAVELENGTH-INDEPENDENT MEMS
`
`A. Data Modulators
`
`The first practical application of MEMS devices in fiber
`communications was as an optical data modulator, originally
`intended for a low-cost fiber-to-the-home network. A modulator
`is essentially a 1 × 1 switch, operated in either transmission
`(two fibers) or reflection (single fiber). The optical power is
`provided by a constant-intensity remote source, and the mod-
`ulator imprints a data signal by opening and closing in response
`to an applied voltage. Signaling in DWDM fiber networks
`usually requires an expensive wavelength-controlled laser at
`each remote terminal. Passive data modulators offered a poten-
`tially inexpensive solution, but waveguide modulators were too
`expensive and too narrow in optical spectral bandwidth to be
`practical. MEMS offered a new and practical solution.
`The mechanical antireflection switch (MARS) modulator
`is a variable air-gap etalon operated in reflection. The basic
`structure is a quarter-wave dielectric antireflection (AR) coating
`suspended above a silicon substrate [5]. The quarter-wave layer
`is made of silicon nitride with 1/4λ optical path (index times
`thickness), which is roughly 0.2 μm for the 1550-nm telecom
`wavelength. The mechanically active silicon nitride layer is
`suspended over an air gap created by a 3/4λ-thick phospho-
`silicate glass sacrificial layer (0.6 μm). Without deformation,
`the device acts as a dielectric mirror with about 70% (−1.5-dB)
`reflectivity. Voltage applied to electrodes on top of the mem-
`brane creates an electrostatic force and pulls the membrane
`closer to the substrate, while membrane tension provides a
`linear restoring force. When the membrane gap is reduced
`to λ/2, the layer becomes an AR coating with close to zero
`reflectivity. A switching contrast ratio of 10 dB or more was
`readily achieved over a wide (30-nm) spectral bandwidth.
`The initial MARS device shown in Fig. 1 consisted of a
`22-μm optical window supported by X-shaped arms and had
`a resonant frequency of 1.1 MHz. Later devices used a higher-
`yield structure with a symmetric “drum head” geometry [12],
`[13]. These devices were capable of relatively high-speed
`operation: by optimizing the size and spacing of the etch,
`access holes provide critical mechanical damping, and digital
`modulation above 16 Mb/s was demonstrated [14]. While such
`data rates are no longer relevant for telecom, even for fiber-to-
`
`the-home, related modulators are useful for low-power dissi-
`pation telemetry from remote sensors using free-space optical
`communications.
`These early devices provided a proving ground for the reli-
`ability and packaging of optical MEMS telecom components.
`Initial skepticism from conservative telecom engineers was
`combated by the parallel testing of device array operated for
`months to provide trillions of operating cycles. The packaging
`of optical MEMS devices provided new challenges for MEMS
`engineers, but the simple end-coupled configuration was rela-
`tively straightforward to implement. Fig. 2 shows the config-
`uration for a duplex modulator incorporating a MEMS etalon,
`where data can be received by a photodiode and transmitted by
`modulating the etalon reflectivity [15].
`
`B. Variable Attenuators
`
`Data modulators are operated with digital signals, but the
`fundamental response of an etalon modulator is analog. Elec-
`trically controlled VOAs at that time were constructed with
`bulk optical components with electromechanical actuation, with
`10–100-ms response. Erbium fiber amplifiers can use VOA to
`suppress transient power surges, but the time scale required
`was 10 μs, much slower than the data modulation rate. MEMS
`provided an attractive replacement for optomechanical VOAs,
`and this turned out to be the first volume application for MEMS
`devices in telecom networks.
`The first MEMS VOA was fabricated by scaling the opti-
`cal aperture of a MARS modulator from 25 to 300 μm so
`that it could be illuminated with a collimated beam. The re-
`flected signal was focused into a separate output fiber, avoiding
`the need for external splitters or circulators to separate the
`output signal [16]. The first such VOA device is shown in
`Fig. 3. The wavelength dependence of a simple etalon was
`reduced using a more complex three-layer dielectric stack as the
`mechanically active structure, where the original 1/4λ silicon
`nitride layer is sandwiched between conductive polysilicon top
`(1/2λ thickness) and bottom (1/4λ thickness) layers. This at-
`tenuator provided fast (3 μs) response with 30-dB controllable
`attenuation over the 40-nm operating bandwidth, with 0.06-dB
`polarization-dependent loss, and also supported the 100-mW
`power level present in amplifiers. However, the 3-dB insertion
`loss was excessive.
`
`

`
`WU et al.: OPTICAL MEMS FOR LIGHTWAVE COMMUNICATION
`
`4435
`
`Fig. 3. MEMS etalon variable attenuator using a 0.5-mm diameter drumhead
`geometry. The lighter area covers an air gap between the silicon substrate. The
`hexagonally distributed spots are etch access holes.
`
`Fig. 5. Schematic of 2-D MEMS optical switches.
`
`C. Two-Dimensional MEMS Switches
`Protection switches are made of 1 × N or small N × N
`switches. This can be realized by a 2-D array of vertical micro-
`mirrors commonly known as a 2-D MEMS switch. Fig. 5 shows
`the generic schematic of such a switch. The optical beams
`are collimated to reduce diffraction loss. The micromirrors are
`“digital”: They either direct the optical beams to the orthogonal
`output ports or pass them to the drop ports. Generally, only one
`micromirror in a column or row is in the reflection position
`during operation.
`The first MEMS 2-D switch (2 × 2) was reported in [22]
`and quickly followed by related work [23], [24]. For 2 × 2
`switches, low insertion loss (0.6 dB) can be achieved without
`using collimators, especially when the micromirror is immersed
`in index-matching fluid [25]. Latchable 2 × 2 switches incor-
`porating MEMS bistable structures were later commercialized
`[26], [27]. Larger switches require optical collimators to reduce
`diffraction loss. Switches with 8 × 8 and 16 × 16 ports were
`demonstrated [28], [29]. There are two basic approaches for the
`actuation of the micromirror. The first is based on the rotation
`of the micromirror [22], [28], [30], [31]. The mirror is initially
`parallel to the substrate (OFF position). When actuated, it is ro-
`tated to the vertical position (ON). The second approach moves
`the vertical micromirrors in and out of the optical paths without
`changing the mirror angle [23]–[25], [29], [32], [33]. The
`2-D switches have been realized by both bulk-micromachining
`[22]–[25] and surface-micromachining [28]–[30], [32] technol-
`ogies. Electrostatic actuation is most commonly used [22]–[29],
`[32]. Magnetic actuation has also been demonstrated [23], with
`some in conjunction with electrostatic clamping [30].
`The port count of 2-D switches is determined by several
`factors, including mirror angle, size, fill factor (mirror width
`divided by unit cell width), and curvature. The expandabil-
`ity of the 2-D switch has been studied in [34] and [35].
`To minimize optical diffraction loss, a confocal geometry
`is used with the average optical path length equal to the
`Rayleigh range, which is proportional to the square of the
`optical beam waist. Larger mirrors are therefore required to
`support longer Rayleigh length in higher port-count switches.
`In an N × N switch, the mirror size scales as N, whereas
`
`Fig. 4. Lightconnect’s diffractive MEMS VOA.
`
`The most direct possible approach to attenuation is to use
`a MEMS actuator to insert an optical block between the input
`and output fiber. This was implemented with a surface micro-
`machining (MUMPS process) [17] and with a comb-driven
`silicon-on-insulator (SOI) device [18]. Such VOAs offered
`excellent dynamic range (measurement limited at 90 dB), but
`the polarization-dependent loss could be large ((cid:2) 1 dB) at high
`attenuations.
`Further improvement was needed and was made. Combin-
`ing the collimated beam geometry with a first-surface torsion
`mirror reflector provided a low-insertion-loss structure with
`excellent spectral and polarization performance. For example,
`the device demonstrated by Isamoto et al. [19] achieved 40-dB
`attenuation with a 600-μm mirror driven with 5 V to tilt up
`◦
`. Similar configurations were commercialized, although
`to 0.3
`the specific designs have not been published.
`Another commercial MEMS VOA is based on a diffractive
`MEMS device [4] also used with a collimated beam. This
`device provides excellent optical performance as well as high
`speed: stable operation with 30-dB contrast and less than
`40-μs response time using an 8-V drive. A novel structure with
`circularly symmetric features, shown in Fig. 4, was used to
`suppress the polarization-dependent loss to under 0.2 dB [20].
`This device was one of the first Telcordia-qualified MEMS
`components, with 40 000 units reportedly shipped by 2005 [21].
`
`

`
`4436
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 24, NO. 12, DECEMBER 2006
`
`Fig. 7. Schematic of a 3-D MEMS switch.
`
`D. Three-Dimensional MEMS Switches
`
`A transparent optical crossconnect (OXC) with large port
`count can be realized by 3-D MEMS switches illustrated in
`Fig. 7. The input and output fibers are arranged in 2-D arrays.
`The optical beams are steered in three dimensions by two stages
`of dual-axis micromirrors, directing it toward the desired output
`port. The 3-D MEMS switch has a favorable scaling law with
`respect to port count: Assuming the maximum scan angle of the
`mirror is fixed, the optical path length is proportional to N in
`an N × N switch. To maintain confocal configuration for min-
`√
`imum loss, the beam waist, and therefore the mirror size, needs
`N · √
`√
`N. As a result, the linear dimension of the mirror
`to scale as
`N = N [39]–[41]. In addition, it has low
`chip scales as
`and uniform insertion loss. The 3-D MEMS OXC is a subject
`of intense interest during the telecom boom around the turn
`of the century [42]–[46]. Early efforts (before 2002) focused
`on OXCs with port count ∼1000 × 1000 [47], [48], driven by
`the explosion of Internet data transport. Recently, interest has
`shifted to applications in metropolitan area networks, including
`metro access and metro core networks, which requires OXC
`with medium port count (∼100 × 100), with emphasis on low
`cost, low-power consumption, and small footprint [44], [49].
`Our discussion here will focus on this trend.
`Detailed design tradeoffs and system implementations of the
`3-D MEMS OXC have been reported recently [42]–[46]. Two
`schemes have been proposed to reduce the size of the switch
`and tilt angle of the micromirror. Lucent inserted a Fourier lens
`between the two micromirror chips with the focal length equal
`to the Rayleigh range of the optical beam (Fig. 8) [50]. This
`√
`reduces the required scan angle of the mirror. In addition, the
`2 times
`mirrors can be placed at the beam waist, resulting in
`smaller optical beams. This permits the use of smaller mirrors
`and/or reduction of the crosstalk. Fujitsu used a “rooftop”
`mirror to connect two adjacent micromirror chips (photograph
`show in Fig. 9) [44]. The rooftop mirror shifts the optical beams
`laterally, reducing the tilt angle requirement. Folding of the
`optical beam also shrinks the footprint of the switch.
`In the compact switch category, Lucent’s 64 × 64 switch
`has a size of 100 × 120 × 20 mm3, which can be mounted
`on a standard circuit board [49]. The insertion loss is 1.9 dB.
`Fujitsu’s 80 × 80 switch has a packaged size of 77 × 87 ×
`53 mm3 [44]. The average insertion loss is 2.6 dB. Impressively,
`the switch continues to operate under vibration or 50G shock
`
`(a) SEM of OMM’s 16 × 16 switch (reprinted from [29] with
`Fig. 6.
`permission). (b) Photograph of the packaged switch (reprinted from [36] with
`permission).
`
`the linear dimension of the chip scales as N 2 [35]. Large
`chips are more susceptible to imperfections in mirror angles,
`which cause walkoff of optical beams at the receiving fibers.
`Ultimately, the chip size will be limited by the fabrication
`precision of the micromirrors. 16 × 16 switches have been
`realized, and 32 × 32 switches are within the capability of
`today’s technology.
`Fig. 6(a) shows a scanning electron micrograph (SEM) of
`OMM’s 2-D switch [29]. A vertical mirror is attached at the
`tip of a cantilever. The tilted cantilever can be pulled down
`◦
`during
`electrostatically. The mirror angle is maintained at 90
`switching. The switch is fabricated using a standard three-
`polysilicon-layer surface-micromachining process. The mirrors
`are assembled into vertical position with angular distribution of
`(90 ± 0.1)◦
`. The hermetic switch package is shown in Fig. 6(b)
`[36]. Maximum insertion losses of 1.7 and 3.1 dB have been
`obtained for 8 × 8 and 16 × 16 switches, respectively, and
`the crosstalk is less than −50 dB. The switching time is less
`than 7 ms. Packaging is critical to attain long-term reliabil-
`ity and satisfy Telcordia qualification for telecommunication
`applications [36].
`There were also significant efforts in nonmirror-based
`MEMS 2-D switches [37], [38]. Both Agilent’s Champaign
`switch [37] and NTT’s OLIVE switch [38] used microfluidic
`actuation to switch light between intersecting waveguides. The
`Champaign switch used thermally generated bubbles to dis-
`place index-matching fluids at waveguide intersections, causing
`the light to bend by total internal reflection (TIR). The OLIVE
`switch used thermal-capillary force to move trapped bubbles.
`One drawback of these approaches is the cumulative losses
`and crosstalks through multiple waveguide intersections. The
`maximum port counts achieved are 32 × 32 and 16 × 16 for
`the Champaign and the OLIVE switches, respectively.
`
`

`
`WU et al.: OPTICAL MEMS FOR LIGHTWAVE COMMUNICATION
`
`4437
`
`Fig. 8. Lucent’s optical system layout for OXC (reprinted from [50] with
`permission). A Fourier lens is inserted between the two MEMS chips to reduce
`the required tilt of the mirror and beam size.
`
`(a) Dynamic spectral equalizer package and (b) transmission spectra
`Fig. 10.
`showing the improvement in channel uniformity for a 36-channel DWDM
`transmission.
`
`increases the complexity of electronics [58]. Micromirrors with
`vertical comb drive actuators, first reported in [59], offer many
`advantages. They have a much larger torque, which one can use
`to reduce the operating voltage as well as increase the resonant
`frequency. In addition, they are free from the pull-in effect,
`further increasing the stable tilt angles. It should be mentioned
`that lateral pull-in between comb fingers is a potential issue but
`could be mitigated by MEMS design (such as V-shaped torsion
`beam [60] or off-centered combs [61]). Several variations of
`vertical comb drive mirrors have been reported, including self-
`aligned vertical combs [62], [63], angular vertical combs [64],
`[65], electrostatically assembled vertical combs [66], and thick
`vertical combs (100 μm) attached to mirror edges on double-
`sided SOI wafers [44], [60].
`
`III. WAVELENGTH-SELECTIVE MEMS
`
`A. Spectral Equalizers
`
`The natural extension of a single variable attenuator is to
`provide a VOA for each channel of a DWDM transmission
`system. The surface-normal geometries of the etalon mirror-
`and grating-based attenuators discussed in Section II-B were
`all compatible with a free-space imaging spectrometer. An
`input fiber is imaged through a diffraction grating so that each
`spectral channel is laterally shifted to illuminate one modulator
`in a linear array. The reflected signal, attenuated to the desired
`value, is collected into a single output fiber by a second pass
`through the imaging spectrometer. The first such MEMS spec-
`tral equalizer used a continuous etalon membrane [67]. This
`approach was later implemented in the compact package shown
`in Fig. 10, which located the MEMS device array next to a
`single input/output (I/O) fiber. A single lens is to collimate
`the multiwavelength beam onto a blazed reflective grating and
`
`Fig. 9. Photograph of Fujitsu’s 80 × 80 OXC with a rooftop reflector
`connecting the two MEMS chips (reprinted from [44] with permission). The
`packaged size is 77 × 87 × 53 mm3.
`
`without any signal degradation. The total power consumption
`of Fujitsu’s switch is only 8.5 W, thanks to the low operating
`voltage of the mirrors. NTT’s 100 × 100 switch has a size of
`80 × 60 × 35 mm3 with an insertion loss of 4 dB [43].
`The two-axis micromirror array is the key enabling device
`of the 3-D switch. Important parameters include size, tilt an-
`gle, flatness, fill factor, and resonant frequency of the mirror.
`Additionally, the stability of the mirror plays a critical role
`in the complexity of the control schemes. Early development
`focused on surface-micromachined two-axis scanners [51],
`[52]. The residue stress limits the mirror size to approximately
`1 mm, and the different thermal expansion coefficients be-
`tween the mirror and the metal coating also cause the mirror
`curvature to change with temperature. Bulk-micromachined
`single-crystalline silicon micromirrors are often used in high-
`port-count OXCs that require larger mirror size [46], [53]–[56].
`Electrostatic actuation is most commonly used because of
`its low-power consumption and ease of control. Early devices
`use parallel-plate actuators, which have high actuation voltage
`and limited scan angle due to pull-in instability [57]. Although
`the pull-in effect can be mitigated by nonlinear controllers, it
`
`

`
`4438
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 24, NO. 12, DECEMBER 2006
`
`Fig. 11. Optical schematic for a 2 × 2 MEMS wavelength add/drop switch.
`
`refocus the spectrally separated signals with a second pass onto
`the MEMS array. A third and fourth pass through the lens
`reintegrates the signal into the I/O fiber, where it is separated
`by an external optical circulator. The use of such equalizers is
`illustrated by the before and after spectral traces at the bottom of
`Fig. 10, showing the improvement in uniformity of 36 channels
`sent through a two-stage amplifier. The equalizer setting was
`generated by an iterative algorithm running on the computer
`controller [68].
`Dynamic spectral equalization quickly went from an option
`to a practical requirement as the channel transmission rate
`increased from 2.5 to 10 and then to 40 Gb/s. The simplest
`and least expensive dynamic gain equalizers (DGEs) use a
`mid-amplifier filter that can be spectrally uniform (a VOA,
`as discussed above) or provide a constant spectral slope [69].
`Two distinct categories of spectral equalizers emerged. DGEs
`provide a smoothly varying spectral profile used to compensate
`for the varying gain profiles in amplifiers, while dynamic chan-
`nel equalizers (DCEs) provide the discrete channel-by-channel
`power adjustment needed to compensate for nonuniform trans-
`mission source intensity or path-dependent loss. Channel equal-
`izers are preferable in general but require accurate matching
`of the equalizer passband to the transmission grid to avoid
`passband narrowing.
`Channel equalizers were implemented using discrete VOAs
`attached to waveguide spectral multiplexers [70] and using an
`oversampled array of digital tilt mirrors [71]. However, the best
`performance in channel equalizers was achieved by combining
`the type of free-space grating demultiplexer shown in Fig. 10
`with either diffractive MEMS modulators [72] or analog tilt
`mirrors [73]. The optical setup is similar to that in Fig. 11
`except without circulators. These components typically have
`40–80 channels spaced at 100 or 50 GHz with 6- and 7-dB
`insertion loss and 20- to 30-dB dynamic range. The most ad-
`vantageous characteristic of MEMS equalizers is the extremely
`flat passband transmission profile along with low chromatic
`
`dispersion at the edges. This performance was achieved after
`studying the effects of various mirror geometries [74].
`After understanding the effects of mirror profile on disper-
`sion, it became possible to use the same basic component struc-
`ture as the equalizer to provide channel-by-channel dispersion
`compensation, although this functionality has yet to be adopted
`in the deployed network [75].
`
`B. Wavelength Add/Drop Multiplexers
`
`Wavelength switching allows network operators to use opti-
`cally transparent components to pass through a network node
`without detecting and regenerating the data signal, and com-
`ponents that enable this have been the subject of intense re-
`search and development. The most basic wavelength switch is
`the dynamically reconfigurable WADM, which is essentially a
`1 × 2 or 2 × 2 optical switch operating independently on each
`wavelength channel.
`WADM was a natural extension of MEMS equalizers, and the
`first demonstration of a MEMS add/drop switch based on dig-
`ital tilt mirrors occurred almost simultaneously with the equal-
`izer [76], [77]. Add/drop requires four ports, twice as many
`as the equalizer, and so, the basic structure is slightly more
`complex (Fig. 11). The system is still based on a blazed diffrac-
`tion grating, which is now illuminated with an upper and lower
`beam path. The active device is a linear array of 16 digital tilt
`mirrors fabricated with surface micromachining in the MUMPS
`process. Each mirror defines a DWDM channel and, in switch-
`ing, directs the reflected signal back along the input direction
`or tilted into a new path. Optical circulators on the two I/O
`fibers separate the forward and reverse propagating signals. The
`mirrors in this switch tilted by ±5◦
`under a 20-V signal, switch-
`ing in 20 μs. A quarter-wave plate is used to achieve 0.2-dB
`polarization dependence on a total insertion loss of 7.5 dB.
`The DCE is closely related to the WADM, and in fact, it is
`possible to use high-contrast equalizers as 1 × 1 switches in a
`
`

`
`WU et al.: OPTICAL MEMS FOR LIGHTWAVE COMMUNICATION
`
`4439
`
`Fig. 12. Equivalent circuit of a 1 × 4 WSS. Eight wavelength channels are
`shown in this example.
`
`“broadcast and select” architecture [78]. The primary disadvan-
`tage of this architecture is that it is intrinsically lossy: Signals
`are power split, and then unwanted signals are blocked before
`combining into the output fiber. This does allow multicasting,
`i.e., duplicating signals to multiple output fibers. Broadcast and
`select was actually the first to be implemented in the network
`but is generally expected to be phased out in favor of multiport
`WSSs, which in addition to switching also provide channel
`equalization [79] with no additional cost or complexity.
`
`C. Wavelength-Selective Switches (WSSs)
`
`As optical networks evolve from a simple ring topology with
`WADM nodes to optical mesh networks, WSSs with more than
`one output port are needed to link the node to three or four
`neighboring nodes with each link carrying two-way traffic. The
`WADM concept can be extended to switches with N output
`ports, where N is larger than 2. This is called 1 × N WSS
`[80]–[82]. Fig. 12 shows the equivalent circuits of a 1 × 4 WSS.
`It consists of a WDM demultiplexer, Nλ of 1 × N space divi-
`sion switches (Nλ is the number of wavelength channels) and
`N WDM multiplexers. The WSS can be realized by a similar
`grating spectrometer configuration as the WADM, with the dig-
`ital micromirrors replace by “analog” ones. A large continuous
`scan angle is required to direct the output beam to any of the N
`output fiber collimators. High fill factor is desired to minimize
`the gaps between wavelength channels. The mirror size is
`usually several times larger than the focused optical beam to
`attain a wide and flat passband for minimal signal distortion.
`A detailed review paper on WSS was published recently
`[80]. The optical setup for Lucent’s WSS is shown in Fig. 13.
`The first subassembly maps all fiber I/Os to a common spot
`(point B), and the second subassembly (resolution lens and
`grating) separates and focuses the wavelengths onto the mi-
`cromirror array at the image plane. Tilting of the mirror changes
`the direction of the reflected beam at point B and sends the
`optical beam into a different output fiber. A refined design
`incorporates anamorphic optics in the input stage to reduce the
`physical size of the switch while maintaining the same spectral
`resolution at the expense of longer micromirrors.
`Experimentally, 1 × 4 WSSs with 128 channels spaced on
`a 50-GHz grid and with 64 channels spaced on a 100-GHz
`
`Fig. 13. Schematic optical setup of 1 × 4 WSS (reprinted from [80] with
`permission).
`
`grid have been demonstrated. The typical optical insertion loss
`ranges from 3 to 5 dB. The channel passband is directly related
`to the confinement factor, which is defined as the ratio of the
`mirror size to the Gaussian beam diameter. A confinement
`factor of > 2.7 is needed to produce a flattop spectral response
`with > 74% passband width measured at −1 dB point. JDSU
`has reported a similar 1 × 4 WSS with 3.5-dB insertion loss
`[81]. UCLA has reported a similar WSS with excellent open-
`loop stability [82].
`The analog micromirror array plays a key role in the per-
`formance of the WSS. Several types of WSS micromirror
`arrays have been reported, including electrostatic [83], [84]
`and electromagnetic [85] actuations. The key parameters are
`large continuous scan angle and high fill factor, with the mirror
`size and pitch matching those of the optical system. Lucent
`employed a fringe-field actuated SOI micromirror array [83]
`◦
`at 175 V. The
`and achieved a mechanical tilt angle of 9.2
`resonant frequency is 3.8 kHz for 80-μm-wide mirrors.
`More efficient actuation has been obtained using vertical
`comb drive actuators. Hah et al. reported a low-voltage analog
`micromirror array for WSS [84]. The schematic and the SEM
`of the micromirror are shown in Fig. 14. The mechanical struc-
`tures are completely covered by the mirror; therefore, a high
`fill factor is achieved along the array direction. The actuation
`voltage is as low as 6 V for mechanical tilt angles of ±6◦
`. High
`resonant frequency (3.4 kHz) and high fill factor (98%) are also
`achieved [86]. The excellent stability of the mirror (±0.00085◦)
`enables open-loop operation of the switch with insertion loss
`variation of < ±0.0035 dB over 3.5 h [82].
`Scaling of WSS has been analyzed in [86]. The figure
`of merit is the ratio of the port count and channel spacing
`(N/Δλch). It is proportional to the product of the effective
`aperture of the resolution lens and the grating dispersion. Most
`of the reported WSSs have a port count of four. A larger port
`count (N ≥ 8) is desirable for mesh optical networks, where it
`is necessary to provide two-way links to three or four adjacent
`neighboring nodes. The port count can be increased from N
`to N 2 by arranging the output collimator in a 2-D array. This
`is referred to as 1 × N 2 WSS [86]–[88]. Micromirror arrays
`providing two-axis beamsteering functions are needed for this
`architecture. This can be accomplished by using either a linear
`array of two-axis micromir