OPTICAL SWITCHING
`
`MEMS: The Path to Large
`Optical Crossconnects
`
`Patrick B. Chu, Shi-Sheng Lee, and Sangtae Park, Tellium, Inc.
`
`ABSTRACT
`Continuous growth in demand for optical
`network capacity and the sudden maturation of
`WDM technologies have fueled the develop-
`ment of long-haul optical network systems that
`transport tens to hundreds of wavelengths per
`fiber, with each wavelength modulated at 10
`Gb/s or more. Micro-electromechanical systems
`devices are recognized to be the enabling tech-
`nologies to build the next-generation cost-effec-
`tive and reliable high-capacity optical
`crossconnects. While the promises of automati-
`cally reconfigurable networks and bit-rate-inde-
`pendent photonic switching are bright, the
`endeavor to develop a high-port-count MEMS-
`based OXC involves overcoming challenges in
`MEMS design and fabrication, optical packag-
`ing, and mirror control. Due to the interdepen-
`dence of many design parameters, manu-
`facturing tolerances, and performance require-
`ments, careful trade-offs must be made in
`MEMS device design as well as system design.
`In this article we provide a brief overview of the
`market demand, various design trade-offs, and
`multidisciplinary system considerations for
`building reliable and manufacturable large
`MEMS-based OXCs.
`
`INTRODUCTION
`To meet the growing demand for high data
`bandwidth, service providers are building opti-
`cal networks around the globe using the latest
`wavelength-division multiplexed (WDM) tech-
`nologies with mesh network architecture [1].
`Lightpaths between access points in a network
`are created using fiber links containing many
`wavelength channels in each fiber, where each
`channel or port can have a data rate of up to
`2.5 or 10 Gb/s. At the edge of the networks are
`the clients (IP/ATM routers, optical add-drop
`multiplexers, etc.) that use these lightpaths as
`high-capacity pipes for data/voice traffic. Data
`rate per port is expected to continue to increase
`(40 Gb/s in the very near future). The number
`of wavelength channels (or ports) per fiber will
`
`80
`
`0163-6804/02/$17.00 © 2002 IEEE
`
`also continue to rise as WDM technologies
`mature.
`For long-haul core networks, core switching
`is needed for two main purposes: network pro-
`visioning and restoration (Fig. 1). Provisioning
`occurs when new data routes have to be estab-
`lished or existing routes modified. A network
`switch should carry out reconfiguration
`requests over time intervals on the order of a
`few minutes. However, in many core networks
`today, provisioning for high-capacity data
`pipes (OC-48 — 2.5 Gb/s and OC-192 — 10
`Gb/s) requires a slow manual process, taking
`several weeks or longer. High-capacity recon-
`figurable switches that can respond automati-
`cally and quickly to service requests can
`increase network flexibility, and thus band-
`width and profitability.
`On the other hand, restoration must take place
`in events of network failures (e.g., an accidental
`cable cut). A network switch needs to reroute traf-
`fic automatically in a time interval on the order of
`100 ms, thus restoring operation of the network.
`Traditionally, network restoration is performed pri-
`marily by digital electronic cross-connects and syn-
`chronous optical network (SONET) add-drop
`multiplexers, operating at a data rate of about
`45–155 Mb/s. For switches in a core network han-
`dling hundreds of gigabits per second of traffic,
`restoration at a coarser granularity is desirable in
`terms of both cost and manageability. Provisioning
`and restoration at coarse granularities also makes
`sense in light of the development of high-speed
`service-layer equipment such as IP routers with 10
`Gb/s interface and Gigabit Ethernet.
`These provisioning and restoration require-
`ments of next-generation optical networks
`demand innovations in switching technologies.
`In the following sections, a vision and tech-
`nologies for next-generation optical crosscon-
`nects (OXCs) are described, with a focus on
`MEMS technologies as the leading choice for
`photonic switching. Key challenges associated
`with the development of MEMS-based OXCs
`are discussed. Finally, an outlook on MEMS-
`based OXC development and deployment is
`presented.
`
`Capella 2009
`
`Ciena/Coriant/Fujitsu v. Capella
`IPR2015-00816
`IEEE Communications Magazine • March 2002
`
`

`

`NEXT-GENERATION
`CROSSCONNECTS
`
`An emerging vision of the next-generation cross-
`connects for optical networks is one that allows
`network reconfiguration in the optical layer (Fig.
`2a): provisioning and restoration in large units
`(e.g., the wavelength). Since the number of wave-
`lengths per fiber has already reached hundreds
`today (160 wavelengths for 10 Gb/s) and is expect-
`ed to increase, the desired port counts for such
`OXCs are expected to be in the thousands, where
`scalability is a paramount concern. Such a switch
`must also operate in a fully nonblocking manner,
`where every input must be allowed to connect to
`every output with no restriction. In addition, inser-
`tion loss, physical size, polarization effects, and
`switching times are also critical considerations.
`Equipped with intelligent provisioning and restora-
`tion capabilities, the next-generation OXC must
`also meet the stringent telecommunication require-
`ments with an operating lifetime of 20 years.
`
`OPTICAL-LAYER SWITCH WITH AN
`ELECTRICAL SWITCHING CORE
`
`An optical layer switch can be implemented using
`opto-electronics interfaces and high-speed elec-
`tronics. Due to the advancement of state-of-the-
`art integrated circuit (IC) technologies, multiple
`vendors currently offer electronics-based optical
`switches, also known as O-e-O (Optical-electrical-
`Optical) switches, with a few hundred 2.5–Gb/s
`ports residing in several equipment bays. These
`state-of-the-art switching systems provision and
`mesh-restore wavelengths at a granularity of 155
`Mb/s to 2.5 Gb/s. For example, Fig. 2b shows Tel-
`lium’s Aurora Optical Switch™ that has 512 OC-
`48 (2.5 Gb/s) input ports and 512 OC-48 output
`ports, and can deliver a total aggregate capacity
`of 1.28 Tb/s. They also provision and mesh-restore
`
`Transponders
`
`1… n
`
`1
`
`n
`
`WDM
`mux/demux
`
`Optical-layer
`switch
`
`Sonet
`
`IP
`
`ATM
`
`(a)
`
`Node
`Optical path
`Provisioned path
`Break
`
`■ Figure 1. Illustration of data path provisioning and restoration in a core
`transport mesh network.
`
`10 Gb/s wavelengths (OC-192) via inverse multi-
`plexing down to the basic switch rate, with the
`capability of grooming such subrate signals within
`a given 10 Gb/s pipe. Intelligence of this switch
`allows dynamic and automatic provisioning and
`protection as well as in-service system upgrades.
`Based on multiple stages of Clos structures [1],
`these switches are also scalable to thousands of
`switching ports.
`
`OXCS WITH MEMS-BASED
`OPTICAL SWITCHING CORE
`
`OXCs with electrical switching cores like the
`Aurora Optical Switch will continue to be
`deployed and remain in service for quite some
`time. Higher-speed and higher-capacity electron-
`ics switches are expected to reach the market in
`the near future as IC technology advancement
`
`(b)
`
`■ Figure 2. a) Illustration of an optical-layer switch connected to DWDM transport systems and client
`equipment; b) Tellium’s Aurora Optical Switch™ with 512 OC-48 (2.5 Gb/s) input ports and 512 OC-48
`output ports, 1.28 Tb/s of aggregate switching capacity deployed, and carrying commercial traffic today.
`
`IEEE Communications Magazine • March 2002
`
`81
`
`

`

`Free-rotating
`switch mirror
`
`Fiber-
`collimator
`array
`
`Fiber-collimator
`array
`
`(a)
`
`(b)
`
`■ Figure 3. a) Illustration of a 2D switching architecture; b) 2D N ¥ N switches first demonstrated by AT&T [8].
`
`continues. On the other hand, the possibilities of
`improved scalability, footprint, manageability,
`and cost continue to fuel the quest for techno-
`logical solutions beyond the proven state of the
`art. A new concept that has arisen is an all-opti-
`cal OXC: an optical-layer switch with an optical
`switching core. All-optical switches are also
`known as O-o-O (Optical-optical-Optical) switch-
`es, which can be realized using arrays of MEMS-
`fabricated micro-mirrors.
`
`MEMS for Photonic Switching — MEMS
`technology enables the fabrication of actuated
`mechanical structures with fine precision that
`are barely visible to the human eye. MEMS
`devices are by nature compact and consume
`low power. A batch fabrication process allows
`high-volume production of low-cost devices,
`where hundreds or thousands of devices can be
`built on a single silicon wafer. While the
`MEMS field is young compared to traditional
`semiconductor electronics, MEMS technology
`is based on fabrication technology fundamental
`to IC fabrication and many mature engineering
`disciplines such as mechanics, electromagnet-
`ics, and material science. Applied research in
`MEMS over the past two decades has led to
`numerous successful commercial devices,
`including valves and pressure sensors for auto-
`motive and medical applications, accelerome-
`ters, and angular rate sensors for airbags, toys,
`and instrumentation on land, at sea, in air, and
`in space. On the other hand, technological
`wonders such as injectable micromachines per-
`forming heart surgery inside the human body
`will remain fantasies of fiction writers for many
`decades to come.
`Optical MEMS, nevertheless, is a promising
`technology to meet the optical switching need
`for large-port-count high-capacity OXCs. Within
`the last decade, the realization that tiny micro-
`machined structures can steer light by generating
`small tilting motions has opened doors to many
`exciting applications of MEMS in photonic
`switching [2–4]. Current (nonelectronics) com-
`peting technologies for building are thermal
`bubble switches, which make use of total internal
`reflection and index-matched fluid, and wave-
`guide-based switches, which make use of inter-
`
`ferometric effects of light in planar waveguides.
`Potential benefits of an all-optical MEMS-based
`OXC include scalability, low loss, short switching
`time, low power consumption, low crosstalk and
`polarization effects, and independence of wave-
`length and bit rate. Therefore, MEMS has
`become the leading choice of technology for
`building large all-optical OXCs.
`The most notable commercial MEMS optical
`devices to date are Texas Instruments’ Digital
`Mirror Devices (DMD) [5], which have found
`applications in consumer visual display and pro-
`jectors. While different MEMS-based solutions
`for critical transmission applications such as
`gain equalization [6] and dispersion compensa-
`tion [7] are under investigation, add-drop multi-
`plexers and small protection switches are among
`MEMS-based optical products that are slowly
`reaching the market. In recent news, small opti-
`cal switch products have been announced to
`pass rigorous Telcordia telecommunications
`specifications, beginning to cast away healthy
`doubts about the long-term reliability of MEMS
`devices. Large MEMS-based OXCs as fully
`qualified products are expected to be a reality in
`the near future.
`
`Two-Dimensional MEMS Switches — The
`OXCs of main interest are fully nonblocking opti-
`cal switches with N input and N output ports.
`Two architectures for MEMS-based OXCs have
`emerged. In the first architecture, often known as
`2D switching (Fig. 3) [2, 8, 9], a square array of N
`¥ N mirrors is used to couple light from a linear
`array of N fibers on one side of the square to a
`second linear array of N fibers on an adjacent
`side of the square. The (i, j) mirror is raised up to
`direct light from the ith input fiber to the jth out-
`put fiber. Mirror control for these 2D switches is
`binary and thus straightforward, but the trade-off
`of this simplicity is optical loss. While the path
`length grows linearly with N, the number of ports,
`the optical loss also grows rapidly due to the
`Gaussian nature of light. Therefore, 2D architec-
`tures are found to be impractical beyond 32 input
`and 32 output ports. While multiple stages of 32 ¥
`32 switches can theoretically form a 1000-port
`switch, high optical losses also make such an
`implementation impractical.
`
`82
`
`IEEE Communications Magazine • March 2002
`
`

`

`Optical
`signal
`
`Lens
`array
`
`Fiber
`array
`
`Optical
`path
`
`MEMS
`array
`
`Fiber
`array
`
`MEMS
`array
`
`Lens
`array
`
`(a)
`
`(b)
`
`(c)
`
`■ Figure 4. a) Illustration of 3D switching architecture; b) illustration of beam steering using a two-axis gimbaled mirror; c) fabricated
`MEMS gimbaled mirror array.
`
`Three-Dimensional MEMS Switches — In
`the 2D case, all the light beams in the switch
`reside on the same plane, resulting in unac-
`ceptably high loss for large port counts. The
`second architecture (Fig. 4a), known as 3D
`switching [10–12], makes use of the three-
`dimensional space as an interconnection
`region, allowing scaling far beyond 32 ports
`with acceptable optical losses (< 10 dB). In
`this architecture, there is a dedicated movable
`mirror for each input and each output port.
`Each mirror must now operate in an analog,
`rather than binary, mode, tilting freely about
`two axes (Fig. 4b, c). This elegant architecture
`offers the virtue that the optical path length
`now scales only as ÷N instead of N, so port
`counts of several thousand are achievable with
`losses below 10 dB. This 3D optical architec-
`ture clearly presents real hope for developing a
`scalable large-port-count OXC.
`
`THE PATH TO A MEMS-BASED
`OPTICAL CROSSCONNECT
`MEMS-based OXCs are no doubt feasible in
`concept. Substantial challenges must be over-
`come for any switch design; these challenges
`include MEMS mirror manufacturing, optome-
`chanical packaging, and mirror control. Many
`aspects of these three challenges are interdepen-
`dent. Complex trade-offs must be weighed in
`designing a MEMS-based OXC.
`MEMS DESIGN AND FABRICATION
`Components of a large MEMS-based OXC
`include thousands of actuated mirrors, lenses,
`collimators, and fiber arrays. With no doubt
`MEMS mirrors, the key active element in the
`optical system, are the most critical technology
`for large OXCs.
`
`MEMS Design — A two-axis actuated tilting
`mirror can be divided into three elements: the
`mirror, the springs as the mechanical support,
`and the actuator, all of which determine impor-
`tant OXC system parameters. Examples of these
`parameters include maximum port count (depen-
`dent on the mirror tilt angle), switch settling
`time (dependent on the mirror response time),
`insertion loss (dependent on the mirror size,
`reflectivity, and maximum tilt angle), and power
`dissipation (dependent on power required for
`mirror actuation and control). For a 1000-port
`switch, each mirror may require a diameter on
`the order of 1 mm, with mirror radius of curva-
`ture (ROC) greater than a few tens of centime-
`ters. Reflectivity of each mirror is desired to be
`above 97 percent. The tilt angle requirement
`ranges from a few degrees to ±10˚ depending
`on the optical train design of the OXC.
`The challenges in MEMS design come from
`the different trade-offs between desired proper-
`ties of the MEMS device. As an example, the
`supporting springs for the mirrors must have
`sufficient stiffness to meet the mirror response
`time and vibration immunity requirement. But
`the upper bound of the spring stiffness is deter-
`mined by the desired maximum tilt angle and
`the actuator’s maximum force or torque output
`(as well as the switch power budget). Magnetic
`actuation and electrostatic actuation are two
`viable choices for mirror positioning. Magnetic
`actuation offers the benefit of large bidirection-
`al (attractive and repulsive) linear force output
`but requires a complex fabrication process and
`electromagnetic shielding. Electrostatic actua-
`tion is the preferred method mainly because of
`the relative ease of fabrication and integration.
`However, to achieve large tilt angle using a stiff
`spring, the trade-offs include high actuation
`voltages (on the order of 50–200 V) and nonlin-
`ear torque output.
`
`IEEE Communications Magazine • March 2002
`
`83
`
`

`

`For a typical
`Z-configuration
`1000-port switch,
`coupling losses
`between the
`input and output
`fibers can be
`computed using
`Gaussian beam
`propagation
`methods.
`Component
`fabrication
`tolerances and
`packaging
`tolerances can
`also be
`estimated.
`
`SiO2
`
`Patterned Au
`
`Si
`
`Mirror
`
`Conductive material
`
`Electrode
`
`Bottom substrate
`
`(a)
`
`(b)
`
`(c)
`
`■ Figure 5. a) Top view of a MEMS mirror; illustration of an SOI-based electrostatic MEMS mirror; b)
`before; and c) after structural release of the gimbaled mirror.
`
`A particular challenge for MEMS mirror
`design is to maximize ROC. A stable metal
`coating such as of gold, along with necessary
`additional metal adhesion and diffusion barrier
`layers, is often used as a reflective surface.
`These metal coatings can create an undesirable
`temperature-dependent mirror curvature due to
`intrinsic stress of the metal layers and the dif-
`ference in thermal expansion coefficients of the
`metal coating layers and the bulk mirror made
`of a different material. This problem is espe-
`cially severe if the metal coating is applied only
`to one side of the bulk mirror. A thick mirror
`can best counteract curvature from stress
`induced by metal coating on the mirror. Unfor-
`tunately, large mass leads to slow mirror
`response time and high sensitivity to stochastic
`vibration.
`
`MEMS Fabrication Choices — In principle,
`the bulk mirror can be made of any material as
`long as reliability, reflectivity, and optical flat-
`ness requirements are met. Single-crystal silicon
`(SCS), commonly used in MEMS, is recognized
`to be the most suitable choice over polysilicon or
`electroplated metal due to low intrinsic stress
`and excellent surface smoothness. The choice of
`material for the mirror springs is arguably even
`more important because the mirror springs will
`constantly be twisted and bent. Superior mechan-
`ical characteristics make SCS the best candidate
`for the mirror springs. Alternative materials such
`as polysilicon and metal are poor substitutes
`because of potential stress, hysteresis, and
`fatigue problems. In most cases, the same mate-
`rial is chosen for both the bulk mirror and the
`springs in order to yield a straightforward fabri-
`cation process.
`A plethora of fabrication processes can be
`used to create two-axis actuated SCS mirrors or
`mirror arrays [11, 13, 14]. Besides typical litho-
`graphy, deposition, and etching procedures, nec-
`essary fabrication steps may include deep
`reactive ion etches (DRIE), silicon wafer bond-
`ing, and chemical mechanical polishing (CMP)
`[5]. Silicon-on-insulator (SOI) wafers are a con-
`venient starting material to create SCS bulk
`mirrors with uniform thickness and low intrinsic
`stress (Fig. 5), but these wafers are unfortunate-
`ly expensive with few supply vendors today.
`Applying clever silicon etching and wafer bond-
`ing techniques to cost-effective [100]-type sili-
`con wafers may also yield mirrors with
`sufficiently low mass and large ROC. The pri-
`
`mary differentiating factor between these
`MEMS mirror processes is device performance
`characterized by mirror size, flatness, reflectivi-
`ty, maximum mirror tilt angle, and ease of mir-
`ror control. Material supply availability, length
`of fabrication cycles, and equipment bottlenecks
`play important roles in shortening product
`development cycle and time to market. Ease of
`circuit integration, achievable mirror array fill-
`factor, mirror array size, and manufacturing
`yield may also influence the overall switch fabric
`design. Arguably, a fabrication process that
`enables monolithic integration of electronics
`with MEMS [14] may lead to MEMS mirrors
`with the greatest functionality and the highest
`performance.
`
`OPTICAL PACKAGING
`The optical system as shown in Fig. 4a requires
`thousands of micro-mirrors, lenses, and fibers
`aligned to each other with tolerances on the
`order of microns and hundreds of micro-radians.
`This multi-element body must endure thermal
`cycles, shock, and vibration during shipping and
`operation, which may lead to short-term and
`long-term mechanical drift in packaging. Obvi-
`ously, tolerance of various pointing errors and
`misalignment errors depends on the robustness
`of the optical architecture design. In addition,
`these thousands of optical components must be
`carefully and compactly packaged with all the
`necessary control electronics in order to meet
`the additional space constraints and front panel
`accessibility requirements of telecommunications
`equipment.
`For a typical Z-configuration 1000-port
`switch like Fig. 4b, coupling losses between the
`input and output fibers can be computed using
`Gaussian beam propagation methods. Compo-
`nent fabrication tolerances and packaging tol-
`erances can also be estimated [4]. For example,
`±1percent of focal variation in a single port
`lens in a lens array could account for up to 1
`dB of optical loss. ±2 mm of relative position
`error in a fiber array can also lead to similar
`losses. One method to facilitate packaging is to
`make use of large fiber bundles, lenslet arrays,
`and monolithic dies with thousands of mirrors.
`The number of optical elements in the system
`may then be reduced to half a dozen or so.
`However, fabrication and packaging of such
`large fiber bundles, lenslet arrrays, and MEMS
`mirror dies poses formidable challenges of
`their own.
`
`84
`
`IEEE Communications Magazine • March 2002
`
`

`

`Snap-down
`voltage
`
`Applied voltage
`
`Linear
`spring torque
`
`Stable
`region
`
`Unstable
`region
`
`Nonlinear
`electrostatic torque
`
`Increasing
`voltage
`
`Torque
`
`Snap-down angle
`
`Mirror tilt angle
`
`Snap-down angle
`
`Mirror tilt angle
`
`(a)
`
`(b)
`
`■ Figure 6. a) For a given applied voltage, two intersection points are found between the nonlinear electrostatic torque curve and linear restor-
`ing spring torque curve, each corresponding to an equilibrium tilt angle; b) however, only the first intersection point is open-loop stable.
`
`Packaging of MEMS structures such as
`accelerometers and pressure sensors requires
`special attention beyond traditional integrated
`circuit packaging because of their sensitivity to
`contaminants, physical contact, and shocks.
`Packaging of optical MEMS structures on
`large (> 10 cm2) dies introduces new complex-
`ities more challenging than ever before. To
`guarantee long-term operation of the MEMS
`mirror, the MEMS die should be hermetically
`sealed in a package with an anti-reflection
`(AR) coated optically clear window. Rigorous
`thermal management of the MEMS die pack-
`age may be required since mirror ROC can be
`a strong function of temperature. Signal rout-
`ing and inputs/outputs (I/Os) to the die are
`also paramount considerations. Due to the
`large number of die I/Os (1000 or more), a
`large die package with matching bonding pads
`and output pins is required. Fortunately, the
`latest land-grid array (LGA) and ball-grid
`array (BGA) with 0.5–1 mm pitch can easily
`meet the signal density requirements. Never-
`theless, caution must be taken to minimize
`crosstalk and signal attenuation from routing
`inside the packages and through various con-
`nectors and cables.
`
`A CONTROL SYSTEM FOR A
`MEMS-BASED OXC
`
`At the heart of a high-speed large capacity
`MEMS-based OXC is a robust mirror con-
`troller. The two main objectives of the con-
`troller are the following: first, guarantee that
`new port connections are successfully estab-
`lished within the allowed switching time; sec-
`ond, guarantee uninterrupted port connection
`after the new connections are established. In
`other words, upon request the controller must
`change the tilt angle of the MEMS mirrors
`quickly (within 5–10 ms after receiving the
`command) and then maintain the new position
`of the MEMS mirrors until a new connection
`request is received. A valid connection is char-
`
`acterized by achieving an insertion loss within
`0.5 dB of optimum loss of the switch, which
`corresponds to a pointing error for each mirror
`of less than 100–200 mrad. This requirement
`alone poses a substantial challenge. Additional
`challenges come from the nonideal behavior of
`fabricated MEMS mirrors.
`
`MEMS Mirrors with Nonideal Behavior —
`The MEMS mirror system to the first order can
`be characterized by the mirror mass, the mirror
`spring constant, and the damping coefficient.
`The mechanical behavior of the mirror (i.e., its
`response to sinusoidal excitations and step
`inputs) roughly matches that of a typical spring-
`mass system. In theory, a properly behaved
`mechanical system should be straightforward to
`design. Unfortunately, these three mechanical
`system parameters are not free variables that can
`be freely chosen. The mirror mass is governed by
`the mirror size and ROC (thus mirror thickness)
`requirement of the optical system. Likewise, the
`spring stiffness is bounded by the tilt angle range
`requirement, the available peak voltage (or cur-
`rent), and the maximum actuator force output.
`The damping factor also cannot be easily tuned
`by varying the mechanical designs.
`In addition to the mechanical design con-
`straints, ideal mechanical response may not be
`readily achievable depending on the choice of
`mirror actuation method. Consider the electro-
`statically actuated MEMS mirror in Fig. 5. This
`class of actuated mirror is among the simplest to
`fabricate. However, the system is inherently non-
`linear and also unstable for large tilt angles (Fig.
`6) [15]. The unstable open-loop region begins at
`the snap-down angle, which is independent of
`spring stiffness. When a voltage greater than the
`snap-down voltage is applied to the mirror, the
`mirror will swing to the most slanted position,
`hitting the substrate below the mirror. Using
`open-loop control, the MEMS mirror simply
`cannot rest at a tilt angles greater than or near
`the snap-down angle. Alternative electrostatic
`actuator designs based on comb-drive do not
`
`IEEE Communications Magazine • March 2002
`
`85
`
`

`

`Presently there
`are numerous
`commercial
`efforts developing
`MEMS-based
`all-optical
`switches. Well-
`known subsystem
`suppliers for
`MEMS-based
`switching include
`Analog Devices,
`Corning,
`Integrated
`MicroMachines,
`OMM, and ONIX.
`
`have an equally severe stability problem; howev-
`er, a more complex fabrication process may be
`required [13].
`To complicate matters further, MEMS
`devices may not be fabricated exactly as
`designed. Real devices will have fabrication
`imperfections and variations. During operation,
`the MEMS mirrors may also experience stochas-
`tic perturbation from the environment, including
`vibration from equipment cooling fans, heavy
`truck deliveries, door slams, earthquakes, etc.),
`and even interference from neighboring MEMS
`mirrors. Therefore, only an intelligent control
`system can guarantee timely and reliable port
`connections by the MEMS mirrors.
`
`Open-Loop Control vs. Closed-Loop Control
`— Two control options are available: open-loop
`control and closed-loop control. Open-loop con-
`trol can be straightforward to implement. A rela-
`tionship between the mirror angle and the applied
`voltage must first be established by simulation or
`measurement. Then an appropriate voltage can
`be applied to the MEMS device to achieve a
`desired tilt angle. This control method requires
`minimal processing power, which is a definite
`benefit since the optical switch system must incor-
`porate shelves of electronics to control 1000 or
`more MEMS mirrors. In addition, no mirror
`angle sensing hardware is needed. However, in
`such an open-loop system, a slight calibration
`error (due to simulation or measurement error or
`fabrication nonuniformity) or electronics drift will
`lead to steady-state error in the tilt angle for
`which there is no possibility of self-correction or
`compensation. In addition, an open-loop con-
`troller cannot adequately optimize settling time
`or overshoot characteristics. In terms of system
`stability and stochastic immunity, an open-loop
`controller in fact can offer no benefit. Therefore,
`open-loop control s not a robust solution.
`From the system performance standpoint, the
`superior alternative to open-loop control is
`closed-loop feedback control. With feedback, it
`may be possible to extend the mirror tilting
`range beyond the snap-down angle. Using a
`feedback controller with a modest gain, the set-
`tling time, overshoot, and steady-state error can
`all be fine tuned according to system specifica-
`tion, even in the presence of mirror imperfection
`from nonideal MEMS fabrication. Most impor-
`tant, a MEMS mirror under feedback servo can
`be immune to random external shock and vibra-
`tion. Potential performance benefits from feed-
`back control are indeed overwhelming. However,
`an OXC with closed-loop controlled MEMS mir-
`rors requires the development of a servo-control
`algorithm, the incorporation of sensing mecha-
`nisms for computing the proper control feedback
`signal, and the implementation of control elec-
`tronics that offer sufficient computing power to
`control 1000 or more mirrors within the power
`and space budget of the switching fabric.
`
`THE ALL-OPTICAL HORIZON
`Beyond the engineering challenges already
`described, deployment of all-optical MEMS-
`based OXCs as a network element still encoun-
`ters additional hurdles. Network operators in
`
`general require switches with intelligence and
`functions such as performance-monitoring, con-
`nection verification, fault localization, bridging,
`keep-alive generation, and topology discovery
`[1]. Unlike all-optical switches, switches with
`competitive electronics-based technologies such
`as Tellium’s Aurora Optical Switch can offer
`these functions at bit rates up to 10 Gb/s (OC-
`192). However, these technologies may not be
`optimal at higher bit rates, at or above 40 Gb/s
`(OC-768), in terms of cost, power, floor space,
`and complexity. On the other hand, transparent
`all-optical switch fabrics can uniquely offer raw
`aggregate capacity independent of bit rate. The
`best solution in the long run may be an optical-
`layer switch that encompasses a transparent opti-
`cal fabric with the proper opto-electronic
`interfaces. Network architects thus carry the bur-
`den to exploit the benefits of these optical-layer
`switches.
`Presently there are numerous commercial
`efforts developing MEMS-based all-optical
`switches. Well-known subsystem suppliers for
`MEMS-based switching
`include Analog
`Devices, Corning, Integrated MicroMachines,
`OMM, and ONIX. The latter two companies,
`OMM and ONIX, have recently announced
`focusing their technology development on 2D
`MEMS switching products instead of 3D
`MEMS products. Among many different fac-
`tors, this change in development focus may be
`attributed to the more pressing need for small-
`er-size optical switches than large ones in the
`near term. Smaller less costly machines are
`expected to extend sales opportunity from the
`long haul to the metropolitan markets. In addi-
`tion, smaller-port-count machines will support
`the concept of O-o-O and O-e-O machines at
`the same node of a network.
`While many heated debates on network archi-
`tectures still have not subsided, MEMS-based
`OXCs are slowly making the move from the lab-
`oratory to the network. However, the market for
`all-optical switches to date remains very limited.
`Limited deployment of small (256 ¥ 256 or small-
`er) all-optical OXCs may take place as early as
`the first quarter of 2002. A sizeable market is
`expected to develop eventually in two to three
`years, likely followed by demand for larger-port-
`count (> 256) all-optical switches. While the
`surmounting engineering challenges for large
`OXCs seem numerous today, this market
`demand for large-port-count OXCs may be
`matched just in time by development efforts
`already underway.
`
`CONCLUSION
`MEMS technology offers the tantalizing possibil-
`ity of advanced optical crossconnects with large
`port count, scalability, and switching capacity
`that can meet the switching demands in the ever
`faster, denser, rapidly growing WDM optical
`networks today and in the future. However,
`demonstration of field-tested and qualified
`large-port-count MEMS-based optical switches is
`still in the distant future. Exquisite engineering
`is necessary to overcome challenges in areas
`such as MEMS mirror fabrication, opto-mechan-
`ical packaging, and mirror control algorithm and
`
`86
`
`IEEE Communications Magazine • March 2002
`
`

`

`implementation. While the available reliability
`data on MEMS devices from their brief history
`continue to improve, MEMS-based systems still
`must endure the test of time in order to estab-
`lish trust and confidence in the telecommunica-
`tions industry. Without a doubt, these
`engineering challenges as well as other market-
`ing challenges will be overcome in due time. As
`MEMS technology continues to advance, one
`thing is clear: the powerful impact of MEMS as
`a disruptive technology for the telecommunica-
`tions industry will never be forgotten.
`ACKNOWLEDGMENTS
`The authors would like to thank Dr. Lih Lin and
`Dr. K. Daniel Wong for many help comments
`and suggestions.
`
`REFERENCES
`[1] T.E. Stern, and K. Bala, Multiwavelength Optical Net-
`works: A Layered Approach, Reading, MA: Addison-
`Wesley, 1999.
`[2] M. C. Wu, “Micromachining for Optical and Optoelec-
`tronic Systems,” Proc. IEEE