1620
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 23, NO. 4, APRIL 2005
`
`Wavelength-Selective 1 K Switches Using
`Free-Space Optics and MEMS Micromirrors:
`Theory, Design, and Implementation
`
`Dan M. Marom, Member, IEEE, David T. Neilson, Senior Member, IEEE, Member, OSA, Dennis S. Greywall,
`Chien-Shing Pai, Nagesh R. Basavanhally, Vladimir A. Aksyuk, Daniel O. López, Flavio Pardo, Maria Elina Simon,
`Yee Low, Paul Kolodner, and Cristian A. Bolle
`
`Abstract—The design and performance of several generations
`switches are reviewed. These
`of wavelength-selective 1
`optical subsystems combine the functionality of a demultiplexer,
`per-wavelength switch, and multiplexer in a single, low-loss unit.
`Free-space optics is utilized for spatially separating the constituent
`wavelength division multiplexing (WDM) channels as well as for
`space-division switching from an input optical fiber to one of
`output fibers (1
`functionality) on a channel-by-channel basis
`using a microelectromechanical system (MEMS) micromirror
`array. The switches are designed to provide wide and flat pass-
`bands for minimal signal distortion. They can also provide spectral
`equalization and channel blocking functionality, making them
`well suited for use in transparent WDM optical mesh networks.
`
`Index Terms—Gratings, microelectromechanical devices, micro-
`electromechanical system (MEMS), optical add/drop multiplexing
`(OADM), optical filters, optical switches, wavelength-selective
`switch.
`
`I. INTRODUCTION
`
`T RANSPARENT switching, where the optical signal
`
`does not undergo conversion to the electrical domain for
`switching purposes, can greatly simplify and reduce the cost
`of implementing optical networks by the elimination of optical
`to electrical to optical (OEO) conversions [1], [2]. The use of
`transparent switching within wavelength division multiplexed
`(WDM) systems further necessitates that switches be either
`wavelength-selective or be preceded by a demultiplexer and
`followed by a multiplexer for channel access [3]. The former
`is typically more desirable in many switching scenarios, since
`it avoids multiple components and will typically have lower
`losses and wider passbands. At an optical add/drop multiplexer
`(OADM) node, a subset of the optical channels, or wavelengths,
`propagating in the optical fiber is extracted for local detection
`(known as drop channels) and new optical channels are inserted
`in their place (known as add channels). The optical add/drop
`functionality can be achieved by the use of a channel blocking
`
`Manuscript received June 7, 2004; revised January 7, 2005.
`D. M. Marom and D. T. Neilson are with Bell Laboratories, Lucent Technolo-
`gies, Holmdel, NJ 07733 USA (e-mail: dmarom@lucent.com).
`D. S. Greywall, C. S. Pai, N. R. Basavanhally, V. A. Aksyuk, D. O. López,
`F. Pardo, M. E. Simon, Y. Low, P. Kolodner, and C. A. Bolle are with the Bell
`Laboratories, Murray Hill, NJ 09999 USA.
`Digital Object Identifier 10.1109/JLT.2005.844213
`
`filter [4]–[6] placed between a passive splitter (for dropping
`channels) and a passive combiner (for adding channels). The
`device blocks the dropped channels from continuing to propa-
`gate in the line system and interfering with the added channels.
`A more efficient solution utilizes a wavelength-selective 2
`2
`switch [7], [8]. This switch has two inputs, the line system
`input, and the add channels and two outputs, the line system
`output, and the drop channels. These wavelength-selective
`switches use internal switching elements to route the individual
`WDM channels to the proper port.
`As optical networks evolve from simple ring architecture
`with OADM nodes to optical mesh networks [1], the trans-
`parent switching requirements change. Mesh network nodes
`are typically linked to three or four neighboring nodes with
`each link carrying two-way traffic. Transparent switching at
`each node’s network links, or cross connect functionality, is
`required for implementing an all-optical network. Furthermore,
`a modular cross connect fabric may be more desirable from
`an economic standpoint, as the node interconnecting links are
`deployed gradually. Finally, the cross connect may be required
`to support a power equalization feature [9] for optimal optical
`transport.
`The wavelength-selective
`switch fulfills all the mesh
`networking requirements above [10]–[15]. The switch has a
`single input fiber that carries the WDM signal consisting of
`channels, and distributes these
`channels in a reconfig-
`urable and independent fashion across the
`output fibers.
`The switches [10]–[14] use a microelectromechanical system
`(MEMS) mirror array for the beam steering elements. Owing
`to the reciprocal nature of light propagation, the same switch
`fabric may be operated in reverse for wavelength-selective
`switching functionality. A complete wavelength-se-
`lective
`cross connect ( WDM inputs and WDM
`outputs) is implemented by utilizing
`switch modules and
`passive splitters [16]. A wavelength-selective
`cross
`connect can also be constructed using blocking filters [5],
`[6], [17], but has additional loss since it requires both passive
`splitting and combining and further requires
`blocking filters
`component count.
`In this paper, we review the technology of the wavelength-se-
`lective
`switch, from design choices and tradeoffs, through
`a description of various switch implementations we constructed,
`to the performance the switches exhibited.
`
`0733-8724/$20.00 © 2005 IEEE
`
`FNC 1044
`
`

`
`MAROM et al.: WAVELENGTH-SELECTIVE
`
`SWITCHES USING FREE-SPACE OPTICS AND MEMS MICROMIRRORS
`
`1621
`
`Fig. 1. Optical system configuration for wavelength-selective switch. The
`system is composed of a subsystem that converts fiber position to angle and a
`second subsystem, which uses a lens and diffraction grating to provide spatial
`dispersion to separate the channels.
`
`II. DESIGN OF WAVELENGTH-SELECTIVE
`
`SWITCHES
`
`switch design is based on the
`The wavelength-selective
`guiding principle of optical imaging, leading to simple assembly
`and alignment. The optical system partitions the aperture to pro-
`vide for multiple ports [7] and uses a coaxial relay imaging
`system to map the input beams onto the MEMS micromirror
`array and back. The coaxial arrangement facilitates assembly
`and packaging, as all elements are aligned along one-dimen-
`sional space, and can be housed in a robust tubular holder (sym-
`metric structure with no weak bending axis). It is useful to con-
`sider the switch to be comprised of two major subassemblies
`(Fig. 1). The role of the first subassembly is to image the input
`and output optical fiber end faces onto a common magnified spot
`B. This subassembly converts the distinct spatial locations of the
`fibers to unique angular propagation directions at position B. A
`tilting mirror could be placed at this image plane to reflect the
`switch
`light and implement a nonwavelength-selective
`[18]. Here the light originating from the input fiber A would be
`imaged on the desired output fiber F. Attenuation level control
`can be obtained by deliberately misaligning the image location
`from the output fiber, by tilting this mirror away from the ideal
`coupling angle. The second subassembly introduces the desired
`wavelength-selectivity property. It spatially disperses the input
`magnified common spot, consisting of the WDM channels,
`onto the MEMS micromirror array, such that each channel is
`imaged upon a separate mirror in the array for independent ad-
`dressing. Each micromirror in the array is tilted to a desired
`angle, which determines the output fiber to which the reflected
`light will couple upon imaging back to the fiber array, on a
`WDM channel basis. A typical beam path in the switch origi-
`nates from the input fiber A and is imaged with magnification to
`B by the first subassembly. The second subassembly images one
`of the WDM channels from B to D, according to wavelength.
`The micromirror at D is tilted to a prescribed angle, and the
`reflected light that is propagating in a new direction is imaged
`back to point B by the second subassembly. Finally, the first sub-
`assembly images the reflected signal to the output fiber location
`F by a last imaging operation.
`Due to the independent imaging operations each subassembly
`performs, we may analyze the operation of each subassembly
`independently. The first subassembly determines the magnifica-
`tion ratio, the fiber array layout, and the required mirror tilts to
`
`Fig. 2. Position to angle subsystem showing unequal spacing of fibers and
`lenses to introduce gaps for the variable attenuation function and polarization
`diversity optics.
`
`reach each output fiber. The second subassembly determines the
`amount of spatial dispersion for separating the WDM channels
`and obtaining the necessary passband width. We will establish
`the characteristics of each subassembly, as well as the overall
`design tradeoffs of the wavelength-selective switch.
`
`A. Position-to-Propagation Angle Subassembly
`The optical subassembly responsible for imaging the optical
`fiber end faces onto a common magnified spot is comprised of a
`fiber array, a matching microlens array, polarization diversity
`optics, and a condenser lens whose aperture subtends all the
`beam apertures from the fibers (Fig. 2). The fiber array con-
`fibers, where one fiber is assigned to carry the
`sists of
`fibers are the output fibers.
`input signal, and the remaining
`The optical axes of the individual lenses and fibers are coaxially
`aligned and arranged in a one-dimensional array to accommo-
`date mirrors with a single tilt axis. Furthermore, the fibers and
`lenses are placed on an irregularly spaced grid to introduce gaps
`between some of the lens apertures. This supports the attenua-
`tion function without giving rise to crosstalk. We also employ a
`polarization-diversity solution to eliminate the polarization sen-
`sitivity of the diffraction grating employed in the second op-
`tical subassembly. The polarization diversity is provided by an
`anisotropic uniaxial crystal and a half-wave plate. The uniaxial
`crystal separates an input beam into two distinct (non overlap-
`ping), copropagating, orthogonally polarized beams. The half-
`wave plate rotates the polarization state of one of the beams
`such that the two beams are copolarized. The two beams prop-
`agate within the optical subsystem, and are merged back to a
`single-beam before coupling to the selected output fiber by the
`waveplate and uniaxial crystal combination. Due to the imaging
`operation, the two beams exchange their positions in the return
`path toward the output fibers. Thus, path length differences be-
`tween the two beams, as experienced by the beam traversing
`the half-wave plate, are compensated in the return path. This
`ensures that the system will also have low polarization mode
`dispersion (PMD).
`The optical arrangement of the first subassembly implements
`a telescopic imaging system via the lenses from the microlens
`
`

`
`1622
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 23, NO. 4, APRIL 2005
`
`).
`) and the condenser lens (focal length
`array (focal length
`The imaging operation magnifies the optical beam emerging
`. The
`of the
`from the single mode fiber by factor
`lenses in the microlens array are matched to the optical beam’s
`numerical aperture (NA). Using a Gaussian beam waist of 10.5
`m for the beam from a single-mode fiber, the lens
`should
`. It is
`be at most 3.5 to prevent significant beam clipping
`desirable to pack the microlenses in the lens array as tightly as
`of the condenser lens as large as pos-
`possible, to keep the
`sible. At minimum, the microlens pitch will equal the individual
`, and the condenser lens aperture is, therefore,
`lens diameter,
`. However, for implementing the spectral equaliza-
`tion functionality, increased spacing between microlenses is re-
`quired to allow for attenuation by beam displacement for in-
`tentional imperfect coupling to the output fiber. To conserve the
`condenser lens aperture, the intermicrolens spacing, or gaps, are
`inserted between every pair of microlenses. This ensures that
`there is a gap available to only one side of each microlens in the
`2
`array for attenuation by beam displacement, and a total of
`gaps, where the symbol
`denotes the div operation.
`The intermicrolens spacing, or gap size, is determined from
`the required attenuation dynamic range and minimum crosstalk
`requirements. At the maximal attenuation setting required, we
`must still suppress the crosstalk to the adjacent output fiber. We
`define the microlens pitch at locations where a gap is inserted
`(Fig. 3). The gap size is, therefore,
`. For a given
`as
`from the microlens optical axis, the attenuation
`beam shift
`to the desired output fiber and the crosstalk to the neighboring
`fiber is calculated by the power overlap integral [19], yielding
`
`and
`
`(1)
`
`(2)
`
`. Note
`where the collimated beam profile is denoted by
`that the finite extent of the microlens aperture assists in the at-
`tenuation functionality, as a fraction of the beam power is lost.
`In our designs, the criteria are for 10-dB attenuation range while
`maintaining crosstalk below 40 dB. Using a Gaussian beam ap-
`3.5,
`proximation for the beam profile and microlenses of
`is approximately 1.5 times the lens
`then the necessary pitch
`(or gap size is one half the lens diameter).
`diameter
`The condenser lens aperture with added gaps of half diameter
`. Given the
`size is, therefore,
`condenser lens aperture and focal length, we can now evaluate
`as
`its
`
`(3)
`
`Fig. 3. Position-to-angle conversion optics showing configuration of optics for
`providing variable attenuation function while maintaining maximum density.
`The displacement of the beam x causes attenuation by changing the coupling
`angle at the fiber.
`
`ality, no gaps are required, and the denominator of (3) simplifies
`. Since
`is determined by the required functionality,
`to
`the only free parameter available for the designer is the magni-
`fication factor. A high magnification factor would be desirable
`for implementing the condenser lens. This would also reduce the
`mirror tilt angle ranges. If the input fiber is in the middle of the
`fiber array, as in Figs. 1 and 3, then the mirror must tilt roughly
`, in radians. Alternatively, the
`within the range
`input fiber may be at the edge of the array, requiring the mirror to
`.
`tilt in only one direction but at a doubled range of
`However, as the magnification ratio
`increases, the resulting
`mode size at the output of the first subassembly also increases.
`This places a burden on the second subassembly responsible for
`the spectral resolution, as described in Section II-B.
`
`B. Spatial Dispersion Subassembly
`The second optical subassembly spatially disperses the mag-
`nified mode that was generated by the first subassembly and im-
`ages it on the micromirror array. Its design is similar to a spec-
`trometer with a Littrow mounted grating. A single lens colli-
`mates the light that is then incident on the grating. The diffracted
`light, which is propagating back toward the lens and is angu-
`larly dispersed, is imaged by the same lens onto the micromirror
`array. As is well known from classical spectrography, the spec-
`tral resolution of the instrument increases with increasing focal
`length and grating frequency, and with decreasing input slit size.
`In our switch, the magnified mode size is equivalent to a spec-
`trograph slit size. Therefore, it is desirable to minimize the spot
`) for obtaining high spectral resolution.
`size (decrease
`The spatial dispersion, expressed in meters per hertz, pro-
`vided by the second subassembly is given by [20]
`
`decreases as the magnification factor
`The condenser lens
`decreases and as the number of output fibers
`increases.
`If there is no need to support the spectral equalization function-
`
`is the
`is the center wavelength of the WDM system,
`where
`is the speed of light, and is
`focal length of the resolution lens,
`
`(4)
`
`

`
`MAROM et al.: WAVELENGTH-SELECTIVE
`
`SWITCHES USING FREE-SPACE OPTICS AND MEMS MICROMIRRORS
`
`1623
`
`simple expressions for the passband and stopband widths of a
`WDM channel using (6) with the approximation that the con-
`tribution of the second error function is constant, which is valid
`when designing for flat and wide passbands. The passband width
`normalized by the channel spacing and measured at
`transmissivity level (a characteristic level is
`0.5 or
`3 dB
`passband) is defined by
`
`erf
`
`(7)
`
`Fig. 4. Schematic of magnified Gaussian mode at a single frequency
`component imaged on the mirrors of width D and pitch P . Mode size is
`M times larger than the output of a single mode fiber, 2! , where M is the
`magnification factor of the imaging system.
`
`is the inverse error function. Similarly, the stopband
`where erf
`, originating from crosstalk of neighboring mirrors
`width
`(or adjacent channels) is
`
`the Littrow grating mounting angle. The Littrow angle is given
`, where
`is the grating spatial frequency
`by
`). Therefore, the mirror pitch of the micromirror array
`(in
`, where
`is the WDM channel fre-
`will be
`quency spacing (in Hz). Note that we are assuming constant spa-
`tial dispersion across the total bandwidth of the optical system.
`In reality, especially for gratings of high spatial frequency, the
`mirror pitch will not be constant due to the wavelength depen-
`dence in the grating diffraction formula [20].
`The spatially dispersed image of the magnified Gaussian
`mode, Fig. 4, present on the micromirror array can be expressed
`as
`
`(5)
`
`where
`is the Gaussian mode field radius of the beam from a
`in (5) defines
`single mode fiber (5.25 m). The term
`the center location of the magnified Gaussian mode as a function
`of mirror
`of the temporal frequency. The dimensionless ratio
`size to the magnified Gaussian mode size
`measures how well the Gaussian mode is confined within the
`micromirror, and will be shown to determine the passband per-
`formance. The frequency-dependent, power-coupling efficiency
`integral is calculated by performing the traditional overlap inte-
`gral over the extent of a single mirror at the device plane [19].
`More elaborate modeling taking into account the effect of the
`neighboring mirror states has been performed elsewhere [21].
`With the simple model, the coupling efficiency is defined by
`
`erf
`
`erf
`
`(6)
`
`is the physical width of the micromirror in the spa-
`where
`tial dispersion direction, and the mirror is modeled as infinite in
`is slightly smaller
`the orthogonal direction. The mirror size
`, due to the presence of a gap to pre-
`than the mirror pitch
`vent physical contact between adjacent mirrors. We can derive
`
`erf
`
`(8)
`
`10
`The stopband width is typically measured at the
`40 dB level. The two parameters influencing the passband
`and
`and stopband widths are the fill factor of the array
`the confinement ratio . It is desirable to maximize both band-
`widths for minimal signal distortion and crosstalk, which can be
`satisfied by an increasing confinement ratio . A high fill-factor
`micromirror array also maximizes the passband width, yet de-
`creases the stopband width. Nevertheless, the mirror arrays are
`typically fabricated with minimal gap size as technically fea-
`sible for maximizing the passband width. In the limiting case of
`, the fill-factor approaches 1 and the confinement pa-
`is the only parameter determining the passband shape,
`rameter
`controlling the extent of the passband flatness and the roll-off
`rate (Fig. 5). Thus, passband requirements can be accommo-
`dated by varying the ratio of the mirror size to the magnified
`Gaussian mode size, which sets .
`The available degrees of freedom remaining in designing the
`second optical subassembly is choice of diffraction grating and
`lens focal length. For obtaining high spectral resolution within a
`small package, it is always desirable to select a high spatial fre-
`quency diffraction grating. Other factors influencing the grating
`selection process are the diffraction efficiency and polarization
`dependence in the telecom (1500–1620 nm) wavelength range.
`We employ polarization diversity in our switch, implemented
`in the first subassembly, as the chosen grating does have sig-
`nificant polarization dependence. Once the magnification factor
`of
`and diffraction grating have been selected, the focal length
`the resolution lens can be established to meet the passband per-
`is equal to the condenser’s,
`formance metrics. The lens’s
`as evaluated by (3). However, this lens has a field of view diam-
`eter determined by the physical extent of the micromirror array,
`. These two requirements, combined with the spec-
`or
`tral range, imply that the resolution lens will require multiple
`elements to obtain good imaging characteristics, and warrant a
`custom design.
`As outlined above, the design process of a wavelength-selec-
`switch is straightforward. Given the switch require-
`tive
`ments (number of output fibers, spectral equalization dynamic
`range, and channel passband characteristics), the optical param-
`eters are established. The designer can vary the magnification
`factor
`, which will affect the lenses’
`, overall system size,
`
`

`
`1624
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 23, NO. 4, APRIL 2005
`
`Fig. 5. Calculated passbands as a function of the confinement parameter
`. Assumes a 100% fill-factor micromirror array, and effect of neighboring
`is the
`mirror is neglected.  is the ratio of mirror size to beam size and 
`channel-to-channel frequency spacing.
`
`, until a
`and micromirror scan range, and confinement factor
`suitable design space is achieved. However, having so few crit-
`ical parameters influencing the switch design is often too re-
`strictive. We introduce anamorphic optics to obtain an additional
`design parameter that may lead to more efficient switch imple-
`mentations.
`
`C. Anamorphic Optics
`
`The use of free-space optics in our switch design allows us
`to better utilize the three-dimensional volume of the switch
`package. We observe that the lenses’
`is determined by the
`beam apertures. In addition,
`extent of the linear array of
`the spectral resolution is determined by the magnified Gaussian
`beam width in the dispersion direction only. We introduce
`anamorphic optics to convert
`the circular Gaussian beam
`profiles to elliptical ones. The anamorphic optical elements
`are inserted into the first optical subassembly to generate a
`magnified elliptical beam whose narrow axis is in the spatial
`dispersion direction of the second optical subassembly. This
`ellipse orientation continues to satisfy the minimal beam size
`requirement for the spectral resolution.
`The anamorphic elements are placed between the microlens
`array and the condenser lens, in the collimated beam region, and
`.
`serve to compress the beams’ vertical dimension by factor
`The fiber and microlens arrays are also both oriented vertically
`when employing the anamorphic optics (Fig. 6). Orienting the
`fiber array vertically means that the MEMS mirrors must tilt
`about an axis parallel to the dispersion direction. This is desir-
`able for maximizing the channel passband [21] and reducing the
`sensitivity to mirror curvature [22]. The anamorphic effect re-
`beam apertures by
`, increasing
`duces the extent of the
`the condenser lens
`
`to
`
`(9)
`
`Fig. 6. Schematic of optical system showing the effect on beam size and shape
`of inserting anamorphic optics. The anamorphic optics ensures that high spectral
`resolution can be maintained while the apertures of the optics are minimized.
`
`of the resolution lens,
`The benefit further extends to the
`as well as to the micromirror scan range, without effecting the
`. Therefore, we can now modify three pa-
`confinement factor
`rameters in order to reach a desirable switch design; the mag-
`and confinement ration for determining the
`nification ratio
`for controlling
`spectral resolution, and the anamorphic ratio
`.
`the lenses’
`The advantages listed above may lead to a conclusion that
`the use of anamorphic optics is purely beneficial. However,
`the magnified elliptical beam requires the mirrors in the mi-
`. This can make the
`cromirror array to be longer by factor
`design of the MEMS mirrors more difficult since they are
`now longer in the direction of tilt, will have greater mass,
`and be more susceptible to curvature. Furthermore, the mirror
`resolution in tilt angle is also finer, requiring greater precision
`in mirror positioning.
`
`III. IMPLEMENTATION OF WAVELENGTH-SELECTIVE
`SWITCHES
`
`We have realized three successful generations of wavelength-
`switches. These switches were designed to
`selective
`support the switching functionality, and provide wide and flat
`passbands for minimal signal filtering. Filtering is particularly
`critical since it is expected that signals will pass through mul-
`tiple switches, and concatenated filtering will narrow the system
`passband [17], [23].
`Common to all our switches is the support of a WDM system
`operating at the extended -band (1554–1608 nm). Low inser-
`tion losses were achieved by using an 1100 lines/mm grating
`with high diffraction efficiency in the grating’s S-plane [24]
`(polarization perpendicular to the groove direction), along with
`the aforementioned polarization diversity. The grating was Lit-
`60.5 (angle for center wavelength
`trow-mounted at angle
`1582 nm). The two logical subassemblies describe above
`were implemented as physical subassemblies, since this made
`building various optical configurations more practical. Partic-
`ular description of each switch version is provided below.
`
`

`
`MAROM et al.: WAVELENGTH-SELECTIVE
`
`SWITCHES USING FREE-SPACE OPTICS AND MEMS MICROMIRRORS
`
`1625
`
`Fig. 7. Electrostatically actuated MEMS mirrors for wavelength-selective
`switches (one micromirror of high-fill-factor array shown). A torsional mirror
`with a rotation axis orthogonal to the dispersion direction is shown in (A). Both
`mirrors in (B) and (C) tilt about an axis parallel to the disperion direction. The
`design in (B) uses a double hinged actuator while that in (C) is fringing-field
`actuated.
`
`A. 128-Channel 1
`
`4 Wavelength-Selective Switch
`
`Our first generation wavelength-selective switch was config-
`ured with a single input and four output fibers. The switch sup-
`ported 128 channels spaced on a 50-GHz grid, and was designed
`to provide channel bandwidth support of 10-Gb/s transmission
`rates.
`The first optical subassembly implemented an imaging
`3.3. No anamorphic optics
`system with magnification of
`were used. The five fibers and lenses were tightly packed,
`as this switch was not designed to support dynamic spectral
`equalization. Using (3) with a correction to the denominator
`due to the tight lens packing, the theoretical condenser lens
`is
`2.3. In practice, an
`of 2 was required, due to the
`increased aperture requirement for the polarization diversity.
`The input fiber was placed at the center of the array, requiring
`5.6 . A five-element, 100-mm
`a micromirror tilt range of
`focal-length resolution lens was designed for the second optical
`subassembly with the prescribed aperture and field diameter.
`The lens and grating combination provide a spatial dispersion
`of 1.86 m/GHz, resulting in a micromirror pitch of 93 m,
`12-mm long. In practice, the mirror
`and the entire array was
`pitch varied from 82 to 108 m due to the nonlinearity in the
`grating’s angular dispersion. The resolution lens focal length
`2.7.
`was chosen to provide a confinement ratio of
`The switch employed a MEMS mirror array, with the mir-
`rors tilting in the direction of the spatial dispersion [Fig. 7(a)].
`The MEMS mirrors were etched in a 1- m-thick silicon-on-
`insulator (SOI) wafer and had two 0.5- m-thick torsion rods
`defining their rotation axis. The SOI mirror chip was flip-chip-
`bonded onto an electrode chip with a 10- m spacer. The mirrors
`were actuated by an electrostatic attractive force imposed by one
`of two electrodes below each mirror on either side of the axis.
`8 ,
`These MEMS micromirrors were designed to rotate up to
`200 V dc applied to either mirror electrode. A
`at a voltage of
`2- m gap between the mirrors provided a 98% fill-factor for the
`array in the spatial dispersion direction.
`
`Fig. 8. Picture of the assembled 128-channel 50-GHz channel spacing 1  4
`wavelength-selective switch. The 100-mm focal length lens is in the center with
`the 1100-lines/mm grating to the right.
`
`The switch prototype was assembled on an optical table
`350 mm.
`(Fig. 8). The length of the optical system was
`
`B. 64-Channel 1
`2 Wavelength-Selective Switch With
`Spectral Equalization
`Our second-generation wavelength-selective switch was con-
`figured with single-input and two-output fibers. The switch sup-
`ported64channelsspacedona100-GHzgrid,andwasdesignedto
`providechannelbandwidthsupportof40-Gb/stransmissionrates.
`One key objective set for the design of the switch was to re-
`duce its physical size. This was achieved primarily by reducing
`the focal length of the resolution lens to 50 mm. However, the
`spectral dispersion of the second subassembly was halved to
`0.93 m/GHz by this action. Since the channel spacing was
`doubled to 100 GHz, the micromirror pitch remained at 93 m,
`6-mm long. The requisite channel
`and the entire array was
`passbands were achieved by increasing the confinement ratio
`3.2, implying a reduction in the magnification ratio to
`to
`2.75. The lenses’
`was maintained sufficiently high by
`using one of the fibers both as an input and an output through
`the use of an optical circulator. Thus, the switch utilized only
`two fibers, resulting in a reduced aperture requirement. Using
`1 yields a theoretical
`of 4.2, but
`(3) with a value of
`was 2.6 in practice due to the polarization diversity. The first
`optical subassembly utilized a dual-fiber collimator (two fibers
`placed at the lens’ front focal plane), followed by an adjustment
`prism to maintain parallelism for the two collimated beams. The
`prism was placed at a location that defined the necessary beam
`separation to provide for the spectral equalization functionality.
`The micromirrors were required to tilt in the direction orthog-
`onal to dispersion, in support of the spectral equalization func-
`tionality. To reduce the electrical I/O requirements, the mirrors
`utilized single-sided actuation (one electrode per mirror). The
`, or
`7 (using theoret-
`mirror tilt range was, therefore,
`value of 4.2). A four-element, 50-mm resolution lens
`ical
`was designed for the second optical subassembly.
`Two different MEMS micromirror arrays were designed for
`the switch; one based on surface micromachining of polysilicon
`and the other on bulk processing of a SOI wafer [25], [26]. In
`the polysilicon approach, a double-hinge activation mechanism
`is defined. An actuation plate, anchored at one edge, is tilted
`to small angles via an underlying parallel plate electrode and
`
`

`
`1626
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 23, NO. 4, APRIL 2005
`
`Fig. 9. Picture of an assembled 64-channel 1  2 wavelength-selective switch
`with spectral equalization. The housing is 190-mm long and 44-mm outer
`diameter.
`
`a 4- m gap. The mirror is attached to the free edge of the ac-
`tuation plate and to the substrate with unequal arm lengths, al-
`lowing large mirror tilt angles out-of-plane via angle amplifi-
`cation [Fig. 7(b)]. The mirror, actuator plate, and springs are
`etched in the 1.5- m-thick polysilicon. Spring features are typ-
`ically 0.5- m wide, and the gap between adjacent mirrors is 0.7
`m (
`99
`fill ratio). In the SOI approach, a 10- m layer of
`polysilicon is deposited over the patterned 1- m-thick single-
`crystalline silicon and is used to define the actuator electrodes
`and ground shields. The electrode attracts the short actuator arm
`via an electrostatic fringing field, resulting in mirror rotation
`out-of-plane about the torsion springs [Fig. 7(c)]. The mono-
`lithic structure does not exhibit rotational snap-down.
`As shown in Fig. 9, the switch was packaged in a Super-
`Invar tube to make it insensitive to temperature variations. The
`190-mm length, and it
`tube size was 44-mm outer diameter
`weighed 1.25 Kg.
`
`C. 64-Channel 4
`1 Wavelength-Selective Switch With
`Spectral Equalization and Anamorphic Optics
`
`Our third-generation wavelength-selective switch was config-
`ured with four input fibers and a single output fiber (the switch is
`4, as the optical path is reciprocal).
`physically identical to a 1
`This configuration is most appropriate for implementing cross
`connect functionality among four WDM systems with broad-
`cast capability and hitless switching [16]. This switch, also sup-
`ported 64 channels spaced on a 100-GHz grid, and was designed
`to provide channel bandwidth support of 40-Gb/s transmission
`rates.
`The form factor and spectral resolution optics of our
`second-generation switch were preserved in this new five-fiber
`switch, which is shown in Fig. 10. Anamorphic optics was,
`therefore, added in the first subassembly to support the higher
`. The fiber
`fiber count without changing the resolution lens’
`and matching lens array were irregularly spaced to support
`the spectral equalization functionality (two gaps of half lens
`3,
`diameter inserted in arrays). The anamorphic ratio was
`which was achieved with a prism pair. The magnification ratio
`2.5. The chosen parameters lead to th