JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 12, DECEMBER 2000
`
`1845
`
`Holographic Optical Switching: The “ROSES”
`Demonstrator
`
`W. A. Crossland, I. G. Manolis, M. M. Redmond, K. L. Tan, T. D. Wilkinson, M. J. Holmes, T. R. Parker, H. H. Chu,
`J. Croucher, V. A. Handerek, S. T. Warr, B. Robertson, I. G. Bonas, R. Franklin, C. Stace, H. J. White, R. A. Woolley,
`and G. Henshall
`
`Abstract—The design, assembly, and performance of a proto-
`8 free-space switch demonstrater using reconfigurable
`type 1
`holograms are reported. Central to the switch fabric is a ferro-
`electric liquid crystal (FLC) on silicon spatial light modulator
`1 array of highly reflective and
`(SLM) deposited with a 540
`planar mirror strips. The input and output ports of the switch are
`fabricated as a linear array of silica planar waveguides connected
`to single-mode fibers, and the holographic beam-steerer operates
`without the need for adjustment or dynamic alignment. The
`waveguide array and the single Fourier transform lens for the
`2 holographic replay system are housed in an opto-mechanical
`mount to provide stability. The switch operates at 1.55 m
`60 nm. The
`wavelength and has a designed optical bandwidth of
`first measured insertion loss and crosstalk figures are 16.9 dB
`19.1 dB, respectively. Improvements in SLM performance,
`and
`the use of new addressing schemes and the introduction of better
`alignment
`techniques are expected to improve these figures
`3 optical
`considerably. The preliminary performance of a 3
`crossconnect is also presented to show that this technology is
`switching fabrics.
`scalable to
`Index Terms—Ferroelectric liquid crystal, optical crossconnect,
`optical switch, reconfigurable hologram, spatial light modulator,
`waveguide.
`
`I. INTRODUCTION
`
`T HE rapidly evolving demands of telecommunications
`
`and avionics systems have created a new market for
`high-capacity all-optical photonic switches. The scale of the
`networks proposed leads ultimately to requirements for very
`large switching fabrics (e.g.,
`and
`). Currently,
`there are several different technologies being developed to
`implement such systems, including thermo-optic waveguide
`(TOW) switches [1], mechanical switches, and micro-electro
`mechanical systems (MEMS) [2]. An alternative approach
`potentially capable of realising such systems uses holographic
`optical beam deflectors based on binary and multiple-phase
`
`Manuscript received April 7, 2000; revised October 2, 2000.
`W. A. Crossland, I. G. Manolis, M. M. Redmond, K. L. Tan, and T. D.
`Wilkinson are with Cambridge University Engineering Department, Trump-
`ington Street, Cambridge CB2 1PZ, U.K. (e-mail: wac@cam.ac.uk).
`M. J. Holmes, T. R. Parker, H. H. Chu, J. Croucher, and V. A. Handerek are
`with the Department of Electrical and Electronic Engineering, Kings College
`London, WC2R 2LS, U.K.
`S. T. Warr, B. Robertson, and I. G. Bonas are with Thomas Swan and Co.
`Ltd., Crookhall, Consett, County Durham DH8 7ND, U.K.
`R. Franklin, C. Stace, and H. J. White are with BAe Systems, Filton, Bristol,
`BS12 7QW, U.K.
`R. A. Woolley is with CRL, Dawley Road, Hayes, Middlesex UB3 1HH, U.K.
`G. Henshall is with Nortel Telecommunications, London Road, Harlow,
`Essex CM17 9NA, U.K.
`Publisher Item Identifier S 0733-8724(00)10997-1.
`
`liquid crystal spatial light modulators [3]. These are moderate
`speed, optically transparent, switches capable of handling any
`data format or bit rate. In addition, as the light is deflected
`through free-space, multiple signal beams can be simulta-
`neously interconnected allowing the switch to be scaled up
`to hundreds of channels. Such switches are ideally suited to
`applications in network protection and restoration after failure,
`dynamic connection provisioning, and polling networks of
`optical sensors.
`In this paper, the design and operation of a
`free-space
`holographic optical switch developed for sensor polling will be
`described. In addition, in order to show that this technology is
`scalable to
`switches, the preliminary results for a bench
`top
`switch will also be presented.
`switch uses a hologram recorded onto a ferroelectric
`The
`liquid crystal over silicon spatial light modulator (FLC/Si SLM)
`[3], [4] to steer the incoming beam to the desired output port [5].
`The binary FLC is ideally configured as a reflective half-wave
`plate. FLCs have been shown to record binary-phase holograms
`[6] and to replay lossy binary phase holograms without polar-
`ization sensitivity regardless of the wave-plate thickness and
`FLC switching angle [7]. A previous free-space optical switch
`demonstration using FLC on glass holograms in a
`transmis-
`sive bench set-up has been reported [8]. The ROSES [9], [10]
`switch demonstrator uses a reflective FLC/Si SLM as the holo-
`gram recording device; a waveguide array (WGA) to provide
`the input and output ports, and a custom-made opto-mechanical
`mount with six-axis alignment adjusters to house the WGA, a
`single collimation/transform lens, and the SLM.
`Although prototype photonic switches have already been
`demonstrated using thermo-optic waveguide and MEMS-based
`technologies, it is believed that holographic beam deflecting
`switches offer a more scalable and reliable approach to pro-
`ducing high-capacity all-optical photonic switches. Thus,
`throughout this paper the advantages and challenges associated
`with this technology will be discussed and compared against
`these competing technologies.
`
`II. DESIGN OF ROSES SWITCH
`A. Operation of Switch
`The primary aim of the ROSES project was to develop the
`technology required to implement a large scale
`holo-
`graphic beam deflecting switch. As a first step toward this goal,
`a prototype polarization independent linear
`fiber optic
`routing switch has been demonstrated. The switch was designed
`
`0733–8724/00$10.00 © 2000 IEEE
`
`FNC 1005
`
`

`

`1846
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 12, DECEMBER 2000
`
`Fig. 1. Operation of the ROSES 1  8 holographic routing switch. SLM deflects signal beam through an angle ' dependent on period of hologram. Can control
`which output port light is diffracted into.
`
`to take a signal beam from a single-mode fiber, and dynam-
`ically route it to one of eight single-mode output fibers. The
`switch is shown schematically in Fig. 1, and consists of four
`main components: a waveguide array providing spatial fan-in to
`and from the input and output fibers; a Fourier lens, a reflec-
`tive binary-phase ferroelectric liquid crystal silicon backplane
`SLM that acts as a beam-steering element, and a custom inter-
`face board. The Fourier lens was a commercially available com-
`ponent, while the silicon backplane and silica-on-quartz wave-
`guide array were custom-designed within the project and fabri-
`cated at external foundries. Custom optomechanics were used
`to align and control the relative positions of all optical compo-
`nents. In addition, the optomechanics, ribbon fiber connectors
`and electronic interface board were all mounted on a common
`aluminum baseplate.
`The basic operation of the switch is as follows: the optical
`signal is launched down the input fiber, and couples via a con-
`nector into the ribbon fiber. The signal then couples into the
`input waveguide and propagates to the launch end of the wave-
`guide array, from where it is collimated (Fig. 1) by the Fourier
`lens onto the SLM. The SLM acts like a phase-only diffrac-
`tion grating with a tuneable period and pattern, creating a de-
`, of the reflected beam. By controlling the de-
`flection angle,
`flection angle, the beam may be made to return to a selected
`point on the waveguide array, and hence couple into the chosen
`output waveguide. The output signal then propagates across the
`waveguide array and into the ribbon fiber and output fiber. The
`deflection angle depends on the phase modulation pattern (the
`hologram), which is downloaded from a PC via an electronic in-
`terface board. The maximum deflection angle for 20 m pixels
`is 2.22 at 1550 nm.
`Due to the binary-phase nature of the hologram, light is dif-
`1 and
`1 orders. The
`fracted symmetrically into both the
`1 and
`1 orders al-
`switch was designed such that both the
`ternately carried the optical signal (i.e., the 1st, 3rd, 5th, and 7th
`
`channels were to the left of the input channel, and the 2nd, 4th,
`6th, and 8th channels were to the right of the input waveguide, as
`shown in Fig. 1). In order to minimize crosstalk caused by the
`unwanted symmetric order, the waveguide was designed with
`the unwanted orders landing between the output waveguides.
`
`B. The Liquid Crystal over Silicon (LCOS) Spatial Light
`Modulator
`The SLM consists of a linear array of 540 pixellated elec-
`trodes, each 18 m 6 mm, with a 2 m dead space between
`the pixels (Fig. 2). On the silicon backplane, each pixel is con-
`nected to an integrated SRAM pixel drive circuit and suitable
`access circuitry to which the required hologram patterns were
`down-loaded from a custom interface board (the PC controlled
`frame store shown in Fig. 1). An extra silicon-processing step
`was carried out within the project, to lay down a layer of optical
`quality aluminum “mirror” over the pixels [11]. Light incident
`on the SLM passes through a layer of birefringent liquid crys-
`talline material [12] with an optical axis orientation controlled
`by the voltage applied to the pixel. It is then reflected from
`the aluminum mirror and passes back through the liquid crystal
`layer. The first-order diffraction efficiency of a binary-phase
`, and molecular tilt angle
`hologram with FLC layer thickness
`is given by (1), where
`is the birefringence and
`is the
`operating wavelength [7]
`
`(1)
`
`Equation (1) shows that efficiency is maximized when the de-
`vice acts as a half-wave plate in reflection at the wavelength
`of operation (1550 nm), and the ferroelectric tilt angle is 45 .
`Hence, we aimed to use high tilt FLC materials.
`However, the presence of a high tilt angle in a FLC material
`is not compatible with the presence of a Sa phase [12] within
`
`

`

`CROSSLAND et al.: HOLOGRAPHIC OPTICAL SWITCHING: THE “ROSES” DEMONSTRATOR
`
`1847
`
`and a tilt angle of 34 at room temperature. We subjected them to
`fields of up to 10 V/ m at frequencies of approximately 1–2 kHz
`with a small dc offset (a few mV) at temperatures just below
`their I-Sc and N -Sc phase transition temperatures in order
`to obtain monodomains. In addition we were able to rotate the
`layer orientation of the monodomain achieved in CDRR8 using
`asymmetric fields and thus obtain bistable operation [16]. Char-
`acterization of the performance of both compounds confirmed
`that, at fields currently available to us from our silicon active
`backplane, trade-offs have to be made between the speed of re-
`sponse of the device and its efficiency when selecting a suitable
`demonstrator de-
`FLC. Thus we chose CDRR8 for our
`switch [17]. The
`switch
`vice and CS2005 for our
`is based on all glass SLMs using ITO electrodes, which allow
`higher operational fields to be applied to the FLC and permit
`us to demonstrate the relatively high efficiencies which can be
`obtained in binary SLMs using FLC materials with tilt angles
`approaching 45 .
`
`C. Hologram Design
`The hologram refers to the phase modulation pattern and pe-
`riod applied by the SLM. In practice, the effect of the phase
`modulation is to split the beam incident on the SLM into a
`number of reflected beams, each corresponding to a different
`diffraction order of the phase modulation. Only one of these
`diffraction orders is created intentionally: the others are un-
`wanted. The intended diffraction order carries the signal into the
`selected output waveguide, while the unwanted diffraction or-
`ders, unless controlled and positioned appropriately, can lead to
`crosstalk and back reflection. The output angle of each diffrac-
`tion order, measured from the optical axis is given by (2), where
`is the period of the phase modulation, usually known as the
`is an integer that identifies a particular
`hologram period, and
`diffraction order.
`
`(2)
`
`As illustrated previously in Fig. 1, the SLM is pixellated, and
`so the possible hologram periods must be an integer multiple of
`the pixel pitch. On arriving at the launch end of the waveguide
`array, the tails of the beams in the unwanted diffraction orders
`will couple into their nearest waveguides, leading to crosstalk.
`:
`Of particular concern are the zeroth-order reflection (
`angle of reflection) and also the order
`at angle of incidence
`, op-
`symmetric to the one used for switching (same value of
`posite sign), since for binary phase modulation as provided on
`the demonstrator SLM, these orders cannot be eliminated. The
`hologram patterns were calculated using Fourier series analysis
`under the constraint of suppressing crosstalk from unwanted
`diffraction orders to an acceptable level: the targets for total
`20 dB, and total backreflection below 40 dB.
`crosstalk were
`The number of pixels used to display the phase modulation was
`chosen to be sufficient to avoid excess crosstalk and insertion
`loss penalties due to clipping of the tails of the beam incident
`on the SLM. The theoretical diffraction efficiency of the holo-
`grams varies between 4 dB and 4.8 dB: 4 dB of this is inherent to
`binary-phase modulation, while the 0.8 dB variation is the loss
`penalty resulting from modifications to the pattern in order to
`
`Fig. 2. Photograph of the assembled SLM.
`
`the FLC phase sequence and this has implications on the align-
`ment properties of the material. Firstly, the absence of a Sa phase
`means that on cooling into the Sc phase [12] two equivalent
`orientations of the smectic layers appear so that, in order to
`achieve a monodomain, one of the two must be selected by suit-
`able alignment techniques. Secondly, once the monodomain is
`obtained one of the two-switched states is preferred to the other
`and so the material is monostable not bistable, which is a dis-
`advantage from the point of view of addressing schemes. Also,
`since the orientational viscosity of an FLC material typically in-
`creases with increased tilt angle, high tilt angle FLC materials
`tend to exhibit lower operational speeds than standard FLC ma-
`terials for the same applied electric fields. However, in compar-
`ison with standard FLC materials, high tilt angle FLCs exhibit
`a tilt angle, which is fairly temperature independent. Hence, a
`device using a high tilt FLC should have an efficiency which
`is much less temperature sensitive and hence possess a broader
`operational temperature range.
`Operation at
`telecom wavelengths also has implications
`on the device performance. Firstly, in order to achieve a half
`wave plate thickness at 1550 nm the SLM thickness must be
`approximately three times that required for visible wavelengths.
`Secondly, at telecoms wavelengths, the birefringence of the
`FLC may be less than that at visible wavelengths. Again this
`would necessitate the use of a thicker FLC layer to maximize
`the diffraction efficiency. Our SLM was fabricated using a
`fixed voltage active silicon backplane and hence increases in
`the cell thickness results in a decrease in the available applied
`electric field, and correspondingly a decrease in the speed of
`response of the FLC.
`We selected two FLC materials:—CS2005 [14] a commer-
`cially available high tilt angle compound with a phase sequence
`)-Cr and a tilt of 43 at room temper-
`I-(73)-N -(65)-Sc -(
`ature; and CDRR8 [15] an organosiloxane-based experimental
`compound with the phase sequence:—I-(57)-Sc —(room temp)
`
`

`

`1848
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 12, DECEMBER 2000
`
`TABLE I
`REQUIRED WAVEGUIDE ALIGNMENT TOLERANCES AND PERFORMANCE OF OPTOMECHANICS
`
`Fig. 3. ROSES 1  8 switch performance (input power of 1 mW at 1550 nm).
`
`suppress crosstalk from unwanted diffraction orders. Addition-
`ally, in order to maintain electro-chemical stability of the FLC,
`the pixels were DC balanced by scrolling the hologram pattern
`(one pixel at a time) across the SLM.
`
`optomechanics. In use the device has no moving parts and the
`deflection angle produced by the beam deflector only depends
`on the binary pattern presented on the FLC SLM. No dynamic
`alignment is required.
`
`D. Waveguide Array
`A waveguide array was used to provide spatial fan-in in order
`to increase the wavelength range per port, while avoiding the
`need for microlenses. The waveguide index and cross-section
`were optimized to maximize the wavelength range of the
`system for a given crosstalk specification, including the effect
`of lens aberrations (assuming perfect alignment), and subject
`to the constraint of acceptable coupling losses to external
`ribbon fiber. The waveguide array was designed to provide a
`1 dB optical bandwidth of 60 nm. In perfect alignment, the
`theoretical 1 dB bandwidth is 79 nm, which would increase to
`89 nm with a custom lens.
`
`E. Optomechanics
`The full range of alignment tolerances was analyzed, with the
`most severe being that of the transverse position of the wave-
`guide array, with sub-micron accuracy required to maintain the
`required wavelength range. A custom-made optomechanical
`mount was procured with six-axis alignment adjusters to
`hold the waveguide array (glued to a waveguide carrier), the
`Fourier lens and the SLM. Table I lists the required waveguide
`alignment tolerances and the actual tolerances provided by the
`
`III. PERFORMANCE OF SWITCH
`In order to determine the capabilities of the system, the optical
`insertion loss, crosstalk, wavelength response, and temporal re-
`sponse of the switch were experimentally investigated. The first
`step in testing the full switch involved accurately aligning the
`waveguide with respect to the signal beams. Once this was com-
`pleted, the SLM was configured to direct light into each of the
`output channels in turn and power measurements made to deter-
`crosstalk matrix (power diffracted into the
`mine the full
`signal channels and corresponding crosstalk into the other chan-
`nels). Due to the slow response time of the power meter com-
`pared to the hologram update rate (100 frames/s), the optical
`signals were time-averaged. The results of these measurements
`are shown in Fig. 3. For some switch configurations there was a
`0.25 dB,
`noticeable fluctuation in the crosstalk power of up to
`although this was not the case for the signal beam (signal power
`coupled into the required output waveguide), which remained
`extremely stable and had a measurement error of better than
`0.1 dB.
`The theoretical loss for a perfect binary-phase SLM was cal-
`culated to be 5.6 dB (Table II). This figure includes the intrinsic
`
`

`

`CROSSLAND et al.: HOLOGRAPHIC OPTICAL SWITCHING: THE “ROSES” DEMONSTRATOR
`
`1849
`
`4 dB loss due to the binary-phase nature of the hologram, a
`0.66 dB loss due to the nonideal liquid crystal tilt angle of the
`material used (CDRR8), and a dead space loss of 0.92 dB. The
`calculation assumes a perfect cell thickness and no absorption
`loss in the device. In order to determine whether the SLM had
`been perfectly fabricated, a separate test rig was set up to mea-
`sure the diffraction efficiency (defined as the ratio of power
`1 order when a binary-phase hologram is displayed to
`in the
`the power in the zeroth-order when no hologram is displayed)
`and real efficiency of the device (defined as the ratio of power
`1 order to the incident power). The diffraction effi-
`in the
`ciency for channel eight was measured to be 18.1% (7.4 dB
`loss). This difference between this value and the ideal figure of
`5.6 dB is due primarily to an nonoptimized cell thickness, and
`may be corrected for by fine-tuning the cell width. The real ef-
`ficiency, which takes into account absorption within the SLM
`%, giving the device an overall loss
`was found to be
`dB. Measurements of the ITO coated coverplate
`of
`showed that this excess loss was almost entirely due to absorp-
`tion in the ITO layer, a problem that can easily be corrected for
`by using a thinner layer of this material.
`The maximum theoretical insertion loss for an ideally aligned
`switch should be of the order of 9.4 dB. This assumes a per-
`fectly optimized, nonabsorbing SLM, lens aberration losses of
`0.6 dB, and measured propagation losses through the waveguide
`and connectors of 2.4 dB. In addition, there is a variation in effi-
`ciency between the various grating designs of 0.8 dB due to the
`optimization of certain hologram patterns to minimize crosstalk
`between channels. Thus, the minimum insertion loss is 8.6 dB
`dB).
`and the maximum insertion loss is 9.4 dB (average
`Taking into account the real efficiency of the SLM, measured to
`dB, we obtain an average theoretical insertion loss
`be
`dB.
`of
`crosstalk matrix
`After the switch was aligned, the full
`was measured. The insertion loss of the switch was found to
`dB, which compares relatively well with the pre-
`be
`dB. The excess 2.5 dB loss may
`dicted average loss of
`be accounted for by a number of mechanisms including nonop-
`timum alignment, increased aberrations introduced by misalign-
`ment, beam distortion due to a nonuniform curvature of the
`SLM and nonuniform cell thickness, and the presence of dead
`pixels. The variation of insertion loss across all eight channels
`0.5 dB. By comparison, the theoret-
`was measured to be only
`ical variation in hologram diffraction efficiency was calculated
`0.4 dB. The signal-to-peak noise crosstalk varied be-
`to be
`19.1 dB to optical isolations as high as
`tween a worst case of
`40.5 dB. These results are extremely promising, as the target
`20 dB and the theoretical worst crosstalk
`crosstalk figure was
`figure was calculated to be 21.0 dB. The fact that the crosstalk
`failed to meet specifications may again be accounted for by mis-
`alignment. If excess aberrations (caused by a nonuniform SLM
`or a tilt in the system) are present, the beam size will increase
`and light from the symmetric diffraction order will spill over
`into the wrong output channel. Finally, it should be noted that in
`cross-connect, theoretical crosstalk values
`the case of a
`below 40 dB are possible. This will be discussed in the fol-
`lowing section.
`As mentioned in the previous paragraph, the power meter
`response was far slower than the hologram update rate (100
`
`TABLE II
`THEORETICAL LOSS FIGURES FOR SLM AND SWITCH
`
`frames/s). Thus in order to monitor the temporal stability of
`the switch, the output from each signal channel was monitored
`using a photodiode and digital oscilloscope. The temporal vari-
`ation in the power launched down each signal waveguide was
`found to be relatively small, with the measured voltage varying
`3.6% and
`5.3% of the average signal. The biggest
`between
`dip occurred at the start of each new hologram frame. This re-
`sult shows that the hologram update scheme used in the system
`(scrolling the pattern one pixel at a time) does not unduly inter-
`rupt the optical signal being sent through the switch. In addition,
`the wavelength response of the switch was measured as having
`an insertion loss variation of 1.3 dB over a 30 nm range and
`3 dB over a 60 nm range (centered at 1520 nm). The limited
`bandwidth of the switch, and the offset of the minimum inser-
`tion loss from 1550 nm to 1520 nm was due to alignment er-
`rors, an error in the focal length of the lens, and a rotation of the
`waveguide with respect to the signal beams. The actual central
`operating wavelength can be adjusted for by optimizing the po-
`sition of the output waveguides and hologram patterns. Finally,
`the back reflection from the switch was investigated and a worst
`35.1 dB was measured.
`case value of
`Considering the preliminary nature of these measurements,
`and the on-going development of the SLM, and alignment pro-
`cedures, the switch performance results presented in this sec-
`tion are quite encouraging. It is expected that the switch per-
`formance will improve significantly as SLM devices more pre-
`cisely optimized for 1550 nm are made available, better ad-
`dressing schemes are implemented, and more alignment expe-
`rience is gained. In particular, the use of thinner ITO layers and
`optimization of the SLM cell thickness should reduce the av-
`erage insertion loss to well below 10 dB.
`
`OPTICAL SWITCH
`IV.
`One of the main aims of the ROSES project was to develop
`holo-
`the technology required to implement a large-scale
`channel switch can
`graphic optical crossconnect. An
`be built using a single SLM, however, the fan in to the output
`fibers limits scalability and performance. A well-known solu-
`tion to this problem is to build a holographic switch using two
`arrays of sub-holograms, as shown in Fig. 4 [5], [8]. This ar-
`chitecture ensures that the output beams are collinear with the
`axes of output fibers. Although the beam-steering angle has a
`
`

`

`1850
`
`JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 12, DECEMBER 2000
`
`Fig. 4. Two hologram optical switch architecture.
`
`TABLE III
`SUMMARY OF 3  3 SWITCH TARGET PARAMETERS
`
`wavelength dependence, in a double hologram system the “com-
`plementary” nature of the holograms significantly reduces the
`overall wavelength dependence of the switch. This approach
`also has the added advantage of improving the crosstalk char-
`acteristics of the switch as the second hologram array tends to
`reject unwanted diffraction orders from the output channels. Fi-
`nally, it should be noted that the insertion loss for such a switch
`architecture remains constant as the size of the system is scaled
`switching fabrics scaling to many hun-
`up. Thus large
`dreds of channels are possible using this technology.
`switch was designed and con-
`Such a two hologram
`structed as part of the ROSES project. The target performance
`figures for the switch are presented in Table III. Note that Fig. 4
`only illustrates the two-hologram array switch architecture. For
`simplicity the optics required to interconnect the signal beams
`have been omitted from this diagram. A detailed description of
`switch will be presented at a later
`the optical design of the
`date.
`This prototype system used high-efficiency transmissive
`SLMs containing the commercially available FLC material
`CS2005 [14], which possesses a tilt angle of 43 . The trans-
`pixel glass SLM
`missive SLM was fabricated as a
`with 20 m pitch pixels and a 2 m dead space. In practice,
`switches, reflective silicon backplane SLMs
`for large
`would be used instead of transmissive devices. This is because
`of the difficulty involved in addressing individual pixels on
`
`a transmissive device (each linear pixel on the SLM had a
`separate connection going to it). By comparison, only two data
`lines were required to address the reflective SLM used in the
`switch.
`switch demonstrator was designed to operate at the
`The
`telecommunications wavelength of 1550 nm. At this wavelength
`of CS2005 is approximately 0.12. This
`the birefringence
`implies a half wave cell thickness for transmissive modulation of
`6.3 m (or 3.15 m for a reflective device). A pair of glass SLMs
`were made using rubbed nylon alignment with a cell thickness
`measured at 6.4 m. The liquid crystal layer was subjected to
`appropriate pre-treatment (thermal and voltage) to ensure good
`alignment
`system
`The aligned transmissive FLC SLM was tested in a
`at 1550 nm and imaged onto a Vidicon camera to test the diffrac-
`tion performance. The results of this can be seen in Fig. 5, where
`the diffraction peaks are shown at a drive voltage of 50 V
`(8 V/ m at 500 Hz). The diffraction efficiency (power into the
`first order) for the SLM was measured to be 35%, which is close
`to the theoretical maximum of 40.5% for a perfect binary-phase
`hologram. Thus, it can be seen that by using a high tilt FLC and
`fabricating a cell with an optimum thickness, very high SLM
`diffraction efficiencies can be obtained. In practice, the same
`reflective SLM de-
`efficiency could be realized for the
`scribed in Section II-B if the cell thickness were optimized and
`the same liquid crystal material (CS2005) material were used.
`switch down to close to
`This would bring the loss of the
`10 dB.
`switch was built as a bench system, using cata-
`The
`logue lens components, which resulted in a switch that was over
`one meter long. Further studies carried out within the ROSES
`project have shown that, by using a system with both custom
`optics and reflective silicon backplane SLMs, this length could
`be greatly reduced. The two SLMs were run without polar-
`isors and at maximum tilt angle. The measured loss through the
`switch was 19.5 dB (worst value) and the crosstalk was mea-
`dB (worst value), though for most signal paths
`sured at
`the crosstalk was well below 40 dB as required. It should be
`stressed that these results are preliminary, and more work has to
`be done to optimize the switch performance. However, even at
`this stage we can clearly see that the switch worked as expected
`
`

`

`CROSSLAND et al.: HOLOGRAPHIC OPTICAL SWITCHING: THE “ROSES” DEMONSTRATOR
`
`1851
`
`Fig. 5. Diffracted orders generated by a transmissive binary-phase SLM driven at 50 V - .
`
`and the performance figures, even with binary phase SLMs, ap-
`proach that required for telecommunication and sensor systems.
`
`V. FUTURE PROSPECTS
`The results presented in this paper show that reconfigurable
`holographic switching technology has the potential of being
`and
`switching fabrics ca-
`scaled up to large scale
`pable of meeting the requirements of both telecommunications
`and aerospace applications. The technology has several benefits
`when compared to competing approaches such as MEMS-based
`switches and planar thermo-optic waveguides. Firstly, it pro-
`vides a robust and repeatable nonmechanical approach to beam-
`steering. Whereas in a MEMS-based system the beam is steered
`by adjusting the tilt of micro-mirrors, a holographic beam de-
`flector controls the propagation direction of a signal beam by
`changing the hologram pattern and period. This “tilt” control is
`digital and therefore not subject to electrical noise and there is
`no need for continuous adjustment to maintain alignment. In ad-
`dition, as there are no moving parts, the long term reliability of
`a holographic beam-steering switch can be expected to be much
`greater than an equivalent MEMS-based switch. Moreover, in
`terms of redundancy, each hologram is inherently more reliable
`than a MEMS or waveguide-based switch as a few individual
`pixels on each sub-hologram can fail without unduly affecting
`the system performance. The SLM mirror surface is well pro-
`tected from the air, so corrosion or oxidation will not occur, and
`the SLM provides a convenient heatsink for the power absorbed
`by the aluminimum mirrors, so there are no power limit prob-
`lems when routing many WDM channels together.
`The two hologram approach described in Section IV is
`switching fabrics. Although
`scalable to large scale
`insertion loss will be higher compared to a MEMS-based
`switch, the loss is independent of functional size, and funda-
`mental signal-to-noise ratios are limited only by the available
`40 dB
`aperture of the hologram (crosstalk values in excess of
`have been demonstrated in the lab). In addition, although
`high-performance thermo-optic switches have been demon-
`strated, these have tended to be small-scale systems. According
`
`to [1],
`switches have been demonstrated
`and
`48 dB in
`with an insertion loss and crosstalk of 2.3 dB and
`switch and 6.6 dB and
`55 dB in the
`switch,
`the
`respectively. However, the size of the waveguide substrate re-
`mm
`mm
`quired for these systems was relatively big (
`chip containing 256 switching units), and it is uncertain if
`thermo-optic waveguide technology can be scaled up to handle
`the hundreds of channels currently required without having to
`use multiple switching stages.
`The obvious drawback of using a two-hologram plane archi-
`tecture is the increase in insertion loss. However, the perfor-
`mance of a holographic beam-steering switch may be improved
`by using multiple-phase holograms instead of binary-phase de-
`, for an ideal mul-
`vices. The theoretical diffraction efficiency,
`tiple-phase hologram (no dead space) is given by (3)
`
`(3)
`
`is the number of phase levels [16]. In the case of a four-
`where
`level phase device, the theoretical diffraction efficiency would
`be 81% (0.9 dB loss). Thus, if the device is correctly optimized
`(ideal cell width and negligible ITO absorption) it should be
`holographic switch with an insertion
`possible to produce a
`switching fabric
`loss of around 6.0 dB. In the case of an
`utilising two arrays of four-phase level SLMs, the loss should
`be in the region of 10 dB, a dramatic improvement over the
`19.5 dB loss obtained using binary-phase SLMs. Thus, it can be
`seen that the switch performance is no longer solely dominated
`by the SLM efficiency, but by connector losses and aberrations
`introduced by the optics and the SLM. Although this insertion
`loss figure is still greater than that achievable using thermo-optic
`waveguide and MEMS-based switches, it is believed that the
`scalability and inherent reliability of holographic optical beam
`deflector, make this