256<tex>$,times,$</tex>256 Port Optical Cross-Connect Subsystem

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 6, JUNE 2004

256

1499

256 Port Optical Cross-Connect Subsystem

David T. Neilson, Senior Member, IEEE, Member, OSA, R. Frahm, Paul Kolodner, C. A. Bolle, Roland Ryf, Member, IEEE, J. Kim, A. R. Papazian, C. J. Nuzman, A. Gasparyan, N. R. Basavanhally, V. A. Aksyuk, and J. V. Gates

Abstract—This paper describes the subsystem design and performance of a 256 256-port micromechanical beam-steering optical cross-connect with 1.33-dB average loss, which can provide 238 238-port cross-connect with a maximum loss of less than 2 dB. This paper describes the design chosen and analyzes the tolerance ranges required to produce low loss and simulate the expected loss distribution of the fabric. The method of establishing and testing the connections is also described. The simulation is compared with the measured system, and the expected and measured static and dynamic crosstalk are compared. Index Terms—Cross-connect, microelectromechanical (MEMS) devices, optical communication systems, optical design, optical fiber switches, photonic switching systems.

I. INTRODUCTION

O

PTICAL cross-connects have been identified as a key technology for building transparent optical networks [1]–[4]. They enable the network to be easily upgraded to higher data rates or different data formats. In order to build large-port-count switches, an architecture that scales well with port count is required. The most practical technology for building large-port-count ( 32), low-loss optical cross-connect fabrics is to use microelectromechanical systems (MEMS) mirrors with free-space beam steering [5]–[14]. In these fabrics, the component count and dimension scale linearly with the port number. While these systems are typically referred to as three-dimensional (3-D) MEMS cross-connects, the MEMS devices are only one of the technologies required for their construction. Previous publications in the field of optical MEMS for telecommunications have concentrated on the design and performance of the MEMS devices themselves [15]–[20]. While some general design spaces for subsystems have been discussed [21]–[25], little has been written detailing the actual design and requirements of the optical system, which is needed to make the MEMS devices into a switching subsystem. The subsystem designer must be cognizant of the MEMS design space. The MEMS mirrors influence the design in terms of such first-order parameters as mirror sizes and achievable angular range, which put limits on the maximum and minimum beam sizes, respectively. They also influence the design through secondary effects such as mirror curvature, reflectivity, and packaging requirements. The role of the other parts of the Manuscript received June 11, 2003; revised March 16, 2004. D. T. Neilson and R. Ryf are with Bell Laboratories, Lucent Technologies, Holmdel NJ 07733 USA. R. Frahm, P. Kolodner, C. A. Bolle, J. Kim, A. R. Papazian, C. J. Nuzman, A. Gasparyan, N. R. Basavanhally, V. A. Aksyuk, and J. V. Gates are with Bell Laboratories, Lucent Technologies, Murray Hill NJ 07974 USA (e-mail: neilson@lucentcom). Digital Object Identifier 10.1109/JLT.2004.829223

Fig. 1. Optical system layout of cross-connect fabric. The input and output fiber arrays, two microelectromechanical system (MEMS) tilt mirror arrays and the Fourier lens are shown in their relative positions. Extreme beam paths are also shown.

subsystem is to efficiently deliver light to the MEMS array and to couple it back into the fiber, while making the best use of the MEMS components. Several of the technologies required by the free-space subsystem, such as fiber arrays and microlenses, have been previously considered through work on free-space optical interconnects [26]–[28]. However, the use of single-mode fiber (SMF) for both input and output while achieving low insertion loss represented new challenges for these technologies. This paper presents the detailed design of the subsystem for a 256 256-port optical cross-connect (OXC). It indicates what the tolerance space afforded by such a design is and demonstrates that it is sufficient to achieve low-loss switch fabrics, which has been experimentally verified. II. OPTICAL SYSTEM DESIGN First, we describe the optical design space chosen for this cross-connect fabric. In an optical switch fabric, large two-dimensional (2-D) arrays of optical fibers [26]–[28] are connected via free-space optics, with MEMS mirrors acting to define and align the connections, as schematically shown in Fig. 1. The switch fabric consists of arrays of microlenses that are used to relay light from the input fiber array to the input array of beam-steering MEMS micromirrors. The MEMS mirrors on the input array can tilt on two axes to steer the beam to hit a MEMS mirror on the output array, which tilts to steer the beam to the output array of microlenses, which couple light into the output fiber array. The input and output fibers are both single mode, and the fabric is designed to operate in a telecommunications band of 1250–1650 nm.

0733-8724/04$20.00 © 2004 IEEE

1500

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 6, JUNE 2004

Fig. 2. Schematic and actual array shows the layout of the mirrors, which are 0.6-mm diameter on a 1-mm pitch with wiring arteries every fourth row and column. One axis is stretched by 3.5% to account for the 15 skew of the chip in the optical system.

The MEMS chip, shown in Fig. 2, is a surface micromachined design described in [19], [20], [29], and [30]. Because of the assembly structure, to lift the mirror clear of the substrate, only moderate ratios of mirror size to mirror pitch, fill factor, are possible. The surface micromachined design also requires the electrical control lines to be routed in arteries between rows/columns of mirrors, further reducing the overall fill factor of the system. The MEMS mirrors are 600- m diameter on a 1- mm pitch, with a wiring artery after every fourth row and column, as shown in Fig. 2, making a 19 19 array of beams. One axis is stretched by 3.5% to account for the 15 skew of the chip in the optical system. 7 on axis. HowThe mirrors have an angular range of ever, in order to keep the angle/voltage sensitivity low and to provide margin on the MEMS design, we wish to work at angles below 5 on each axis [19]. In order to achieve low clipping loss, we need the beam size on the MEMS mirrors to be 400 m. If we 1.5 times smaller than the mirror size, i.e., use a simple design in which the beam propagates through free space from one array to the other, we find that the optimal distance between the MEMS mirrors is two Rayleigh ranges, with the waist at the center. Given the 400- m beam size constraint at the MEMS mirror, this leads to a 40-mm Rayleigh range, or an 80-mm beam throw. The extreme mirror has to scan this beam across an 18.7-mm array, which leads to a nominal angle requirement for the MEMS mirror of 6.25 . The actual angle required would be increased by the skew of the system, which for a 15 skew would be 6.84 . While this is within the range of the mirrors, it does not provide suitable margin or expected robustness. Instead of the simple system, we chose to use an alternative optical configuration with a Fourier lens placed between the two MEMS chips [14], as shown in Figs. 1 and 3. In this system, the beam waist is placed on the MEMS mirror, and the Fourier lens has a focal length equal to the Rayleigh range and is placed one Rayleigh range away from each MEMS mirror array. A simple and thus biconvex lens is used since the beams are suffer negligible aberration. The lens produces some distortion, which results in small changes, from the paraxial condition, in the required angles for the MEMS mirrors to make the connections. This optical configuration results in a beam waist of the same size being created on the other MEMS mirror. The Rayleigh range chosen was 72.5 mm, which gives a beam diameter on the mirror of 378 m, which is 1.59 times smaller

Fig. 3. Gaussian beam path through the OXC. The 1=e beam diameter is 379 m at the MEMS mirror and 474 m at the microlenses. The MEMS mirrors are 600-m diameter, and the microlenses are 800-m diameter.

Fig. 4. Optical subsystem of the MEMS-based OXC showing MEMS mirror array, fiber array, and Fourier lens mounted in the optical housing.

than the mirror size and therefore will result in low optical clipping loss (0.03 dB). The total loss for this system from clipping and imperfect coupling would be 0.1 dB. In order to allow separation of the input and output beams from the cross-connecting beams, the MEMS mirror array is tilted by 15 with respect to the axis from the center of one array to the center of the other, as shown schematically in Fig. 1 and the actual assembly in Fig. 4. The fiber arrays plus microlens arrays must be placed at a least 50 mm away from the MEMS mirror to ensure the beams are not obstructed, and a value of 56 mm was used to give some margin for aligning the components. Thus, the total path length through the system from

NEILSON et al.: 256

256 PORT OPTICAL CROSS-CONNECT SUBSYSTEM

1501

microlens to microlens is 3.52 Rayleigh ranges, or 257 mm. Microlenses were chosen to have only 80% linear fill factor to minimize any interactions between the surfaces during etching. The beam size at microlens, which is 0.78 Rayleigh ranges away from the waist, is 474 m, which is 1.68 times smaller than the 800- m microlenses (0.015-dB clipping loss). Accounting for the Fourier lens and the actual skew effects, the maximum angular range required for the MEMS mirrors is 4.3 in the skew plane and 4.7 in the nonskew direction. Combined with the skew angle of 15 , this keeps the maximum angle of a beam on a mirror at less than 20 , which ensures that the maximum polarization-dependent loss (PDL) contribution from both gold-coated MEMS mirrors is 0.03 dB. Low PDL is important since it can not only impact the performance of transmission systems, but also contributes to the maximum variation in loss for connections at different angles, arising from the mirror coating. The maximum path length variation in this system is 6% of a Rayleigh range, which would result in negligible (0.004-dB) variation in coupled power. A critical subcomponent of this subsystem is the 2-D array of cleaved, polished, and antireflection (AR)-coated optical fibers, aligned with and attached to an array of microlenses so as to produce a precise lattice of collimated laser beams. These laser beams must be accurately directed onto the array of MEMS mirrors. The optical system of the cross-connect must be constructed with extreme precision [32] in order to minimize optical loss, and the performance of the microlenses and fiber array are critical in achieving this. This requires that the microlenses have a nearly optimal surface contour and that the array of fiber cores in the fiber array be aligned with the focal points of the microlenses in the lens array with high dimensional precision. Initially, we must determine the first-order parameters for the microlenses. The expression for the fiber-to-lens distance in , the Rayleigh terms of the magnification range of the beam at the fiber , and the beam throw is (1) The focal length of the lens is given by the expression (2) For a fiber with a beam diameter 10 m and a beam waist at the MEMs mirror of diameter 378 m, this gives a mag, and the required distance between minification crolenses and the MEMs mirror of 56 mm will give a 2.368 mm and a lens-to-fiber distance of focal length 2.405 mm or 37 m from the focal point of the lens. This focal length is the nominal design value, and the average microlens focal length in an array can vary over a range of about 5 without degrading the best-focus insertion loss ( 0.05 dB) of the system. While there is a wide range for the focal length of the array, there is only a narrow range for the variance of the focal length within the array. The excess loss as a function of defocus is shown in Fig. 5. The tolerance for the uniformity within an array is a defocus of 13 m for 0.2-dB excess loss and 22 m for 0.5-dB excess loss. If this tolerance were available entirely for

Fig. 5. Defocus sensitivity of the microlenses. Experimental values (heavy) and beam propagation calculations (thin) give good agreement. The focal range shown corresponds to 2.5 to 1.7 .

0 % + %

Fig. 6. Silicon microlenses made by photoresist reflow and compensated transfer etching. Pitch between lenses is 1 mm, diameter of lenses is 800 m, and sag is 15 m.

focal length, this would correspond to 0.5 and 1 variations, respectively. However, not all of this variance is available for the focal length tolerance since there are additional sources of defocus error that can arise from nonplanarity of the array of fiber ( 2 m), nonoptimal position, or tilt of the lens array with respect to the fiber array ( 2 m) and an allowance for the thermal expansion of the mounting of the microlens ( 0.5 m over 25 C). This gives a more realistic microlens focal length tolerance of 8.5 m for 0.2-dB excess loss and 17.5 m for 0.5-dB excess loss, or 0.3% and 0.6% respectively. We have evaluated low-loss lenses with uniform focal lengths, which were fabricated by etching both silicon and ) glass. The glass lenses have the advantage high-index ( of a lower index, making it easier to achieve better AR coatings and transparency across a broader range of wavelengths. in addition, the lower index allows a larger absolute tolerance in surface contour, but at the cost of a deeper etch profile. Silicon has the advantage of high index, reducing the etch depth and the availability of better understood chemistry and superior lithography. The required relative control between shape and etch depth is the same for both materials. While both glass and silicon lenses can produce low-loss fabrics, we found the best performance was achieved with the silicon lenses shown

1502

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 6, JUNE 2004

Fig. 7. (a) Focal length error and (b) insertion loss variation for 256 microlens array. The insertion loss shows best focus (thick) and defocused insertion loss (thin) distributions.

in Fig. 6. The focal length error distribution and double-pass best-focus insertion loss distribution for a silicon microlens array made on 8-in wafers using reflow, compensation, and transfer etching [33]–[35] are shown in Fig. 7. This yields arrays with less than five defective lenses per thousand. This array shows a defocus error of better than 6 m, which is consistent with 0.1-dB excess loss when the other defocus effects ( 4.5 m) are considered. The variation in double-pass insertion loss, simulating propagation through two identical arrays, is shown as the thick line in Fig. 7(b) at best focus for the array from 0.1–0.4 dB, including the loss due to AR coatings. The effect of the defocus changes the distribution, as shown by the thin line in Fig. 7(b), though the variation remains the same, i.e., in the range from 0.1–0.4 dB. The fiber array must also be laterally aligned with precision to the lens array in order to ensure the beams can be accurately registered on the MEMS mirror array. The tight alignment tolerance arises from the magnification of the lateral errors by the microlens. The lateral magnification of the microlens of focal length at a distance , which differs from the beam , which magnification given previously, is given by for the optical system described is 22.7. The loss sensitivity to decentering of the beam is shown in Fig. 8, as a function of beam decenter on the 600- m diameter MEMS mirror. The corresponding fiber-to-microlens decenter is also indicated. From this, we can see that a 95- m beam decenter (4.1- m lens-to-fiber error) achieves an excess diffraction loss of less than 0.2 dB. It is worth noting that the size of the fiber-to-microlens errors that are acceptable is approximately four times larger than for coupling fibers directly to each other; thus, the fiber array precision required for this fabric is less exacting than for a multifiber connector or coupling to a waveguide array. The fiber arrays are made by inserting fibers into holes etched in silicon faceplates [36] on a 1-mm pitch. They are polished and AR coated with a return loss greater than 30 dB. These have a 97 yield of working fibers or 248 fibers per array. They typically have a fiber-to-uniform-grid error of 1 m maximum. This corresponds to a 23- m maximum error when projected onto the MEMS mirror array, and this would give a negligible contribution to the insertion loss. Alignment of the lensed fiber array to the MEMS mirrors is achieved by monitoring the power reflected off the mirrors with a large-area detector. The sensitivity of the power measurement determines

Fig. 8. Excess double-pass insertion loss due to the decentering of the laser beam on its target MEMS mirror is plotted as a function of the distance from the mirror center and in terms of the corresponding error in fiber-to-microlens misalignment, given the 22.7 magnification. The decentering of the beam must be less than about 95 m (4.1 m for fiber) to achieve an excess diffraction loss of less than 0.2 dB and less than about 120 m (5.1 m for fiber) to have excess loss under 0.5 dB.

2

the alignment precision and will typically lead to a maximum misalignment of 80 m. Combined with the variation of the fiber array, the resulting total misalignment is a maximum of 100 m, which would contribute a worst-case 0.25-dB insertion-loss penalty. The thermal stability of the alignment of the system is also a critical performance metric. With a stainless steel housing (SS416), the expected misalignment of the MEMS and fiber array components is less than 0.7 m C corresponding to 25- m error over 10 to 60 C, which would be expected to contribute less than 0.5-dB excess loss. We have verified that 0.5 dB with acconnection loss can be maintained within tive mirror alignment over the temperature from 10 to 60 C. 4 C) near room temperFor small temperature excursions ( ature, the connection loss is maintained within 0.05 dB without active alignment. Mirror curvature has been highlighted as a significant issue for achieving low insertion loss [12], [22]. We will show that, while it is important, the ranges of acceptable curvature are sufficiently large as to allow thin silicon mirrors with metallic coatings [29] to achieve low loss. If the MEMS mirrors are not perfectly flat, then they will modify the reflected beam propagation, which will result in loss due to additional clipping or

NEILSON et al.: 256

256 PORT OPTICAL CROSS-CONNECT SUBSYSTEM

mode mismatch in the follow-on optics. In general, there will be spatial, thermal, or temporal nonuniformity of the MEMS mirrors in the system and the resulting effect on loss, which must be considered. The effect of a curved MEMS mirror is to modify the wavefront curvature of the beam, causing it to change its divergence through the system. This has two effects on the throughput of the system: 1) it changes the size and focal position of the output beam at the fiber, causing a change in coupling efficiency, and 2) it changes the beam size at the apertures in the system, which results in increased loss due to clipping. Since the MEMS mirrors are arranged in Fourier transform configuration, the beam size at the other mirror is a function only of magnitude of the curvature at the other mirror and not its sign. Thus, the clipping loss is independent of the sign of the curvature of the mirror. The coupling loss, which comes primarily from defocus of the beam caused by curvature of the mirror, does depend upon the relative signs of the mirrors since the curvature of one mirror can compensate the curvature induced by the other. Since the total loss is a combination of the coupling and clipping losses, we would expect it to depend on the relative curvatures of the two mirrors as well as their absolute curvature. Using Gaussian beam-propagation calculations, we have determined the expected loss as a function of the curvature of the two mirrors for our system design. The results of this analysis are shown in Fig. 9. The results show that the relative sign of the curvature of the two mirrors is important for determining the tolerance, with matching curvatures providing approximately twice the tolerance of mirrors with differing curvature signs. From Fig. 9, we can determine that the 0.5-dB tolerance on matching curvatures is 5 m or a radius of curvature of 200 mm, while for random curvature, it is 2 m or a 500-mm radius of curvature. Thin (3- m)-surface micromachined mirrors [29] can achieve stable curvatures flatter than 5 m and are suitable for low-loss cross-connects. The wavelength dependence of the fabric occurs in two ways if we neglect the contributions of AR coatings which can be made arbitrarily flat: 1) from the change in the diffraction of the beams through the system (axial chromatic aberration) and 2) by changes in the angles of the connection path through the system (lateral chromatic aberration). In order to calculate the effects on the diffraction of the beams with wavelength, it is necessary to account for the change in the fiber numerical aperture (NA), the change in the focal length of the microlenses, and the change in the way the beam diffracts in free space. By considering these effects, we can calculate the change in coupling and clipping for wavelengths from 1310 to 1610 nm to be less than 0.1 dB. The change in the refractive index of the Fourier lens introduces a wavelength-dependent change in angle for some of the high-angle connections. However, the connection can always be peaked for any given wavelength, with no additional loss. The Fourier lens is chosen to be Schott LaF2, a high-index glass ) with low dispersion in the 1.31–1.6- m range to ( minimize this effect. If the switch is to be used, in a transparent wavelength network application, with multiple wavelengths 0.1-dB distributed over the 1520–1620-nm range, then wavelength-dependent loss is observed. If the range is the 0.6-dB excess wavelength-depenfull 1320–1620 nm, then

1503

Fig. 9. MEMS mirror curvature sensitivity, showing that the sensitivity depends on the relative curvature as well as the absolute curvature. The contours are 0.1 dB, and the 0= 5 m curvature range is indicated by the box.

0 +

dent loss is observed. While this wavelength dependence is generally acceptable, it is possible to replace the Fourier lens with a curved mirror [31] and eliminate the lateral chromatic aberration entirely.

III. OPTICAL CONNECTION DATABASE AND VERIFICATION An OXC consists of more than just the physical hardware of the MEMS and optics, since it is necessary to know how to conconnections. The advantage trol the mirrors to make all the of the beam-steering approach is that the number of MEMS mirrors scales with the port count , but this means that each mirror must be able to address positions. It is necessary to determine the database of voltages that must be applied to make the connections with minimum loss. It is also important to verify for a given desired connection that an optical path exists and that the measured value of the insertion loss is known. These two functions can be determined simultaneously using a single database and verification test set, shown in Fig. 10. This is composed of a set of voltage driver cards with 2048 individually addressable channels to activate the 512 mirrors in the system (there are two MEMS mirror arrays with 256 mirrors, and each mirror has four electrodes). An optical test set sends light into each input fiber and collects the light from each output port. The input system is composed of 16 1 16 mechanical SMF switches, fed by light from a 1.3- or 1.55- m distributed feedback laser via optical splitters, which are connected with LC fiber connectors to the input ports of the switch fabric under test. The output system consists of 16 2 beam combiners, which combine the light coming from 16 of the ports to both a microsecond-response detector and an optical power meter. The test set is connected to the fabric output ports with LC connectors. A computer controller determines the optical paths to be illuminated and the voltages to be set and reads the optical powers at the output ports. The computer is also responsible for building the database, modeling the optical and electrical performance, and calculating expected trial voltages for new connections. The configuration supports multithreading of up to 16 connections being optimized simultaneously, and the entire database of 524 288 voltages and 65 536 optical measure-

1504

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 6, JUNE 2004

TABLE I INDIVIDUAL TOLERANCE RANGES OF PARAMETERS FOR THE SIMULATIONS

Fig. 10. Connection and verification test set with a switch fabric under test (centered on the table) and fiber loop-back panel (to the right). The test set sources, detectors, and switches used as optical inputs and outputs are located above and connected to the fabric under test with SMFs using 512 LC connectorized fiber jumpers.

ments can be determined and verified in less than 4 h with verification of the repeatability of loss on each connection. In order to establish the loss of the switch fabric, it is necessary to calibrate the loss of the test set. This is achieved by using a loop-back panel, shown to the right of the fabric in Fig. 10, which is the same as the cross-connect fiber interface panel but with fibers directly connecting the input to the output ports on the back side of the panel. The use of the loop-back panel allows one to minimize the test set variance and thus enables measurements to achieve accurate average losses. The use of LC connectors in the fabric, test set, and the loop-back panel introduces relative port-dependent uncertainty in the loss measurement. This causes the test set to produce an apparently broader distribution for the switch fabric than the actual distribution of the fabric itself. In the processes of calibrating the test set and measuring a fabric under test, eight LC connectors are used. This results in a 0.3-dB (one sigma) variability, of which only 0.15 dB may be attributed as being intrinsic to the two LC connectors associated with the fabric itself. The remainder is due to the variability of the LC connectors in the test set and calibration procedure. This variability has been verified by measuring multiple switch fabrics and loop-back panels on several test sets. IV. SIMULATION OF COMBINED TOLERANCES In the previous section, we considered the individual tolerances and their acceptable ranges for our design. However, in the whole system, we know that the errors in components couple and individual bounds do not always give realistic ranges. An example of the coupling is that the errors in focus of the microlenses produce beam size and wavefront curvature changes at the MEMS mirror which interact with the curvature there to determine the size and curvature of the beam at the second MEMS mirror. To address this issue, we have run Monte Carlo

Fig. 11. Comparison of the four tolerance ranges given in Table I. Set 1 is the base set. Set 2 has increased mirror curvature range of 0=6.6 m . Set 3 has an increased microlens defocus range of = 26 m. Set 4 has an increased mirror centering error range of = 150 m.

+ 0

+ 0

0

simulations of the optical system to examine the coupling of mirror curvature, focal-length errors, and decentered beams on the apertures. In order to simplify the computation, we neglect the clipping of the microlenses and use Gaussian beam theory to propagate the input and output beams to their respective MEMS mirrors, which can be decentered and curved. The propagation of the clipped input beams to the output can be calculated by using a fast Fourier transform (FFT), and the mode overlap integral can be calculated at the output MEMS mirror. This keeps the computation to a minimum, making such simulations practical. To simulate a fabric, we randomly generate 256-input and 256-output port errors and then calculate the 65 536 connections between them. This enables us to relate the computed results to the measured values of an actual fabric while providing an average of six sampling points across each parameter described sunsequently. We have simulated the system with four configurations of the three fabrication errors given in Table I. The first configuration is bound by the tolerances that we have concluded for our system design and experimental measurements of components. The other three configurations are variations of one tolerance parameter each to explore the sensitivity space of these parameters. In all cases, we use a uniform probability distribution within the tolerance bounds. The loss due to the other com0.75 dB and is added to ponents, such as optical coatings, is the losses in the simulations. The resulting distributions from simulated fabrics using the parameter sets in Table I are shown in Fig. 11. The distribution for the first set, which is our baseline for the system, indicates a maximum loss of 2 dB, a spread of 1 dB, and an average of 1.3 dB.

NEILSON et al.: 256

256 PORT OPTICAL CROSS-CONNECT SUBSYSTEM

Fig. 12. Effect of variation of the MEMS mirror curvature on the distribution (broken), nominal 0= 5 m of losses for both increased 0= 3.3 m (solid), and reduced 0= 6.6 m (dash) tolerance ranges.

0 +

0 +

0 +

From the other distributions in the figure, we can see that increasing any of the variances results in a significant change in the connection loss distribution. This indicates that the tolerances in our current system design are closely balanced in scale and approaching critical values. We conclude that there is little scope to increase the ranges without significant performance degradation. We have also looked at the effect of loosening and tightening the MEMS mirror curvature tolerance. This is illustrated in Fig. 12 for both increased and reduced tolerance ranges. From Fig. 12, we conclude that, while it would be beneficial to have flatter mirrors, the improvement in worst-case insertion loss will only be a few tenths of a decibel. In addition to the tolerances given here previously, there will also be variances from the optical coatings, 18 surfaces, and the aberration of the microlenses in the system. We estimate this variance to have a sigma of 0.14 dB. This variance was not included in the previous simulations, although the 0.75-dB average loss was included. The optical coating and microlens variance can be convolved with the simulated distributions to determine a “real” distribution since these effects would not be expected to couple to the other variances. V. EXPERIMENTAL RESULTS We have constructed several switch fabrics using the system designs described previously. Typically, fabrics can achieve a measured loss distribution with a 1.5-dB average and 2.5-dB maximum, with a port count of greater than 230 230. The loss distribution of the best performing fabric [14] is shown in Fig. 13. It has a mean insertion loss of 1.35 dB and has 247 input and 251 output ports. The PDL is 0.1 dB, and the polarization-mode dispersion and chromatic dispersion are less than 0.1 ps and 0.2 ps/nm, respectively. We can compare the simulation results to the measured data for a fabric. This requires that we account for the variation of the fiber connectors in the fabric and test set. The effect of the connectors in the simulation is achieved by adding a random loss with a mean value of zero and a sigma width of 0.3 dB to each port. We also add the random variation with one sigma of 0.14 dB to account for variance in coatings and aberrations of

1505

2 6

Fig. 13. Optical loss distribution for a 247 251-port fabric. Comparison between measured (heavy) and simulated loss with all test set connectors 0.3 dB (thin), with only fabric connectors 0.15 dB (dashed) and without connectors (broken). Simulation assumes fiber defocus error of 13 m, curvature 0–5 m , decentering 100 m, and connectors. Other fixed contributions to loss: 0.75 dB.

6

6

6

the microlenses. The new loss distribution can then be determined and compared with the measured distribution, as shown in Fig. 13. We find good agreement between our simulations and measured distribution, indicating that the tolerance ranges we have determined are consistent with the as-built fabric. We can also use the simulation to give us an understanding of what the fabric loss looks like, without connectors (no connector broadening) or if one could measure a fabric with only the connectors that are actually part of the fabric ( 0.15 dB). From these distributions, it appears that the entire fabric is consistent with a maximum loss of less than 2 dB. This result is consistent with a previous paper [14] where we had conservatively described 238 238 ports as being below this threshold. The simulations allow us to better understand the origin and nature of the losses in the system, since we can directly consider correlation of loss with an error in one of the parameters. In general, understanding the significance of the 65 536 connection losses in the fabric is challenging, and we decompose the loss of the connection into three parts. The loss of a connection in the fabrics can be characterized as the sum (in decibels) of an input port loss, plus an output port loss, as suggested in [37], plus an additional connection-dependent loss term (3) The input port loss can be derived from the connection losses for that port by taking the difference between the average loss of the connections to that port and the average loss of all the connections in the fabric and adding it to half the average loss of all the connections (4) The connection-dependent loss is given by the difference between the actual connection loss and the prediction obtained from adding the two computed port losses, as follows: (5) This model produces 256 input losses, 256 output losses, and 65 536 connection-dependent losses. Typically, the dominant fraction of the loss variance of a connection is determined by the

1506

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 6, JUNE 2004

port losses, with the connection-dependent terms being small. Since the port loss can be determined from a subset of the connections, it is possible to determine the expected loss of a connection before measuring all the connections in the fabric. This concept is used in system performance modeling during database determination and verification since we can have estimates of the loss of connections after only a small number of connections have been made. The port losses also allow screening for high-loss ports to improve the loss distribution of the fabric for a fixed number of ports. The usefulness of port losses is predicated on the connection-dependent loss being small. We can assess the validity of this assumption for both the simulation and real data sets. From the simulations, we find that distribution of the port-dependent loss has a standard deviation of 0.05 dB and a minimum-to-maximum variance of 0.5 dB; i.e., the port loss model alone can be wrong by as much as 0.25 dB. From the experimental results, we find that the standard deviation is 0.06 dB, and the minimum-to-maximum variation is again around 0.5 dB. From this, w e can conclude that the connection-dependent loss is real, and we can evaluate the significance of these variations by comparing them to the variation of the loss in the fabrics. The experimental fabric and simulation with connectors have a loss variation of 1.3 dB, indicating that neglecting the connection dependence is accurate to 4 typically and 19 worst case. This range of uncertainty still provides useful information in predicting losses and screening for high-loss ports. That this model works for the real fabric, where fiber connectors play a significant role in the variance of losses, is not entirely surprising, since fiber connectors will contribute only constant port-dependent loss. By considering the simulation results, we can evaluate the significance and origin of connection-dependent loss in the free-space portion of the cross-connect. Here, we find the connection-dependent loss represents 10 typically and 50 worst case of the loss variance. The parameters, which result in higher connection dependent losses, are extreme values of mirror curvature or microlens focal-length variation. The concept that curvature can produce the connection-dependent variance, as well as port loss, can be understood from the coupled nature of the curvature, as shown in Fig. 9, in which the loss is a function of both input and output curvatures and so would not be expected to decompose to static port-dependent values. From this, we would conclude that the port-dependent loss alone is not a representative loss distribution of the free-space portion of the fabric. From the consistency that we find between the simulation and experimental measurements of the connection-dependent loss, we conclude that the simulation provides verification of the origins of the loss variance in the fabrics. The simulations predict a significant contribution to the connection loss variance from compensating effects such as mirror curvature and microlens focal length. VI. CROSSTALK Crosstalk between cross-connect channels must be low in order to ensure that the data is not corrupted or degraded when passing through the switch. There are two crosstalk conditions we must consider, which we will refer to as static crosstalk and

Fig. 14.

Crosstalk distribution for static case for a typical OSM.

dynamic crosstalk, due to the situations under which they typically occur. Static crosstalk is the crosstalk between channels that are making valid connections. In this case, the crosstalk is inherently low because both the tilt of the input and output mirrors involved in the valid connections are wrong to couple light into adjacent connections. Static crosstalk is caused by the small amount of light that is diffracted by the input MEMS mirror aperture ( 22 dB) spreading out on the output MEMS mirrors adjacent to the target mirror. A fraction of the light is then rediffracted into those beams by the other MEMS apertures. Thus, the worst case of static crosstalk is for connections in which the paths are parallel, i.e., when adjacent input and output mirrors are used. From beam-propagation calculations, this worst-case static crosstalk is 67 dB. The measured static 62-K connections in a typical crosstalk distribution for switch fabric is shown in Fig. 14. The experimental result shows a peak in the distribution at 68 dB, corresponding well with the beam-propagation value. Dynamic crosstalk is the case that can occur during active switching, when an input mirror is being scanned to another connection [38] while light is falling on that input mirror. This is a much more difficult situation to evaluate, since light is potentially diffracting off various apertures and structures, depending on the trajectory of the mirror during switching. Testing all possible dynamic crosstalk conditions in a fabric is not practical since there are 256 factorial starting conditions for the switch configuration. Thus, we must consider the most likely mechanism and evaluate it. During a switching event ( 5 ms) [39], it is possible for the scanned input beam to cross another output mirror that is involved in a legitimate connection. In this case, only one mirror, the output mirror, is misaligned for the crosstalk light to couple to the output, and so the crosstalk can be greater than for the static case where both mirrors are misaligned. The worst case of dynamic crosstalk occurs when the beams are overlapping and nearly parallel in propagation. This occurs if an offending input mirror adjacent to the input mirror of an existing connection is scanned across the output mirror for that existing connection. Although the magnitude of dynamic crosstalk can be greater than static crosstalk, in normal operation, dynamic crosstalk will be a transient effect with switching times 5 ms and therefore less significant. (In a real operating system, if a

NEILSON et al.: 256

256 PORT OPTICAL CROSS-CONNECT SUBSYSTEM

1507

Fig. 16. Dynamic crosstalk distribution for a typical OSM. This distribution represents individual measurements from the four surrounding mirrors from the 62 K possible connections ( 248 K measurements).



2

Fig. 15. Dynamic crosstalk as interfering mirror is scanned over 1 1-mm patch. The bold circle indicates the position of the target mirror (600 m). The contours are 2 dB with the peak at 48 dB. The plot is truncated at 70 dB.

0

0

fault occurred elsewhere that caused an input mirror to statically point incorrectly at the wrong output mirror, then that input mirror would not be used in an active connection and therefore would not be carrying light and, hence, would not contribute to crosstalk.) However, we could consider a static measurement of this worst-case dynamic crosstalk as an upper bound on dynamic and static crosstalk. From propagation calculations, the worst-case dynamic crosstalk in this condition is 44 dB and associated with a crosstalk beam that is 300 m from the center of the mirror; i.e., is being partially clipped. To verify this condition, we have measured the coupling when an input mirror is scanned across an output mirror forming a connection to an input mirror adjacent to the scanned mirror. The amount of light coupled is shown in Fig. 15. The asymmetry occurs because the mirror being scanned is logically to the left of the mirror making the connection. The maximum crosstalk occurs when the beam is on the mirror but is displaced from the center toward the edge, indicating that it is an effect related to clipping on the edge of the mirror. The scan also shows there are no other significant sources of light from the surrounding geometrical structures other than from the MEMS mirrors. Sampling this crosstalk condition by independently measuring the individual four surrounding mirrors for each of the possible 61 997 connections ( measurements), we find that the dynamic crosstalk is typically less than 50 dB, and the worst observed case is 40 dB, as shown in Fig. 16. We attribute the bimodal distribution to small reflections from the AR coating used in the package window. Reflected light in the horizontal plane of the central connection can walk through multiple reflections and contribute to the crosstalk. The mirrors physically located above and below the center connection are not in the horizontal plane of the main connection and contribute less crosstalk than the mirrors to the left and right. This observed crosstalk is sufficiently low as



to ensure low-penalty transmission through the system. Note also that dynamic crosstalk during switching can be avoided by controlling the mirror-angle trajectory during the switch step. In this case, the crosstalk will be better than 54 dB during normal operation. In both the static and dynamic cases, the amount of crosstalk is a function of the amount of light clipped at the MEMS mirrors and quality of the AR coating used on the package windows. By designing a switch with low clipping loss and good AR coatings, it is possible to reduce the worst-case crosstalk in the switch. Another issue with arrays of MEMS mirrors can be the mechanical, hydraulic, or electrical interaction with neighbors when one mirror is actuated. This does not produce crosstalk in the optical signals but may cause power fluctuations in the output signals. We see no evidence of this effect ( 0.1 dB) in our fabrics. VII. CONCLUSION This paper has described the design of a 256 256 OXC subsystem. It has outlined the design with a Fourier transform lens and indicated some of the advantages of this approach. The requirements of each of the major components were identified, and the tolerance space for the significant parameters of the design were quantified. This paper related these to the experimentally achieved fabrication tolerances for the components, has shown how the tolerances couple, and has related these to experimentally measured fabrics. The methods of measuring and interpreting the performance of optical switch fabrics were discussed. Finally, the source and magnitude of the optical crosstalk in these switch fabrics were determined. REFERENCES [1] Y. Tze-Wei, K. L. E. Law, and A. Goldenberg, “MEMS optical switches,” IEEE Commun. Mag., vol. 39, pp. 158–163, Nov. 2001. [2] D. J. Bishop, C. R. Giles, and G. P. Austin, “The Lucent LambdaRouter: MEMS technology of the future here today,” IEEE Commun. Mag., vol. 40, pp. 75–79, Mar. 2002. [3] A. Neukermans and R. Ramaswami, “MEMS technology for optical networking applications,” IEEE Commun. Mag., vol. 39, pp. 62–69, Jan. 2001. [4] E. Goldstein, L. Lin, and J. A. Walker, “Lightwave micromachines for optical networks,” Optics Photonics News, Mar. 2001.

1508

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 6, JUNE 2004

[5] D. T. Neilson, V. A. Aksyuk, S. Arney, N. R. Basavanhally, K. S. Bhalla, D. J. Bishop, B. A. Boie, C. A. Bolle, J. V. Gates, A. M. Gottlieb, J. P. Hickey, N. A. Jackman, P. R. Kolodner, S. K. Korotky, B. Mikkelsen, F. Pardo, G. Raybon, R. Ruel, R. E. Scotti, T. W. VanBlarcum, L. Zhang, and C. R. Giles, “Fully provisioned 112 112 micro-mechanical optical cross connect with 35.8 Tb/s demonstrated capacity,” in Proc. Optical Fiber Communication Conf. (OFC 2000), vol. 4, 2000, pp. 202–204. [6] P. M. Hagelin, U. Krishnamoorthy, J. P. Heritage, and O. Solgaard, “Scalable optical cross-connect switch using micromachined mirrors,” IEEE Photon. Technol. Lett., vol. 12, pp. 882–884, July 2000. [7] A. Keating, “Optical MEMS in switching systems,” in IEEE/LEOS Annual Meeting 2001 Dig., San Diego, CA, Nov. 2001, pp. 8–9. [8] R. Ryf, J. Kim, J. P. Hickey, A. Gnauck, D. Carr, F. Pardo, C. Bolle, R. Frahm, N. Basavanhally, C. Yoh, D. Ramsey, R. Boie, R. George, J. Kraus, C. Lichtenwalner, R. Papazian, J. Gates, H. R. Shea, A. Gasparyan, V. Muratov, J. E. Griffith, J. A. Prybyla, S. Goyal, C. D. White, M. T. Lin, R. Ruel, C. Nijander, S. Arney, D. T. Neilson, D. J. Bishop, P. Kolodner, S. Pau, C. Nuzman, A. Weis, B. Kumar, D. Lieuwen, V. Aksyuk, D. S. Greywall, T. C. Lee, H. T. Soh, W. M. Mansfield, S. Jin, W. Y. Lai, H. A. Huggins, D. L. Barr, R. A. Cirelli, G. R. Bogart, K. Teffeau, R. Vella, H. Mavoori, A. Ramirez, N. A. Ciampa, F. P. Klemens, M. D. Morris, T. Boone, J. Q. Liu, J. M. Rosamilia, and C. R. Giles, “1296-port MEMS transparent optical crossconnect with 2.07 Petabit/s switch capacity,” in Proc. Optical Fiber Communication Conf. and Exhibit, 2001 (OFC 2001), vol. 4, 2001, Paper PD28-P1–3. [9] J. Kim, J. V. Gates, C. J. Nuzman, B. Kumar, D. F. Lieuwen, J. S. Kraus, A. Weiss, C. P. Lichtenwalner, A. R. Papazian, R. E. Frahm, N. R. Basavanhally, D. A. Ramsey, V. A. Aksyuk, F. Pardo, M. E. Simon, V. Lifton, H. B. Chan, M. Haueis, A. Gasparyan, H. R. Shea, S. Arney, C. A. Bolle, P. R. Kolodner, R. Ryf, and D. T. Neilson, “1100 1100-port MEMS-based optical crossconnect with 4 dB maximum loss,” IEEE Photon. Technol. Lett., vol. 15, pp. 1537–1539, Nov. 2003. [10] O. Jerphagnon, R. Anderson, A. Chojitacki, R. Heikey, W. Fant, V. Kaman, A. Keating, B. Liu, C. Pusaria, J. R. Sechrist, D. Xu, S. Vuan, and Z. Xuezhe, “Performance and applications of a large port-count and low-loss photonic cross-connect system for optical networks,” in 15th Annu. Meeting IEEE Lasers and Electro-Optics Society, 2002 (LEOS 2002), vol. 1, 2002, pp. 299–300. [11] J. Kim, A. R. Papazian, R. E. Frahm, and J. V. Gates, “Performance of large scale MEMS-based optical crossconnect switches,” in 15th Annu. Meeting IEEE Lasers and Electro-Optics Society, 2002 (LEOS 2002), vol. 2, 2002, pp. 411–412. [12] A. Neukermans, “MEMS devices for all optical networks,” in Proc. SPIE, vol. 4561, MOEMS and Miniaturized Systems II, 2001, pp. 1–10. [13] Y. Uenishi, J. Yamaguchi, T. Yamamoto, N. Takeuchi, A. Shimizu, E. Higurashi, and R. Sawada, “Free-space optical cross connect switch based on a 3D MEMS mirror array,” in 15th Annu. Meeting IEEE.Lasers and Electro-Optics Society, 2002 (LEOS 2002), vol. 1, 2002, pp. 59–60. [14] V. A. Aksyuk, S. Arney, N. R. Basavanhally, D. J. Bishop, C. A. Bolle, C. C. Chang, R. Frahm, A. Gasparyan, J. V. Gates, R. George, C. R. Giles, J. Kim, P. R. Kolodner, T. M. Lee, D. T. Neilson, C. Nijander, C. J. Nuzman, M. Paczkowski, A. R. Papazian, F. Pardo, R. Ryf, H. Shea, and M. E. Simon, “238 238 micromechanical optical crossconnect,” IEEE Photon. Technol. Lett., vol. 15, pp. 587–589, Apr. 2003. [15] M. C. Wu, D. Hah, P. R. Patterson, and H. Toshiyoshi, “Microelectromechanical scanning devices for optical networking applications,” in 2002 IEEE Int. Solid-State Circuits Conf. Tech. Dig. (ISSCC 2002) , vol. 1, 2002, pp. 358–359. [16] Y. Mizuno, O. Tsuboi, N. Kouma, H. Soneda, H. Okuda, Y. Nakamura, S. Ueda, I. Sawaki, and F. Yamagishi, “A 2-axis comb-driven micromirror array for 3D MEMS switches,” in 2002 IEEE/LEOS Int. Conf. Optical MEMS Dig., Aug. 2002, pp. 17–18. [17] R. Sawada, J. Yamaguchi, E. Higurashi, A. Shimizu, T. Yamamoto, N. Takeuchi, and Y. Uenishi, “Single Si crystal 1024 ch MEMS mirror based on terraced electrodes and a high-aspect ratio torsion spring for 3-D cross-connect switch,” in 2002 IEEE/LEOS Int. Conf. Optical MEMS Dig. , Aug. 2002, pp. 11–12. [18] M.-H. Kiang, O. Solgaard, and K. Y. Lau, “Electrostatic combdrive-actuated micromirrors for laser-beam scanning and positioning,” J. Micromech. Syst., vol. 7, pp. 27–37, 1998. [19] V. A. Aksyuk, F. Pardo, D. Carr, D. Greywall, H. B. Chan, M. E. Simon, A. Gasparyan, H. Shea, V. Lifton, C. Bolle, S. Arney, R. Frahm, M. Paczkowski, M. Haueis, R. Ryf, D. T. Neilson, J. Kim, R. Giles, J. Gates, and D. Bishop, “Beam-steering micromirrors for large optical crossconnects,” J. Lightwave Technol., vol. 21, pp. 634–642, Mar. 2003.

2

2

2

[20] V. A. Aksyuk, D. J. Bishop, C. A. Bolle, R. C. Giles, and F. Pardo, “Micro-electro-mechanical optical device,” U.S. Patent 6 300 619, Oct. 9, 2001. [21] R. R. A. Syms, “Scaling laws for MEMS mirror-rotation optical cross connect switches,” J. Lightwave Technol., vol. 20, pp. 1084–1094, July 2002. [22] K. Bergman, N. Bonadeo, I. Brener, and K. Chiang, “Ultra-high capacity MEMS based optical cross-connects,” in Proc. SPIE, vol. 4408, Design, Test, Integration, and Packaging of MEMS/MOEMS 2001, Cannes, Frances, 2001, pp. 2–5. [23] K. Bergman, “Overview of high capacity optical cross-connects,” in 14th Annu. Meeting IEEE Lasers and Electro-Optics Society, 2001 (LEOS 2001), vol. 1, Nov. 2001, pp. 224–225. [24] P. B. Chu, S. S. Lee, and S. Park, “MEMS: The path to large optical crossconnects,” IEEE Commun Mag., vol. 40, pp. 80–87, Mar. 2002. [25] X. Zheng, V. Kaman, Y. Shifu, Y. Xu, O. Jerphagnon, A. Keating, R. C. Anderson, H. N. Poulsen, B. Liu, J. R. Sechrist, C. Pusarla, R. Helkey, D. J. Blumenthal, and J. E. Bowers, “Three-dimensional MEMS photonic cross-connect switch design and performance,” IEEE J. Select. Topics Quantum Electron., vol. 9, pp. 571–578, Mar.–Apr. 2003. [26] N. R. Basavanhally, “Opto-mechanical alignment and assembly of 2D-array components,” in 1993 IEEE Princeton Section Sarnoff Symp. Dig., Mar. 26, 1993, pp. 0_23–0_27. [27] G. M. Proudley, C. S. Stace, and H. White, “Fabrication of 2-dimensional fiber optic arrays for an optical crossbar switch,” Opt. Eng., vol. 33, no. 2, pp. 627–635, Feb. 1994. [28] J. M. Sasian, R. A. Novotny, M. G. Beckman, S. L. Walker, M. J. Wojcik, and S. J. Hinterlong, “Fabrication of fiber bundle arrays for free-space photonic switching systems,” Opt. Eng., vol. 33, no. 9, pp. 2979–2985, Sept. 1994. [29] F. Pardo, V. A. Aksyuk, S. Arney, H. Bair, N. R. Basavanhally, D. J. Bishop, G. R. Bogart, C. A. Bolle, J. E. Bower, D. Carr, H. B. Chan, R. Cirelli, E. Ferry, R. Frahm, A. Gasparyan, J. V. Gates, C. R. Giles, L. Gomez, S. Goyal, D. S. Greywall, M. Haueis, R. C. Keller, J. Kim, F. P. Klemens, P. Kolodner, A. Kornblit, T. Kroupenkine, W. Lai, V. Lifton, J. Q. Liu, Y. Low, W. Mansfield, D. Marom, J. F. Miner, D. T. Neilson, M. Paczkowski, C.-S. Pai, A. Ramirez, D. Ramsey, S. Rogers, R. Ryf, R. Scotti, H. Shea, M. E. Simon, H. T. Soh, H. Tang, J. A. Taylor, K. Teffeau, J. Vuillemin, and J. Weld, “Optical MEMS devices for telecom systems,” in Proc. SPIE’s Microtechnologies New Millennium Conf., Canary Islands, Spain, May 19–21, 2003, pp. 5116–59. [30] A. Gasparyan, V. A. Aksyuk, P. Busch, and S. Arney, “Mechanical reliability of surface-micromachined self-assembling two-axis MEMS tilting mirrors,” in Proc. SPIE MEMS Reliability Critical Applications, Santa Clara, CA, 2000, vol. 4180, pp. 86–90. [31] M. Kozhevnikov, N. R. Basavanhally, J. D. Weld, Y. L. Low, P. R. Kolodner, C. A. Bolle, R. Ryf, A. Gasparyan, A. R. Papazian, A. Olkhovets, J. Kim, D. T. Neilson, V. A. Aksyuk, and J. V. Gates, “Compact 64 64 micromechanical optical crossconnect,” IEEE Photon. Technol. Lett., vol. 15, pp. 993–995, July 2003. [32] P. H. Malyak, D. L. Kent, P. R. Kolodner, and J. Crawford, “Interferometric measurements of back focal length and insertion loss of precision microlens arrays,” in Proc. SPIE Gradient Index, Miniature, Diffractive Optical Systems II, vol. 4 437 161, T. J. Suleski, Ed., 2001. [33] L. Erdmann and D. Efferenn, “Technique for monolithic fabrication of silicon microlenses with selectable rim angles,” Opt. Eng., vol. 36, no. 4, pp. 1094–1098, 1997. [34] A. Y. Feldblum, K. O. Mersereau, C. R. Nijander, and W. P. Townsend, “Methods for making microlens arrays,” U.S. Patent 5 286 338, Feb. 15, 1994. [35] C. Bolle, “Method for compensating for nonuniform etch profiles,” U.S.s Patent 6 649 073, Nov. 18, 2003. [36] N. R. Basavanhally, R. E. Scotti, and J. D. Weld, “Fiber array coupler,” U.S. Patent 6 633 719, Oct. 14, 2003. [37] R. Capik, J. Hickey, and S. Korotky, “Insertion loss model and input/output losses of single-stage optical cross-connects,” in Proc. Optical Fiber Communications Conf. (OFC 2003), vol. 1, Atlanta, GA, 2003, Paper WM2, pp. 386–387. [38] C. F. Lam, M. Boroditsky, B. Desai, and N. J. Frigo, “A novel dynamic crosstalk characterization technique for 3-D photonic crossconnects,” IEEE Photon. Technol. Lett., vol. 15, pp. 141–143, Jan. 2003. [39] A. Olkhovets, P. Phanaphat, C. Nuzman, D. J. Shin, C. Lichtenwalner, M. Kozhevnikov, and J. Kim, “Performance of an optical switch based on 3-D MEMS crossconnect,” IEEE Photon. Technol. Lett., vol. 16, pp. 780–782, Mar. 2004.

2

NEILSON et al.: 256

256 PORT OPTICAL CROSS-CONNECT SUBSYSTEM

David T. Neilson (M’96–SM’02) received the B.Sc. (Hons.) degree in physics and the Ph.D. degree in physics for work on optical nonlinearities in InGaAs quantum-well devices from Heriot-Watt University, Edinburgh, U.K., in 1990 and 1993, respectively. He was a Postdoctoral Researcher at Heriot-Watt University from 1993 to 1996, working on systems and devices for free-space optical interconnects. From 1996 to 1998, he was a Visiting Scientist at NEC Research Institute, Princeton, NJ, researching optical interconnects for high-performance computing. In 1998, he joined Bell Laboratories, Holmdel, NJ, where he worked on microelectromechanical systems (MEMS)-based cross-connects, wavelength-selective switches, equalizers, and dispersion compensators. He is currently a Technical Manager at Bell Laboratories with responsibility for the optoelectronic device growth and fabrication facility. He has more than 90 publications in the field of optical interconnects and switching. Dr. Neilson is a Member of the Optical Society of America (OSA) and a Senior Member of the IEEE Lasers & Electro-Optics Society (LEOS).

1509

Roland Ryf (M’03) received the B.S. degree in electrical engineering and the M.S. and Ph.D. degrees in physics, working on photorefractive self-focusing and spatial solitons, parallel optical processing based on holographic storage, and fast optical correlation, from the Interstate University of Applied Sciences of Technology, Buchs, Switzerland, in 1990, 1995, and 2000, respectively. He has been working in the Photonic Subsystem Department at Bell Laboratories, Lucent Technologies, Holmdel, NJ, since May 2000, where he has been focusing on the optical design and prototyping of optical microelectromechanical systems (MEMs)-based switches, spectral filters, and dispersion compensator, and the demonstration of their applications in optical networks.

J. Kim, photograph and biography not available at the time of publication.

A. R. Papazian, photograph and biography not available at the time of publication.

R. Frahm, photograph and biography not available at the time of publication.

Paul Kolodner received the A.B. degree in physics from Princeton University, Princeton, NJ, in 1975 and the A.M. and Ph.D. degrees from Harvard University, Cambridge, MA, in 1977 and 1980, respectively. His Ph.D. dissertation was on suprathermal electron emission produced by a laser-induced breakdown of fast shockfronts. He has worked at Bell Laboratories since 1980 on a variety of experimental problems, most importantly in the fields of nonlinear dynamics and pattern formation, photobiology, microlens-array technology, and thermal management of electronics. He is presently a Distinguished Member of Technical Staff in the Network Hardware Integration Department of Bell Laboratories, Lucent Technologies, Murray Hill, NJ.

C. J. Nuzman received the B.S. degrees in electrical engineering and mathematics from the University of Maryland, College Park, in 1996 and the Ph.D. degree in electrical engineering from Princeton University, Princeton, NJ, in 2000. He now works in the Mathematical Sciences Research Center of Bell Laboratories, Murray Hill, NJ. His research interests center on communications systems, including optical networks and data traffic modeling, and on optimization.

A. Gasparyan, photograph and biography not available at the time of publication.

N. R. Basavanhally, photograph and biography not available at the time of publication.

V. A. Aksyuk, photograph and biography not available at the time of publication.

C. A. Bolle, photograph and biography not available at the time of publication.

J. V. Gates, photograph and biography not available at the time of publication.