`AN OPTICAL CROSS-CONNECT
`
`Andrew S. Dewa and John W. Orcutt
`Telecommunications Business Unit, Materials & Controls Group
`Texas Instruments
`Dallas, TX 75243
`Marshall Hudson and David Krozier
`Telecommunications Business Unit, Materials & Controls Group
`Texas Instruments
`Attleboro, MA 02703
`
`Alan Richards and Herzel Laor
`Astart~ Fiber Networks
`Boulder, CO 80301
`
`ABSTRACT
`
`This paper describes the development of a two-axis silicon
`micromirror for a 480 × 480 optical cross-connect (OXC). The
`micromirror is an electromagnetically actuated, analog silicon
`mirror. The range of motion for the micromirror is +8°. Full range
`of motion open-loop moves have been demonstrated in under 5 ms,
`which enables a full 480 ×480 switch reconfigatration in 10 ms.
`The micromirror and OXC have been optimized for low loss. The
`average optical insertion loss is -2.5 db at 1550 nm and -2,9 db at
`1310 nm, the two primary wavelengths in fiber optic data
`traz~smission, SONET data at 2.4 Gbits/sec has been transmitted
`through the OXC with no loss penalty.
`The preliminary life and environmental validation data for the
`NxN Micromirror is reported. Bare micromirror samples have
`passed both the GR-63-Core vibration specification (1-100 Hz
`0.5G, 3-axis; 100-500 Hz, 3G, 3-axis) and temperature extreme
`specification (-40°C to 70°C) for devices in their shipping
`containers. We also have shock tested a small sample of
`micromirrors to failure. These fail at 800-900G shock.
`Micromirrors have been cycled to full deflection for 30 million
`cycles (6× specified life for the product) with no change in
`performance.
`
`INTRODUCTION
`
`The need for all-optical switching is being driven by
`exponentially increasing demand for bandwidth and the use of
`dense wavelength division multiplexing (DWDM) in all optical
`networks. As the number of wavelengths carried on each fiber
`grows, current methods of optical/electfical/optical (OEO)
`switching become unfeasible. In OEO switching, the light from
`one fiber must be first separated into its individual wavelengths.
`The time domain multiplexed signal on each wavelength is
`demulitplexed down to lower data rate electrical signals. The
`electrical signals are then switched. The signals are mulitplexed
`back up in data rate and the wavelengths are recombined for
`transmission to the output fiber. As the number of wavelengths per
`fiber grows, the space and power consumption of OEO quickly
`becomes too great.
`In an all-optical cross connect (OXC), no electrical conversion
`is needed. All wavelengths are switched simultaneously. At
`present, all announced, large-port-count (N_>256) OXCs are
`MEMS-based, because of the ability of this technology to produce
`high-precision, micromechanical moving mirrors. Typical single-
`mode optical fiber cores are 9 pan in dianaeter, so that the emerging
`optical beam is compatible with the size of typical MEMS optical
`components. The actual size of the micromechanical mirror
`
`depends on a complicated relationship of angle of view, the beam
`diameter, optical magnification, and the length of the optical path.
`Furthermore, since each mirror needs individual control and an
`optical fiber interface, the control electronics together with optical
`connector interfaces to support hundreds to thousands of
`micromirrors can dwarf the optical MEMS. In addition, the
`telecommunications industry requirements on system reliability
`mandate redundant power supplies and system controllers, so that
`the final size of the OXC system can be several equipment racks.
`Texas Instruments (TI) and Astart6 Fiber Networks (AFN) are
`jointly developing a 480 x 480 OXC. The TI/AFN approach
`utilizes individually-packaged, micromachined silicon mirrors to
`perform the switching function and active optical feedback to
`maintain low-loss, long-term connections. This approach provides
`a high degree of modularity, so that the customer can purchase the
`number of ports needed now, and further populate the switch frame
`as demand increases. Modularity also makes the OXC field
`serviceable-a failed module can be replaced without disrupting
`traffic on other modules.
`This paper will discuss the development of the two-axis
`silicon micromachined mirror. It will include a general description
`of the OXC architecture, the fabrication and operation of the
`micromirror, preliminary life and environmental validation tes~ting
`results, and the optical performance of the micromirror in the
`switch matrix. For clarity, the silicon mirror will be referred to as
`the NxN Micromirror throughout the rest of the paper.
`
`OPTICAL CROSS CONNECT
`
`Texas Instruments and Astart6 Fiber Networks are jointly
`developing a 480 x 480 free-space, optical cross connect (OXC)
`[1,2]. The OXC is single-stage and non-blocking, allowing any-.to-
`any connectivity. TI is developing the Optical Unit (OU) capable
`of switching the optical signals from one fiber to another, while
`AFN has developed the overall OXC architecture. The NxN
`Micromirror provides the beam steering function for the OU. The
`OU also contains the focnsing optics and the servo control emitters
`and detectors. AFN combines 4 OU’s with a control electronics
`board into an Optical Module. The optical module is the minimum
`granularity of the OXC. Two hundred and forty modules, arranged
`in a 5 by 24 array make up a fully populated switch. All the optical
`units are pointed at a fixed mirror one meter away to give a 2-meter
`optical path length. A schematic showing four Optical Modules is
`given in Fig. 1.
`The fundanrental optical path of the OXC is shown in Fig 2.
`The light exits the fiber and goes through focusing optics. The
`expanded beam is reflected off a fixed mirror to the NxN
`Micromirror. The beam is steered by the NxN Micromirror to the
`
`0-9640024-3-4
`
`93
`
`Sofid-State Sensor and Actuator Workshop
`Hilton Head Island, South Carolina, June 4-8, 2000
`
`Petitioner Ciena Corp. et al.
`Exhibit 1020-1
`
`
`
`Figure3. Optical photograph of an NxN Micromirror die.
`
`top of the wafer and the bottom of the wafer is 0.2 p,m for the
`100gm etch depth.
`The packaged NxN Micromirror is shown in Fig. 4. The
`package was partitioned into a hermetic portion, the ceramic header
`and window that holds the NxN Micromirror, and a non-hermetic
`portion, where the coils for the actuator, wiring harness and
`mounting bracket are located. The micromirror is hermetically
`packaged to guarantee the 20-year life expected by the
`
`Figure I. A schematic drawing showing 4 optical modules.
`
`target OU via a large fixed mirror. This folding of the optical path
`allows for a smaller OXC footprint. It also allows for the doubling
`of the size of the switch, by replacing the mirror by another array
`of optical modules.
`The two Micromirrors are servo controlled by emitters and
`detectors using the same optical path as the light from the fibers.
`Active servo control is key to achieving and maintaining very low
`insertion loss. The active servo control means that the connection
`is not susceptible to vibration, shock and long term drifts.
`
`NxN MICROMIRROR FABRICATION
`
`The NxN Micromirror is an electromagnetically-driven,
`analog, two-axis silicon mirror. A photograph of the mirror die is
`
`Singie-mode
`Fiber
`
`Unit
`
`-Free space span: 2 meters
`-Beam diameter: 2.6 mm
`-Servo controlled
`micromirror
`
`Single-mode
`Fiber
`
`Fixed
`mirror
`
`Optical emitters and detectors for servo control not shown
`
`Figure 2. The optical path of the 480 x 480 OXC.
`
`shown in Fig. 3. The active part of the silicon mirror is a 3.2 x 3.6
`mm oval which is supported by 2 sets of silicon torsional hinges,
`one set between the rmrror and the gimbals, and the second set
`between the gimbals and the flame. The size of the mirror is set by
`the beana diameter in the optical path. The two sets of torsional
`hinges allow the independent movement of the mirror in two axes.
`The overall die size is 7.2 mm x 9.0 ram. The mirror is about 0.1
`mm thick and is gold coated to maximize its reflectivity at the
`wavelengths of interest for telecommunications, 1.25 lain to 1.65
`~m.
`
`The NxN Micromirror is fabricated using photolithography
`and reactive ion etching via the Bosch Process [1]. The etch recipe
`was optimized for vertical sidewalls with little undercut. The
`typical difference between the width of the torsional hinges at the
`
`Figure 4. A photograph of the packaged NxN Micromirror.
`A US dime is shown for scale.
`
`telecommunications industry.
`
`MICROMIRROR ACTUATOR
`
`The electromagnetic actuator for the mirror ntilizes drive coils
`below the micmmirror and permanent magnets mounted on the
`mirror and gimbals frame, as shown in Fig. 5. This actuator was
`chosen to allow the tips of the mirror m~d the gimbals to move
`several hundred microns providing the required deflection of _+8°
`in both axes with the millimeter-size of the micromirror.
`Figure 6 shows actual deflection data from an NxN
`Micromirror prototype plotted on the magnetic model curves.
`Because there is no ferromagnetic material in the magnetic circuit
`other than the permanent magnets, analytical methods could be
`used to simulate the actuator [4]. This model, along with finite
`element analysis, was used to optimize the actuator design. The
`
`94
`
`Petitioner Ciena Corp. et al.
`Exhibit 1020-2
`
`
`
`Mirror M~ nets
`
`et
`Range
`
`Mirror Drive Ceils
`
`Figure 5. Schematic drawing of NxN Micromirror actuator
`mechanism.
`
`mirror and gimbals hinges are identical in geometry and fall on the
`same straight line. Perfectly linear behavior is what is expected for
`single crystal silicon hinges. The model fits the experimental data
`to within the experimental errors of the measurement.
`The OXC makes a connection by first steering the beam to the
`new location as fast as possible with open-loop control. Then an
`acquisition takes over to quickly move the remaining distance to
`the center position, at which point the closed-loop serv’o control
`takes over. Since the hinges are perfectly elastic, the micromirror
`is an underdamped system. Therefore, the dynamics of the system
`were modeled and a specified waveform is generated for each
`move. A move of 15.6° of the mirror plus gimbals under open-
`loop drive is shown in Fig 7.
`The mirror plus gimbals moved 15.6 degrees in under 5 ms
`and was stable. The stability shown in Fig. 7 is more than enough
`for the acquisition and servo control to take over and optimize the
`connection. The under 5 ms move time is also sufficient to
`guarantee a full OXC reconfiguration in 10 ms.
`
`ENVIRONMENTAL AND LIITE VALIDATON TESTING
`
`Key to deploying a product in the fiber optic
`telecommunication market is demonstrating a reliable product
`which meets the Telcordia specifications. We have begun
`validation tests on the environmental and life of the NxN
`Micromirror. As one would expect with single-crystal silicon
`hinges, cycling mirror over its full range of motion in both axes for
`over 30 million cycles results in no measurable change in the
`
`Figure 6. Measured hinge force vs. mirror deflection
`data plotted on the magnetic force model.
`
`95
`
`Figure 7. Measured position and current waveform for an
`open-loop move of 15.6 ~in under 5 ms (Note: on the Y-axis the
`angle is in radians and the current is in amperes).
`
`micromirror performance. Thirty million cycles is six times the
`product life specification of one reconnect every minute for 20
`years. Since each cycle is move forward and back, one represents
`cycle two connections.
`We have tested the NxN Micromirror to the GR-63-Core
`shipping temperature specification -40°C to +70°C. The initial test
`samples have passed this test with no change in operating
`characteristics. Further testing is underway with additional
`samples to improve out confidence levels.
`Upon seeing the NxN Micromirror, the initial reaction of most
`engineers is to question its shock and vibration sensitivity-because
`there are relatively large magnets at the ends of long moment arms
`supported by thin silicon hinges. To validate the design, we have
`begun testing the NxN Micromirror to the more severe GR-63-
`Core specification for devices shipped in turboprop aircraft. It
`should be emphasized that the GR-63-CORE specification is for
`devices in their shipping containers, including all padding, etc. We
`tested 4 sanlples of the NxN Micromirror bolted directly to the
`shaker table to this specification: 1-100Hz at 0.5G and 100-500 Hz
`at 3G in all three principal axes. (The standard GR-63-CORE
`shipping test is 100-500 Hz at 1.SG.) All samples passed the test.
`Again, further testing is underway to improve our confidence
`levels.
`We have also tested NxN Micromirrors to failure under shock.
`Three mirrors passed a 500G shock test with a 6 ms duration, in all
`three axes. We took 2 of the mirrors to failure under 800-900 G
`shock tests with a duration of 2.5 ms. The duration of the shock at
`higher G levels was reduced due to equipment limitations. Further
`shock testing is underway.
`The environmental and life validation testing results to date
`are summarized in Table 1.
`
`Table 1. Environmental and life test results
`Life Testing
`30 million c),cles (3x expected life)
`Shipping Thermal
`-40°C to 70C with no detectable change
`Extremes
`in performance
`Shock survivability
`Survives 500G 3-axis, 6 ms duration
`Shock failure
`800-900G 2.5 ms duration
`Vibration test
`1-100 Hz 0.5G, 3-axis; 100-500 Hz, 3G,
`3-axis
`(GR-63-Core in
`shipping container)
`
`Petitioner Ciena Corp. et al.
`Exhibit 1020-3
`
`
`
`2. H. Laor, A. Richards, E. Fontenot, "576x576 Optical Cross
`Connect for Single-Mode Fibers", Conference Proceedings of the
`1999 Annual Multiplexed Telephony Conference, San Diego, CA,
`July 19-22, 1999, pp. 343-349.
`
`3. F. Laermer and A. Schilp, Robert Bosch GmbH, "!vlethod
`for Anisotropic Plasma Etching of Substrates", U.S. Patent
`5498312 (1996).
`
`4. E. P. Furlani, "% Three-dimensional Field Solution for Axially
`Polarized Multipole Disks," Journal of Magnetism and Magnetic
`Materials, !35, 205-214 (1994).
`
`OPTICAL PERFORMANCE
`
`The OXC is designed to operate at the two major wavelengths
`utilized in optical fiber communications, 1310 nm and 1550 rim.
`The coatings on the optical unit lens, optical elements and the
`rnicromirror window are optimized for the band-pass from 1250 to
`1650 nm so that the OXC is fully compatible with dense
`wavelength division multiplexing.
`The measured optical insertion loss at 1310 ran averaged -2.9
`db with a maximum value of -3.2 db. At 1550 nm the insertion
`loss averaged -2.5 db with a maximum value of -3.0 rib. Furtker
`testing of the optical loss is ongoing.
`One key metric for an optical switch is to measure any excess
`data loss in a real telecommunications data channe! other than that
`due to the loss of optical power measured by its insertion ioss.
`This is done by comparing the data loss rate for the OXC to that of
`a passive optieaI attenuator with the same attennation as the
`measured insertion loss of the connection. This is called the optical
`loss penalty. The OXC was able to transmit 2.4 Gbits/sec. of
`SONET data with no measurable loss penalty.
`The measured optical performance of the NxN Micromirror is
`
`Table 2. NxN Micromirror OXC Performance
`Optical Loss
`-2.9 db ave.; 4.2 db max.
`(1310 nm)
`Optical Loss
`
`-2.5 db ave.; -3.0 db max.
`
`5 ms for 15.6° move (measured)
`
`.... (15513 nm)
`2.4 Gbits/sec SONET No Measurable L0;~ Penalty
`Data transmission
`Time for a full range
`move
`Time for a full switch
`reconfiguration
`Deflection
`
`10 ms (projected)
`
`_+8° in two independent axes
`
`summarized in Table 2.
`
`SUMMARY
`
`We have demonstrated a two-axis silicon micromirror for a
`480 x 480 optical cross connect. The micromin-or is
`electromagnetically actuated, and can move, open-loop, 15.6° in
`under 5 ms to enable a full switch reconfiguration in 10 ms. The
`average optical insertion loss for the optical switch is -2.5 db at
`1550 nm and -2.9 db at 1310 nm. The optical switch has
`transmitted 2.40bits/sec SONET data with no loss penalty.
`We have begun life and environmental testing of the NxN
`Micromirror. Bare micromirror sampies passed the GR-63-Core
`vibration specification mad temperature extreme specifications for
`devices in shipping containers. We also shock tested a small
`sample of micromirrors to failure. These failed at 800-900(3. We
`tested nficromirrors to 30 million cycles (3x expected life) with no
`change in performance. Further life and environmental testing is
`ongoing to get enough data to fully predict the life and reliability of
`the NxN micromirror.
`
`REFERENCES
`
`1. D. Krozier, M. Hudson, J. D’Entremont, H. Loot, A. Richards,
`and E. Fontenot, "’Performance of a 576 × 576 Optical Cross
`Connect", Technical Proceedings of the National Fiber Optic
`Engineers Conference, Chicago, Illinois, September 26-30, 1999,
`voh 1, p276.
`
`96
`
`Petitioner Ciena Corp. et al.
`Exhibit 1020-4
`
`