"MEMS technology - micromachines enabling the all"
MEMS Technology -- Micromachines Enabling the “All Optical Network" Steven D. Robinson* IEEE Senior Member 6 Dege Farm Rd, Califon, NJ 07830 Email: RobinsonSDR2000@aol.com Recently, this “nano-technology” has found a home in Abstract today’s high speed DWDM optical fiber networks. As bit Research on micromachines or MEMS technology has rates and the number of wavelengths per fiber increase, it is no been ongoing for more than a decade. Early work focused on longer practical to convert optical signals back to the electrical biomedical, printing and airbag applications. Only recently domain. In today’s systems, 80 wavelengths at 10 Gb/s is not has this technology been applied to telecommunications uncommon. Within a couple of years, 160+ wavelengths at 40 networks. This new micro-optics technology, known as Gb/s will be here. Demultiplexing, detecting, Optical MEMS, is enabling a new generation of optical routing/switching, retransmitting and remultiplexing 6.4 components to facilitate deployment of the “all optical Terabits/second of data per fiber electrically at every node is network”. Optical MEMS are miniature optical elements difficult, inefficient and cost prohibitive. Furthermore, system (such as micro mirrors) capable of moving and managing operators would like to transport multiple protocols over the light. Because these devices steer light beams – not electrons, same fiber with dynamic wavelength management. For all they are bit rate, wavelength (channel) and protocol these reasons, there is a great need to eliminate the multiple independent. This paper will discuss the current trends in Optical-to-Electrical-to-Optical (O-E-O) conversions in Optical MEMS research. It will discuss several optical networks. Optical MEMS is a new technology which telecommunications applications including optical switching, allows networks to move and manage light without this OEO dynamic gain equalization, variable optical attenuation and conversion. Instead, through the use of miniature optical dynamic add/drop multiplexing. The author will discuss elements (such as moving micro mirrors), optical networks some of the many challenges in implementing MEMS in field can now dynamically switch, compensate, attenuate, combine grade optical components and review some of the competing and separate signals – all in the optical domain. Optical technologies. Finally, future directions in MEMS research MEMS is driving the development of a new generation of will be discussed -- especially in the area of optical switching; optical components to facilitate this all optical network. the so-called “killer” MEMS application. Optical MEMS Components and Applications MEMS - More Than a Decade of Innovation The number of Optical MEMS applications is growing For years, engineers have worked hard to eliminate daily. In general, they can be classified into 5 main mechanical devices by replacing them with electronic or component areas: Optical Attenuators, Dynamic Gain “solid state” devices. Recently, there has been a resurgence Equalizers, Wavelength Tunable Devices, Optical Switches, back to moving micro-machines or Micro-Eectro Mechanical and Optical Add/Drop Multiplexers (OADM). Systems (MEMS) technology. Built in ways similar to Attenuators. One of the simplest Optical MEMS devices integrated circuits, MEMS devices are fabricated on silicon wafers by patterning various layers of materials and releasing is the Variable Optical Attenuator or VOA. Typically, a (underetching).1 After release, these tiny structures are moving micro-structure is designed to either partially block or capable of motion (see figure 1). If the microstructure is a decouple the lightpath4,5,6. An example of a blocking VOA is mirror, the device can move and manage light. shown below (in figure 2). In this figure, the light from the input fiber is collimated with a lens, manipulated by the Initially, MEMS research was focused on non-telecom MEMS chip (in this case partially blocked or attenuated) and applications such as air-bag sensors, pressure sensors, recoupled to an output fiber. The MEMS chips itself could be biomedical devices, RF Relays and ink jet printing devices.2. actuated horizontally or vertically. Actuators could be Several million devices are shipped every year, accounting for electrostatic, thermal or electromagnetic. Such a device could more than $3B in annual sales.3. be a serve as an on-off switch, as well. Metal (Au) Micro-structure Oxide Motion Poly 2 Poly 1 Poly 0 Electrode Nitride Silicon Substrate “Post Release” Figure 1. Surface Micromachined MEMS Device (Pre & Post release) * Formerly with Lucent Technologies/Agere Systems (Breinigsville, PA) 0-7803-7038-4/01/$10.00 (C)2001 IEEE 2001 Electronic Components and Technology Conference A variation on an optical attenuator is accomplished by actuator is reflective on both sides, the same “drop” bouncing a light beam off of a deformable Fabry-Perot microstructure can be used to “add” the incoming optical cavity7. If properly designed, this MEMS device can signal (represented by the dashed line). Typically, these transform from fully reflective (3/4λ gap) to fully attenuating planar NxN, non-blocking switches are limited to N=32 and (1/2λ gap) through destructive interference. below. Since the number of switch elements increases by N2, Dynamic Gain Equalization. By combining a diffraction N=32 requires 1024 low-loss, reliably working MEMS grating (which separates out wavelengths spatially) with an mirrors. Manufacturing yield and low loss performance become problematic at these switch sizes. Optical packaging of multi-array fibers is non trivial as well – especially if a SM Fiber SM Fiber hermetic package is required. Another drawback is the large loss variation between the closest two fibers and the furthest two (not desirable in optical systems). Lens MEMS Chip For larger Optical Switches, most companies are pursuing the so-called “3D” or beam steering type switch architecture 14,15, 19 Figure 2. Simple Optical MEMS component: VOA . This design requires an array of analog, two axis tilt array of interference type attenuators, Ford et al have Add demonstrated broadband dynamic gain equalization8. Such a device is crucial to simultaneously adjust (and level) multiple wavelengths in today’s high speed, optically amplified DWDM systems. Input Fiber Array Outputs (thru) Wavelength Tunable Devices. By extending the variable cavity concept, it is possible to use MEMS devices to create both tunable lasers and wavelength tunable filters9,10. Optical Switching and OADMs – Perhaps the most common Optical MEMS application today is optical switching11. By redirecting a light beam from one fiber to another, one can use MEMS devices to optically “switch” signals. There are several types of optical switches: MEMS Mirror = Open (down) • Conventional Switches: 1x2, 2x2 and 1xN switches Drop Fibers = Closed (up) • Small NXN switches and switch arrays (with N < 32) • Large NXN Optical Cross Connect (OXC) switches The simplest form is the conventional digital switch. By Figure 4. NxN, Non-blocking Optical Switch having a MEMS mirror “popup” or bisect the light beam during actuation, light is directed or switched from one fiber mirrors. Light comes in from a fiber, is collimated, bounces to another (figure 3). Note that these switches typically use off of the first MEMS tilt mirror, bounces off of a fixed collimating lenses to reduce losses and are usually latching. reflector to the second MEMS mirror, which is then directed In general, these planar or “2D” type conventional to an output fiber. Any input port can be connected to any switches can be extended up to small NxN type switches. A output port -- i.e. it is a strictly nonblocking architecture! 8x8 switch is shown in figure 4, below. Again, a moving Note that the MEMS mirror array is aligned with the microstructure (mirror) bisects a particular lightbeam row, fiber/lens array. Correspondingly, there are 2 mirrors for causing the input port to be “switched” to an output port12,13.. every input port or 2N switching elements (mirrors) total. One feature of this architecture is the ease of using this device Thus, a 1000 port switch would require only 2000 mirrors in as an optical add/drop multiplexer, as well. Note that, if the the beam steering case versus 1,000,000 mirrors in the planar case. This tradeoff comes at a price. Each mirror requires precision analog control. Latchability is not easily achieved. To achieve and maintain pointing accuracy, some method of feedback may be required. Much progress has been made recently in the field of large port count switches. Real world, live traffic proof of this architecture was demonstrated last fall by Lucent Technologies. Lucent’s “Lambda Router” is the world’s first commercial deployment of a 3D MEMS based, 256x256 all optical switch16. Several companies expect to sample 1000x1000 switches this year. figure 3 Simple 2x2 MEMS Optical switch 0-7803-7038-4/01/$10.00 (C)2001 IEEE 2001 Electronic Components and Technology Conference OADMs. By combining a wavelength demultiplexer and to obtain and maintain mirror flatness over temperature. This multiplexer with a number of MEMS switches, it is possible to problem is exacerbated by the fact that mirrors typically need perform optical add/drop multiplexing, as well. For most to be large (at least 500um) to match the collimated light flexibility, a full NXN switch matrix between the demux and beam. mux would be desirable. However, for most early Bulk micromachining. It is also possible to fabricate applications, a simpler, smaller array of 1x2 or 2x2 switches is Single Crystal Silicon (SCS) structures through a “bulk” adequate (since typically only 25% of the wavelengths are micromachining process19. These devices have the advantage dropped at any one node). of a much more rigid structure – allowing designers to fabricate larger, flatter reflective surfaces. This process may use either wet or dry anisotropic etching (or both). Early MEMS Moving devices were difficult to form when the starting wafer was Mirror pure silicon (due to etch times and the lack of an etch stop). Array More recently, the availability of Silicon-On- Insulator (SOI) wafers has caused a resurgence in the use of bulk micromaching techniques17. With SOI wafers, a thin top layer of single crystal silicon (typically 50um—100 um) is separated from a larger silicon substrate (known as the “handle”) by a thin oxide layer (typically a few thousand Fixed angstroms). The oxide serves as an etch stop and a Fiber/ Mirror Lens microstructure release mechanism. As shown in figure 6 Array below, KOH etching is often used to form the backside well (through the handle). Deep RIE etching is typically used to define the top device layer (microstructure) after patterning. SCS Design. As stated above, very rigid, flat, complex structures can be formed using bulk machining. Deep RIE etching (using the Bosch process) can form nearly vertical Figure 5. Large NxN Switch using “3D” beam steering15 walls and high aspect ratio microstructures (greater than 100:1). The two main drawbacks of this technique are 1) Technology Choices and Challenges Potentially long etch times (hours) and 2) Single layer Developing Optical MEMS components requires processing (requiring subsequent wafer bonding for added experience in MEMS device design and fabrication, MEMS functionality). chip packaging, free space optical design and optical Because MEMS device features are relatively big (as component packaging. compared to 0.1 um features sizes on high speed VLSI circuits), these technologies do not push the limits of Chip Design and Fabrication lithographic and wafer processing technology. In general, So how does one create moving micro structures? To last generation semiconductor processing equipment can be begin, Optical MEMS requires microfabrication of a silicon used—often at firesale, fully depreciated costs. Several wafer. The use of silicon as a mechanical substrate has been foundries offer both surface and bulk micromachining process explored for more than 25 years17. There are two commonly capabilities. used silicon microfabrication techniques: surface machining and bulk machining. Deep RIE Etched Surface micromachining (SMM). In this method, layers of Micro- Top layer SCS structure polysilicon, nitrides, oxides and metals are selectively deposited on a wafer surface using LPCVD18. Usually, gold or Oxide aluminum is deposited on top to provide an optically reflective Silicon Substrate surface. After processing, the devices are HF etched to (“handle”) remove the oxide layers only (as shown in figure 1). Often, KOH Etched this is done with a supercritical CO2 dryer to eliminate trapped liquids and solvents which could cause stiction problems. This Figure 6. SOI wafer used to fabricate SCS microstructure 17. so called “release” process frees the suspended structures (such as micro-mirrors) – allowing them to move in one or Making Them Move more axis. In order to move these tiny structures, several methods SMM Design. With the proper choices of materials, it is exist. Most common ways include electrostatics, possible to design 3-dimensional structures which “pop-up” electromagnetics and thermal actuation (or some combination from the surface for later actuation. Most often, this moving of the three). Each has their pluses and minuses. microstructure is a micro-mirror. The big drawback is that Electromagnetic Actuation. Early on, many designs used these suspended structures are typically made of polysilicon. electromagnetic actuation (many still do today—depending on Due to the material and stress mismatches, it is very difficult 0-7803-7038-4/01/$10.00 (C)2001 IEEE 2001 Electronic Components and Technology Conference the application). Electromagnetic control usually requires stops. A unique, 2x2 latching switch was demonstrated by B. one or more coils and a permanent magnet on the Hichwa et al 20 Unfortunately, as described previously, many microstructure. The big disadvantage here is size, weight and Optical MEMS devices use analog actuation. Hence, power dissipation. The size of the coils and the magnets achieving latchability in practice is often difficult if not make fabricating dense arrays (such as used in cross connects) impractical. very difficult. The size and weight of the electromagnetically Control Electronics acutuated microstructures can result in potentially low resonsant frequency structures (making high speed control Since most MEMS devices are dynamic (that’s the whole more difficult). Furthermore, if high current is required to point), they require electronics which can actively control or continuously hold position (non-latching), power dissipation adjust the MEMS device. Whether it’s through high current (and heat) can be a problem—especially in large port count or high voltage, the electronics must quickly and accurately OXCs. position the microstructure. If the transfer characteristics of the device (angle vs. control voltage/current) are variable, Electrostatics actuation. More recently, a number of active feedback and/or individual microstructure testing may MEMS devices have been designed using electrostatic forces be required. In cases where you have multiple actuators, for actuation. By applying a voltage differential between the control electronics size may become an issue. For example, microstructure and a fixed electrode , one can make a device with a 1000x1000 port optical switch, you would have 2000 move. Unfortunately, these forces are relatively weak-- mirrors. If each mirror has 4-8 control leads, a subsystem requiring high voltages (>30V) in some cases. As with other would require up to 16,000 control leads and 16,000 sets of methods, there is a danger of snapdown (usually when the electronics (Buffer, D/A, S/H, analog amp). Clearly this is an device exceeds 1/3 the distance to the electrode). Unless opportunity for further integration. In addition, such a device properly designed, snapdown can cause permanent stiction or requires a micro controller to receive control commands, look even damage to the devices. The major advantage with up current states, drive the Optical MEMS devices to new electrostatic actuation is very low power consumption and positions, process feedback information and hold the new small size. Electrodes can be patterned on or under the states (over time, temperature, shock and vibration). microstructures – no coils or magnets are required. Once a device is moved to it’s position, no additional power is MEMS Chip Packaging required (other than to cover leakage currents and the control As mentioned previously, a particular challenge is properly electronics). releasing and handling such small, often fragile devices. Once Thermal heating actuation. A third actuation method is they are released or etched, it is extremely difficult to separate mechanical expansion and contraction through thermal heating chips out without contaminating or damaging them (traditional and cooling. While possible for simple devices such as wafer sawing sends particles everywhere). New methods are latching 1x2 switches, the power and size implications often being developed now to allow for volume packaging at the preclude using this method for medium and large switch chip and wafer level. Depending on the function, it is matrices. desirable to package devices in a sealed package with a Sensing and Latching - One other challenge is to window to reduce contamination (particle or moisture). determine whether your optical microstructure actually moved Unfortunately, this raises new problems with chip placement, to where you expected it to. Fortunately, if you get it close working distance and optical reflections. enough, you can often use the received or detected light beam Other Challenges to verify this. But, for large numbers of light ports (such as in Optical Design and Packaging. Developing a robust, a large cross connect), this can be an expensive and slow reliable optical MEMS component is non-trivial--many non- solution. Ideally, you would like the device to run open loop, optical MEMS developers overlook the challenges here. It with some local sensing and latching capability. requires experienced free-space optical designers and Sensing. Significant research and several patents on packagers. Providing a temperature insensitive, particle free, MEMS sensing have been filed over the last few years. Most mechanically stable environment requires careful design and methods center around optical, capacitive or piezo electric simulation. Since most of these devices are for single mode techniques. fiber applications, sub-micron alignment tolerances are Latching. Latchable MEMS optical devices would be required. High port count designs need to handle/package desirable for several reasons. For one, significant power can large numbers of fibers, lenses, micromirrors and electrical be wasted in “holding” the state (especially with thermal or control leads. While most of the specific design problems Electromagnetic actuation). Second, drift and movement due have already been discussed above, there is one common to shock, vibration and temperature variations is less of an theme: Optical MEMS designers need to provide light issue. Third, in certain telecommunications applications it is collimation and focusing, wavelength separation, precisely desirable that the state not be lost when power is temporarily controlled microstructures (which are large, flat and highly interrupted. reflective) and significant control electronics. Many of the digital switches can be made latchable through detents, permanent magnets, electrostatics or other 0-7803-7038-4/01/$10.00 (C)2001 IEEE 2001 Electronic Components and Technology Conference Table 1. Comparison of Technology Alternatives for Wavelength Testing. Optically testing dynamically variable Management/Fiber Management Components components is difficult. In some cases, due to MEMS device variability, it is more efficient to fully test after Technology Cost Perf Scale Reliab. Integ. Maturity the electronics has been combined with the moving MEMS ++ + ++ TBD ++ = microstructure. For optical switch matrices, testing and Micro-motors/discr. -- ++ -- = -- ++ test time quickly become a limiting production step. As LCD -- ++ = -- -- + the number of ports grow, the number of possible switch Planar Waveguide + + = + ++ = states to test is still N2 (even with the 3D beam steering Solid State/SOA -- = = TBD + -- architectures). Testing a 1024x1024 switch, for ++ Strong +Good = Moderate -- Weak example, would require switching and verifying more than one million optical connections! While parallel testing Future directions can be done, it would require multiple, expensive laser sources The trends are clear: 1) Lower losses through flatter mirrors, and a flexible test architecture. better control and better quality lenses. Currently many Reliability/Qualification - MEMS devices are still MEMS devices require optical amplification or some O-E-O relatively unproven in telecommunications applications. conversion to make up for higher losses. The move to single Traditionally, optical components must meet stringent crystal silicon devices will help here; 2) More ports to Telecordia type qualification requirements. While several handle the explosion of internet traffic at the core of the field trials are currently on-going, most Optical MEMS network. 4,000x4,000 port MEMS based Optical Cross devices have not been fully qualified yet. Many devices have Connects are now being considered; 3) Creative Chip been cycled billions of times with no failures. Since the base packaging is needed to reliably and cost effectively package material is silicon, most have confidence that any reliability hundreds if not thousands of these tiny, somewhat fragile issues will be overcome. One of the biggest concerns is shock structures. In some cases, thousands of control leads need to and vibration. Fortunately the mass of MEMS devices is very be brought out. Better release, assembly, handling and sealing small. Hence the forces built up during such tests is small. So, processes will be required to make Optical MEMS a volume while somewhat unproven in Telecom applications, MEMS based, low cost business; 4) Electronics integration with the devices have been reliably used in air bag sensors for years. MEMS device will eventually be required to practically Early data suggests that Optical MEMS devices will be implement many proposed systems (providing better control equally reliable. and more dense solutions), and 5) Industry Standards-- Competing technologies Currently there are no multisource or industry standards As shown in table 1, many competing technologies exist defining package footprints, connections or control interfaces. which could realize the various wavelength and fiber In the end, all of the above need to develop for Optical MEMS management component functions described above. components to truly take market share. Companies have been using small DC motors (micro motors) Conclusions to perform switching functions for years. Recently, several The author has attempted to describe the state of Optical companies have developed Liquid Crystal (LCD) devices with MEMS development and Research based on recent blazed gratings to perform optical switching, OADM and publications, talks and seminars. Next generation, high bit dynamic gain flattening21 functions. Planar waveguides can rate optical systems will demand the flexibility provided by also perform many of the same tasks as MEMS and LCDs. dynamically tunable, Optical MEMS devices. Several Unlike these technologies, waveguides require no free-space technical challenges remain, and the competing technologies optics—they are fully integrated. However, insertion loss, are not sitting still. Yet, this new technology has the potential polarization dependent loss, power dissipation and scalability to revolutionize the optical components industry. New issues may limit their uses. Solid state technologies such as functions never before envisioned are now being proven in Lithium Niobate & Semiconductor Optical Amplifiers can 22 the lab and in field trials. Considered all but impossible five perform fast switching but at a high cost . For small switch years ago, large port count (1000+), “All optical” cross matrices (32x32 and below) it will be a horse race between connect switches are now practical due to MEMS. Bit rate these technologies. independent, MEMS based optical components will “future A unique combination of waveguides and bubble jet proof” next generation networks. In the end, moving and technology was demonstrated by Agilent early last year. In managing photons instead of electrons will accelerate March, 2000, Agilent announced one of the first highly deployment of the “all optical network”. integrated, fully non blocking 32x32 switches23. Again, power dissipation, variable loss and scaleability remain a concern here. For larger switch fabrics, most concede that 3D optical MEMS is the only “all optical” technology which can reliably integrate such complex switching functions in a small package. 0-7803-7038-4/01/$10.00 (C)2001 IEEE 2001 Electronic Components and Technology Conference References  A. Neukermans, R. Ramaswami, “MEMS Technology for Optical Networking Applications”, IEEE  Proc. IEEE vol. 86, No. 8: Special Issue on ”Integrated Communications Mag., Jan, 2001, pp 62-69. Sensors, Microactuators,, and Microsystems (MEMS)” (1998).  B. Hichwa et al.,” A Unique Latching 2x2 Fiber optics Switch,”, Proc. 2000 IEEE/LEOS Int’l Conf. Opt.  B. Travis, “MEMS sensors and mechanical gadgets enter MEMS, Kauai, HI, Aug., 2000,post deadline paper. the mainstream”, EDN, May 21, 1998 pp 77-86.  A. Ranalli et al, “Liquid Crystal Based Wavelength  Robert Bratter, “Commercial Success in the MEMS Selectable Cross Connect, ” Proc. Euro. Conference on Marketplace”, Proc. 2000 IEEE/LEOS Int’l Conf. Opt. Optical Comm. (ECOC ’99), V3. MEMS, Kauai, HI, Aug., 2000,pp 29-30.  T. Murphy, A strictly Non-blocking 16x16 Eletro-optic  B. Andersen et al, “MEMS Variable Optical Attenuator Photonic Switch Module,” Proc. Euro. Conference on for DWDM Optical Amplifiers”, Poster Session, Optical Comm. (ECOC ’99), V4, OFC2000, Baltimore, USA. pp 93-94.  R. Wood et al, “ A MEMS Variable Optical Attenuator,”  J. Fouquet, ”Compact optical cross-connect switch based Proc. 2000 IEEE/LEOS Int’l Conf. Opt. MEMS, Kauai, on total internal reflection in a fluid-containing planar HI, Aug., 2000,pp 121-122. lightwave circuit, ”. TuM1-2 Proceedings, OFC2000,  S. Sumriddetchkajorn et al , “Micromachined Fault Baltimore, USA, 204-205. Tolerant , High Resolution, High Speed Programmable Fiber Optic Attenuator,” OFC Tch Dig. 2000, THQ1-1, pp. 240-242.  J. Ford, J. Walker, D. Greywall, K. Goossen, J. of Lightwave technology, V16, 1663, 1998.  J. Ford, J. Walker, K. Goosen and D. Neilson, “ Broad Spectrum Micromechanical Equalizer”. Proc. Euro. Conference on Optical Comm. (ECOC ’99).  D. Vakhsoori et al, “2 mW CW Single-Mode Operation of a Tunable 1550 nm VCSEL with 50nm Tuning range”, Elec. Letters, vol 35, no. 11, May 1999 pp1-2.  C. Chang-Hasnain, “Tunable VCSELs,”, Proc. 2000 IEEE/LEOS Int’l Conf. Opt. MEMS, Kauai, HI, Aug., 2000,pp 29-30.  L. Lin, E. Goldstein, and R. Tkach, “Free space micromachined optical switches for optical networking,” IEEE J. Selected Topics in Quantum Electronics, vol. 5, p 231-7, 1996.  Y. Lin, “Micromachined Free Space Switches with submillisecond Switching Times,” OFC Tech. Digest, 1998, pp. 147-148.  Y. Lin, E. Goldstein, “Lightwave Micromachines for Optical Cross Connects,” Proc. Euro. Conference on Optical Comm. (ECOC ’99), V4.  D. T. Neilson et al, “Fully Provisioned 112x112 Micro- Mechanical Optical Cross connect with 35.8 Tb/s Demonstrated Capacity”, Post Deadline Papers, OFC2000, Baltimore, USA, PD12-1-3  D. Bishop, “Applications of Lightwave Micromachines”, Electronicast Conference on Advanced Photonics, Nov.3, 1999, San Mateo, CA.  M. Fuller, ”Long anticipated all optical switches begin to reach customers,” Lightwave Magazine, Dec, 2000, p1,44.  K. Petersen, “Silicon as a Mechanical Material,” Proc. IEEE, vol. 70, no. 5, 1982, pp. 420.  R. Howe, “Surface Micromachining for Microsensors and Microactuators,” J. Vac. Soc. Tech. B., vol. 6, 1988, pp 1809-1813. 0-7803-7038-4/01/$10.00 (C)2001 IEEE 2001 Electronic Components and Technology Conference