Technology borrowed from disk-drive development could produce more reliable MEMS devices.
DR. CHAK LEUNG, Blue Sky Research
The relentless growth of telecommunications traffic is exceeding that of microprocessor performance, which Moore's Law tells us doubles every 18 months or so. Nowhere are we made more painfully aware of this than when confronted with the performance of optical switches, where electronics and photonics collide.
Today, most optical switching is accomplished using relatively costly SONET-based digital crossconnects. As bit rates and channel counts increase, this optical-electrical-optical switching becomes a growing bottleneck, slowing transmission over the metro and long-haul. The all-optical crossconnect (OXC) switch is a key component that will increase the throughput of optical networks by orders of magnitude.
All-optical switches pass encoded light pulses directly from one optical fiber to another without electrical conversion. Among the various technologies that have been developed to perform optical-switching functions, micro-electromechanical systems (MEMS)-based switching has so far been most widely embraced by systems integrators who seek to exploit the potential these devices have to be small, inexpensive, robust, and scalable. In addition, MEMS-based switches can be constructed using well-established foundry practices. The key mechanical components are micro-machined mirrors that move in response to an electrical signal to route photons from one fiber-optic cable to another.
Current optical-switch architecturesMEMS devices are fabricated using processes similar to those used to make very-large-scale-integration (VLSI) complementary metal-oxide semiconductors (CMOS). Structural elements such as optical-switch micro-mirrors are constructed by selectively etching the oxide layers. The associated circuitry is created using conventional photolithographic techniques.
Commercial MEMS-based all-optical switches are currently designed based on one of two architectures. The 2-D (or N2) architecture uses a two-dimensional matrix consisting of N mirrors on each side arranged to connect N input fibers with N output fibers (see Figure 1). In the three-dimensional (3-D) or (2N) architecture, two groups of mirrors are used, each of which has N mirrors (see Figure 2).
In the 2-D approach, the only two options for mirror position are "on" or "off." 2-D switches are relatively simple to control and have lower power requirements. With the 3-D switches, the mirrors can be moved to at least N different positions. 3-D switches are highly scalable, with N capable of reaching into the thousands.
Actuation, surface phenomenaOne of the most challenging aspects in designing MEMS-based optical switches, or in fact any MEMS-based device, is that the physical forces that affect micro-scale objects are very different than those that govern conventional machines. At the micro-scale, atomic and surface forces such as electrostatics and friction play a far more important role than a body force such as gravity. Residual stress in the MEMS layer materials is also an important factor on the micro-scale. These seemingly subtle differences in micro-scale physical states are the key to designing the actuating mechanisms and related control systems that are responsible for precise movement of the optical-switch micro-mirrors.
Several methods can be used to actuate the mirrors in MEMS switches, but so far, electrostatic actuation is most commonly used, probably because it is most natural to the underlying physical structure of the device. Electrostatic actuation occurs when a voltage differential is applied across the mirror and an oppositely charged material layer. The mirror tilts or deflects in a controlled (and controllable) manner in response to increasing or decreasing voltage. N2 MEMS mirrors need only tilt on one axis to achieve an on or off position. 2N MEMS mirrors are constructed to tilt on two axes to achieve the required 3-D motion.
Although electrostatic actuation has always been assumed to require a relatively low power level, the amount of voltage required to cause the movement is actually directly related to the stiffness of the mirror material-the stiffer the material, the more voltage required. That's why some researchers are looking into mirror materials that are more flexible than silicon, which is quite stiff. Devices constructed out of these alternative, less stiff materials compose what has been termed compliant MEMS, or CMEMS.1
At Sandia Labs, researchers are working on an additional surface phenomenon that negatively affects electrostatically actuated MEMS device performance. "Stiction" refers to a permanent sticking phenomenon caused by residual absorption of de-ionized water used during wafer fabrication. Sandia is working on an alternative design that combines elements of CMEMS to create structures that operate with fewer rubbing surfaces and therefore less susceptibility to stiction.2
Proven technology
In addition to the problems of surface interferences, there is another significant stumbling block to the smooth and predictable operation of electrostatic actuators: the electrostatic force is inherently non-linear over the desired range of motion. Electromagnetic actuation is a viable alternative that has just recently begun to gain the interest of optical-switch manufacturers. However, electromagnetic actuation is actually a very well-proven technology in the data storage industry, having been used successfully in the design of optical pickup mechanisms for CD drives.
Electromagnetics can generate the forces required for mirror positioning in a highly linear and predictable manner, allowing for much more precise control. Performance is further enhanced because mirror hinges can be constructed of metal, eliminating tribological considerations such as stiction and friction. In spite of these advantages, electromagnetics ranks far behind electrostatics in commercial MEMS applications. It has been noted that to date, attempts at designing electromagnetic actuators have usually resulted in devices that required relatively high power. In addition, these mechanisms used magnetic materials that were not common in IC technology, necessitating some type of manual assembly.3
Another issue that arises with electromagnetic actuation is the need for shielding to eliminate crosstalk. Even though electromagnetic shielding has been successfully addressed in hard-disk drives, the MEMS community has remained unenthusiastic about electromagnetic actuation, preferring to concentrate their efforts on resolving the surface phenomena issues that trouble electrostatic actuators.
A new type of OXC design is now being developed that is based on the electromechanical technology that has already proven reliable in high-volume applications such as optical actuators for disk drives. The established optical actuator technology makes use of existing fabrication infrastructure, which is a distinct manufacturing advantage. The design features micro-mirrors mounted on frictionless springs, which results in precise mirror positioning that is highly linear in operation and almost entirely free of hysteresis caused by residual stresses.
Linear operation allows the use of a simpler control circuit and results in greater reliability, lower cost, and lower power consumption. The angulation of the bidirectionally rotatable mirrors is linear over a range of about ±10-degree rotation and is inherently self-aligning with a switching speed of approximately 5 msec. (In contrast, most commercially available electrostatically actuated switches have so far demonstrated switching angles in the ±2-degree range and switching speeds in the 10- to 20-msec range.) The linear servo control results in high pointing accuracy and very low insertion loss (<3 dB).
The design requires low-voltage (5-V), low-power driving circuits. The benefits of low-voltage, linear operation become more pronounced when large arrays of mirrors are used. It is estimated that 2N architectures that deploy electrostatic actuation require in the range of 150 to 250 V.4
$1-billion-plus market
A multitude of vendors have entered the MEMS optical-switch arena. Cahners In-Stat believes sales of the packaged mirror arrays for MEMS switches will surpass the $1-billion mark by 2004.5 The competitive advantage will go to the manufacturer who can produce the fastest, cheapest, most reliable, and most versatile switch. There are ample reasons to believe that the electromagnetically actuated switch will ultimately demonstrate enough significant cost and performance advantages to gain widespread acceptance.
Of course, there are many issues related to successful optical-switch performance that go beyond simply designing the optimum actuator. Other challenges include beam collimation, coupling efficiency, manufacturability, and the development of high-performance, hermetically sealed packages. Nevertheless, rapid advances in design and manufacturing promise to make the all-optical switch the gateway component to infinite bandwidth.
Dr. Chak Leung is vice president of operations at Blue Sky Research (San Jose, CA).
References
- M. Little and B. Jamison, Lightwave, July 2001.
- D. Williams, T. Picraux, and A. Ro mig, Jr., OE Magazine, 1:5, May 2001, pp. 27-29.
- M. Bourne, Lightwave, March 2001.
- Y. Ma and K. Timko, Lightwave, July 2001.
- M. Bourne, Lightwave, March 2001.