In very short order, several types of MEMS all-optical switches have been developed or have become commercially available. Rapid advances and potential benefits for all-optical networking will create a mems market of more than $1 billion by 2004.
As the demand for communications bandwidth accelerates, there is an increased desire to develop an all-optical network. An important part of the equation is the replacement of current switches that must convert optical signals into electrical signals and then back into optical form for further transmission. The gains of avoiding signal conversion could not only significantly reduce costs but also seamlessly move larger amounts of data, voice, and video at higher speeds.
A unique class of switches based on a relatively new technology that can avoid such signal conversion has been generating considerable excitement recently. This all-optical technology is based on microelectromechanical systems (MEMS). MEMS are essentially micrometer-sized devices with three-dimensional (3-D) properties-such as sensors, pumps, and valves-that can sense and manipulate physical parameters.
Created by various micromachining processes, the end result is typically a moveable mechanical structure, most often on a silicon substrate. However, some MEMS have no moving parts, and can be made out of plastic, glass, and even diamond.
MEMS were first commercialized in the early 1990s and can now be found in all kinds of medical, transportation, industrial, and consumer applications. Growth is primarily due to the extremely small size of MEMS, which allows them to be capable of faster, more precise operation than their macroscopic counterparts.
To date, there have only been a few high-volume, high-visibility end-uses, including accelerometers, pressure sensors, and nozzles. Many observers believe the next such application for MEMS is optical switching.
There are two types of optical MEMS switch architectures under development or available-mechanical and microfluidic. Mechanical switches use an array of micromachined mirrors that can range into the hundreds of thousands-all on a single chip. Microfluidic-based switches, on the other hand, have no moving parts. Rather, they rely on the movement of liquid in micromachined channels.
A mirror-based switch incorporates a matrix of micromachined mirrors that are typically fabricated in single-crystal silicon. These mirrors allow for optical connections between input- and output-fiber arrays. Control signals applied to the MEMS chip fix the position of each individual mirror to direct an incoming light signal to the desired output port. Tight control of the mirror angles minimizes the optical-power loss incurred from passing through the switch (see Fig. 1).
Mirror-based switches can be classified according to mirror movement. In two-dimensional (2-D) devices, the mirrors are only able to execute a two-position operation-that is, the mirrors can move either up and down or side to side.
For example, Cronos Integrated Microsystems (part of JDS Uniphase and located in Research Triangle Park, NC) manufactures a 2 × 2 switch that uses a sliding mirror along with a latchable actuator. The mirror is oriented perpendicular to the optical plane and can be moved laterally across the substrate. Latching is incorporated so that when power is removed, the mirror remains in its most recent position.
In 3-D switches, the mirrors can assume a variety of positions by swiveling in multiple angles and directions. Lucent Technologies` (Murray Hill, NJ) WaveStar LambdaRouter is based on MicroStar technology developed at Bell Labs. Its 256 × 256 crossconnect is composed of two-axis (3-D) micromirrors, a fiber/lens array, and a fold mirror that helps direct light from input port to output port (see Fig. 2).
While most major companies are taking the micromirror-based approach, Agilent Technologies (Palo Alto, CA) is taking advantage of its microfluidics knowledge and has developed a switch based on Hewlett-Packard`s ink-jet printing technology. What differentiates this switch most from micromirror-based switches is that it has no moving parts.
The fluid-based switch consists of intersecting silica waveguides, with a trench etched diagonally at each point of intersection. The trenches contain an index-matching fluid that, in default mode, allows transmission of light unimpeded through the switch. To switch the light path, bubbles are formed and removed hundreds of times per second in the fluid using a thermal actuator. The bubbles thus reflect the light from the input waveguide to the output waveguide.
There are three basic means of fabricating MEMS devices: surface micromachining, bulk micromachining, and LIGA. Surface micromachining involves building up various layers of materials that are selectively left behind or removed by subsequent processing. As such, the bulk of the substrate remains essentially untouched.
With bulk micromachining, a large portion of the substrate is removed to form whatever structure is desired. Therefore, structures with greater heights are easier to fabricate because thicker substrates can be used.
LIGA (lithography, plating, and molding) combines the basic process of integrated- circuit lithography with electroplating and molding to obtain depth. In LIGA, patterns are created in a substrate, then electroplated to create 3-D molds.
While these molds can be used as a final product, a variety of materials can also be injected into them to generate a product. This technique has two distinct advantages: many different materials can be used (in particular, metals and plastics), and very high aspect-ratio structures can be built.
Most MEMS switch manufacturers are relying on bulk micromachining methods. Agilent Technologies, however, is using a hybrid process called CMOS MEMS, which combines conventional complementary metal-oxide semiconductor (CMOS) processing with several micromachining steps. Axsun Technologies (Billerica, MA) is the only company so far that has developed a switch fabricated using the LIGA method.
The majority of MEMS switches currently support 32 or fewer bidirectional ports, although companies are working hard to develop devices that can support more than 1000 ports. The big question is whether a large matrix array is better than scaling smaller switch modules together.
Right now, manufacturers are using switch modules smaller than 64 × 64 as building blocks, essentially scaling them to form a large switch. For example, one could construct a fully nonblocking crossconnect supporting up to 1024 × 1024 ports by linking together 32 × 32 or 64 × 64 switch modules.
Supporters of this approach say that the scaling of smaller switch modules provides flexibility, meaning that customers can add to a device as they need extra capacity. Under such circumstances, if part of the switch fails, only a small area is lost, as opposed to a catastrophic failure with a larger switch fabric. Furthermore, a small module is easily replaced.
Others do not consider the scaling approach an efficient solution. They believe the use of one large matrix array (such as 1000 × 1000 switch) is preferable because scaling switches results in power loss. However, it would take a considerable amount of time and difficulty to replace such a switch in the event of a failure.
MEMS switches have demonstrated the ability to switch any type of data, at any protocol (including synchronous optical network/synchronous digital hierarchy [SONET/SDH], asynchronous transfer mode [ATM], Internet Protocol [IP], and Ethernet), and at any speed or wavelength. Nearly three-dozen companies are on the scene actively seeking entry into what is perceived to be a highly lucrative market. This number has increased from only a half-dozen companies involved in producing optical MEMS switches three or four years ago.
The road to commercialization will not be easy. Initial field trials, however, have shown promise to such an extent that many observers are speculating that MEMS is the right tech- nology, in the right place, at the right time.
As a result, we estimate the total market for all MEMS was slightly over $3.5 billion in 2000 and will double in four years. Of that $7.1-billion market in 2004, MEMS-based all-optical switches will account for more than $1 billion.
Marlene Bourne is a senior analyst with Cahners In-Stat Group, 6909 E. Greenway Parkway, Suite 250, Scottsdale, AZ 85254. She can be reached at 480-609-4541; e-mail: email@example.com.
FIGURE 1. A 2-D, 8 × 8 crossconnect shows the typical layout of a micromirror array.
FIGURE 2. A mirror in a 3-D switch demonstrates its ability to swivel in multiple directions.