All-Optical Switching Tutorial, Part 2
Micro-electro-mechanical systems (MEMS) are in widespread use in some other industries, but their use for telecom applications is relatively recent.In telecom, MEMS has become synonymous with the arrays of tiny tilting mirrors used for optical switching fabric, although the same technology is being used to make a wide range of other components as well (see, for instance, LightConnect Comes Into Bloom, Cypress Flexes MEMS Muscles, and Iolon Unveils Tunable Laser).
Since MEMS creates so many mirrors on a single chip, the cost per switching element is relatively low. However, since it involves moving parts, MEMS is fairly slow to switch – requiring milliseconds to do so. This is fine for lambda provisioning or restoration but is too slow for optical burst switching or optical packet switching applications.
Conventional MEMS works by reflecting the beam of light from the surface of a tiny mirror. MEMS systems have moving parts, and the speed at which the mirror moves is limited. By applying more current, the mirror can move faster, but there's a limit to how much current can be sent into the array of mirrors. If this weren't bad enough, it seems that the speed and angular displacement terms in the calculation of the required current have integer powers of around 4 or 5, and so the bottom line is that we have to put a lot of current into the array for a small improvement in speed. By changing the mirror design so that the angle through which light is bent is smaller, it's possible to achieve faster switching speeds. This technique is known as "fast MEMS."
MEMS arrays can be built on a single-chip, single-plane approach. In other words they are 2 dimensional (2D MEMS). In a simplistic approach it’s also possible to stack a number of 2D MEMS arrays on top of each other to create a 3D MEMS array. In fact, real 3D MEMS systems are somewhat more complex than this, but the general principle holds.
A huge drawback of 3D MEMS is the fact that the thousands of mirrors require complex software to coordinate their operations. In particular, one vendor has suggested that there are over a million lines of code in their implementation (although the reference may be to the overall switch software, and not just the MEMS subsystem). While it’s possible to test software extensively, the opportunity for bugs increases geometrically with the size of the code base.
On the upside, MEMS is a very rapidly changing technology. Since it seems to have a monopoly on the high port-count optical switch market for the moment, a huge amount of investment is going into the implementations and into solving the basic problems.
Further reading:
Next Page: Frustrated Reflections
