How to Lose $5B in Optics, or "What do I do with my MEMS Investment?"
Timothy Cahall, Trellis Photonics
Almost $5B has been invested in MEMS (Micro Electrical Mechanical Systems) technology for the telecommunications network. Unfortunately, the vast majority, if not all, of this money, will be lost within the next few years. While a novel science experiment, MEMS is fatally flawed as an optical switching mechanism and will be relegated to bandaging together legacy systems in order to minimize the loss on the money already invested. The true tragedy of this saga lies, not in the enormous amounts of money which are speeding their way to money heaven, or the additional sums which must be pushed down the rat hole in the desperate hope of luring some of the already lost investment back from money heaven. The true tragedy lies in the delay and disillusionment that this waste will cause in achieving the all-optical network.
Long distance network providers are caught in a vicious cycle. They must constantly invest in higher speed technologies in order to compete in a merciless marketplace determined to commoditize their product. As they invest in new technology to derive greater and greater speed from their existing infrastructure, they are forced to scrap or re-deploy equipment which was installed a few years earlier. This causes operators to either shorten their depreciation schedules or take significant one-time charges for decommissioning equipment. While painful, these costs are much less expensive than allowing the market to drive them to irrelevance.
Enter the all-optical network. The all-optical network offers service
providers with hope. It is protocol and speed independent, easily
upgradeable and reduces the cost to upgrade by an order of magnitude. In addition, the time to upgrade to higher network speeds or deploy new services is dramatically reduced. While seemingly too good to be true, the economic and market advantages of an all-optical network are real and easily understood. Unfortunately, it is precisely this clarity of advantage that allowed an unworkable technology, MEMS, to draw enormous investment.
MEMS are made up of tiny moving parts that allow light to be switched
between fibers. This technology is most commonly implemented with
individual wavelengths of light being introduced to the fabric, reflecting off of a mirror and being placed inside a different fiber. This can be done in either two dimensions (2D MEMS) or three dimensions (3D MEMS). While highly intelligent and overcomplex, these systems suffer from fatal flaws. While the fatal flaws most commonly sited are the difficulty in aligning, operating, calibrating and transporting these devices, all of these will likely be overcome(at great expense). Unfortunately, these flaws are merely superficial. The fatal flaws which will prevent a MEMS solution from ever being implemented in an all optical network are its requirement for extensive pre and post processing and its inability to monitor, manage, discipline or power balance wavelengths. The mechanical issues confronting a MEMS installation, while complex, are solvable. The issues confronting MEMS in a real world network are not.
MEMS, by definition, are single wavelength devices. While they are capable of switching the entire spectrum of light between fibers, this capability is hardly worth the expense and complexity a MEMS represents. In order to implement a MEMS in a DWDM (Dense Wave Division Multiplexing) environment, the signal must first be completely de-multiplexed, or pre-processed, to its constituent parts. It is then passed through the MEMS fabric where it must
be re-multiplexed and post processed, into fibers on the DWDM network. This pre and post processing equipment is expensive, bulky and induces unnecessary losses into the network. In addition, doubling the size of the MEMS switch doubles the size of the problem. As MEMS vendors strive to achieve extremely large fabric size, the amount of pre and post processing required to support a single fabric will become larger at the same rate. Clearly, this is a less than ideal situation.
MEMS vendors pride themselves on achieving relatively low loss through their switching fabric. While their progress on this front has been impressive, albeit not unique, their systems are opaque from a network monitoring and management standpoint. While a MEMS system is passing a signal through its fabric with a minimum amount of loss, it is completely unable to see if the signal is good or bad. Accordingly, an installation consisting of multiple MEMS systems cascaded together cannot be diagnosed from a central location.
In order to diagnose problems within a MEMS network, technicians will have to be dispatched to each of the switching sites, interrupt traffic and test to locate the signal problem. It is highly unlikely that network operators will tolerate this return to 19th century trouble shooting technique. The inability of network operators to manage MEMS switching fabrics form a network management viewpoint will prevent them from ever being deployed in volume. A Director of Optical Engineering at a major service provider was recently asked what his dream all-optical switch would look like. His answer was very telling. He said, "I could never get past the fact that
these things couldn`t be managed in the network, so I never bothered to dream". A technology that is so rife with shortcomings that the engineering department doesn`t bother to dream is not going to generate significant revenues in the near term.
If all optical networks are to change the world of networking, they must transition our network paradigm from a fiber-based paradigm to a
wavelength-based paradigm. Unfortunately, MEMS switches cannot perform in this space either. As fundamental a shortcoming as the inability of a MEMS fabric to be managed in a real world network is, the inability of a MEMS fabric to attenuate or discipline individual wavelengths is equally substantial.
Wavelengths of light within a fiber interact. This is a very important feature of optical fiber as it allows light to amplify light within a fiber. This is a fundamental building block of the DWDM networks in deployment today and the future of an all-optical network. However, this interaction of wavelengths within a fiber has a dark side because, unless very carefully balanced, the misbehavior of one wavelength of within a fiber will cause all of the wavelengths within a fiber to go off line. In order to move to a wavelength based paradigm, customers must be assured that the misbehavior of
one wavelength of light within a fiber will not cause any disruption on the other wavelengths within the fiber. MEMS, by their nature, are either on or off. There is no capability to be partially on or partially off. This inability means that any wavelength that passes through a MEMS fabric must pass through at full power. The complete inability of a MEMS fabric to manage the power level of individual wavelengths means that the misbehavior of a single wavelength in a network will disable all of the wavelengths in every fiber that the signal passes through. In a world where a single wavelength is switched between different fibers as it traverses a network, it is possible that a single wavelength operating at too high a power can,
segment by segment, cause a blackout it large portions of the internet. Clearly, this cannot be allowed to happen. Accordingly, wavelength based services will not happen with a MEMS fabric at the heart of the network.
The misbehavior of a single wavelength within a network causing widespread outages is clearly unacceptable. However, the inability of a MEMS fabric to manage the power of wavelengths places network operators in a position that their own network could take itself off line.
Extensive work has been done to ensure that EDFAs (Erbium-Doped Fiber
Amplifiers) amplify light evenly across a relatively wide band. However, EDFAs add the same amount of light to each wavelength regardless of its launch power. If one wavelength is stronger than the rest of the wavelengths, it can, after several stages of amplification, move above its power limit and cause an outage in the network. Accordingly, wavelengths in a fiber must be of approximately the same power level to pass through the EDFA amplifiers in a network.
While it is critical that all of the wavelengths within a network have the same power level, it is a certitude that, in any wavelength switched network, they will not have the same power. The assumption behind a wavelength switched network is that wavelengths from different fibers will be cross connected, or blended, across several output fibers. This means that different wavelengths within the same fiber will have come from different source and be different distances from the switch. Accordingly, wavelengths in an optical switched network will arrive at the switch fabric with dramatically different power levels. As a MEMS fabric is unable to adjust the power of individual wavelengths, a MEMS switch will switch multiple wavelengths, of dramatically different power levels, to the same
fiber, thus causing an outage in the overall network.
The complete inability of a MEMS fabric to be deployed in an all-optical network has delayed the start of deployment of optical switching solutions. However, service providers have found a short-term solution that addresses the problems with MEMS. This solution is commonly known as OEO (Optical Electrical Optical) switches.
OEO switches are essentially the digital cross connects of yesteryear with optical connectors and much higher throughput rates. These systems, due to their electrical nature, are manageable, practical and deployable in a real world network. Unfortunately, they have all of the same impediments of current DWDM systems in that they are speed and protocol dependent. OEO switches will not help service providers break the cycle of throwing away equipment every two to three years. In addition, OEO switches are severely limited in their overall throughput speed by their electrical backplanes.
Clearly, MEMS switching fabrics will not be part of the future all optical network. Unfortunately, an enormous amount of money has been spent on these science projects. Is there anyway to save some portion of that invested capital from a slow death and a trip to money heaven? Tarzan to the rescue!!!
In order to break through the inherent speed limitations of an OEO switching fabric, Tellium has announced that they are moving to O-E-O-E-O. If you say it fast, you can hear Tarzan swinging through the trees. In this solution, complicated though it may be, a MEMS fabric is placed in the heart of what is essentially two OEO switching fabrics. The placement of a MEMS in this position is possible because all of the management and operational issues created by the MEMS can be hidden under the cloak of the electronics. The
electronics benefit in that they are no longer limited by the speed of the electrical backplane, thus enabling larger, non-blocking cross connects.
While complex, this solution will likely work as an interim solution until credible, manageable optical switches reach the market. Unfortunately, this solution will have twice as many lasers and receptors in addition to a MEMS fabric. In short, O-E-O-E-O systems are going to be big, take a lot of power and generate an enormous amount of heat. Fortunately, the cost of the generator and the air conditioner are trivial when compared to the acquisition cost of the switch itself.
The incredible amounts of money which have been invested in an unworkable technology, MEMS, personify the bubble that is growing in the optical space. While these science projects continue to send millions of dollars to money heaven, network operators remain mired in a vicious circle of investments, write downs and reinvestment in continual pursuit of a lower cost per bit. Only by focusing investment on technologies that are deployable in a real world network can the industry ever reach a new level of performance, cost and service.
Over USD $5 billion has been sent to money heaven by the MEMS companies. This tragedy, played out in the popular press, has wasted an enormous amount of money and has distracted the industry from pursuing credible solutions to real world problems. Of greater danger is the coming crash when investors wake up to realize that they have lost $5 billion dollars. When this realization hits them, as it must, the burst in the bubble will wash across the entire optics industry, impacting real and imaginary companies alike.
The real risk to the industry is that photonic switching will become as popular with investors as dot coms are today. If this occurs, we could easily find the implementation of an all-optical network delayed by five years or more. Five billion dollars is a lot of money to lose. However, it pales in comparison to the loss of five years of progress.
Timothy Cahall is the CEO of Trellis Photonics, manufacturer of the Intelligent Lambda Switch®, the industry`s first truly all-optical switching fabric.
216.87.28.62 |
