There's an insatiable demand for more telecommunications bandwidth, and MEMS is viewed as the key technology—the only technology according to many telecommunications experts—that will satisfy this demand. The trend is accelerating toward an all-optical network for moderate- and long-haul telecommunications networks, and eventually to local-access networks, using a variety of optical switching techniques including micromachined mirror arrays.
Much of today's telecommunication is performed in an optical-electronic-optical (O-E-O) fashion. Transmitted optical signals are converted to electronic form for processing and converted back to optical form at the receiver.
Taking an all-optical (O-O) approach can minimize, if not entirely eliminate, speed, bandwidth, and cost bottlenecks imposed by electronic circuits. Also, switching and routing costs can be reduced through integration. As one expert put it, "Our telecommunications networks are advancing at such a rapid pace that they're outstripping the ability of the electronic circuits to process the data that the networks are handling, and an all-optical network is the only foreseeable solution."
"Optical MEMS is one of the fastest-growing applications in MEMS, along with biomedical, chemical, and RF applications," says Roger Grace, president of Roger Grace Associates, a strategic marketing firm specializing in MEMS. Grace estimates the compound annual growth rate for MEMS in these areas as 20.1%, going from $14.1 billion this year to $36.2 billion in 2005.
One of the biggest advantages of an O-O network is that it doesn't depend on the signal format or protocol as an O-E-O network does. This makes it transparent to the signal format and protocol, as well as more reliable. Plus, it provides greater end-to-end scalability and flexibility.
Because the demand for greater bandwidth is mounting, customers may not be able to wait for the all-optical approach to mature. Nitish Mandal, manager of strategic markets for Tellium Inc., discussed this in a recent executive briefing on photonics sponsored by Information Gatekeepers. Mandal said that optical switches are great for transferring high-speed data, but not for processing that data. On the other hand, electronics are great for information processing, but limited on high-speed switching. Having a high-speed optical fabric on the inside of a network and smart optoelectronic interfaces on the outside might be better options.
Last year, Tellium announced its Aurora Full-Spectrum system, which combines all-optical cross-connect switches with electronic control and processing circuitry. An agreement signed this year between the company and Analog Devices allows the latter to manufacture a series of advanced MEMS chips for Tellium's Aurora line.
A Field Of Many
Dozens of startup companies have sprung up in the last couple of years, offering what each perceives as the right switching and routing solution to wideband communications. These companies were formed while mindful that larger enterprises like Nortel, Corning, JDS Uniphase, Cisco, Lucent, and Sycamore have purchased some of the smaller startups and companies specializing in optical MEMS. The startups also are ready to do battle in an increasingly crowded optical-switching marketplace. Even a company like Goodrich (formerly BF Goodrich), which has always been better known for making tires, has gotten into the act. It has aligned itself with Movaz Networks, a producer of optical transport systems, to make next-generation MEMS optical switches. Much of this interest began over the last several years when Lucent Technologies introduced its LambdaRouter, a 3D MEMS switch mirror array.
This interest in the industry is illustrated in studies revealing that about 50% of all data traffic today goes over distances of more than 300 km, distances that can greatly benefit from O-O networks. Today's networks have hundreds of wavelengths, with each wavelength operating at tens of gigabits/s. Switching and routing that many signals via an O-E-O approach is very difficult because several fiber pairs terminate into a telephone company's central office. Switching and routing such signals requires a large array of all-optical switches with thousands of ports. The rush to solve the bandwidth bottleneck has given rise to new terms of terabits and petabits. A terabit is a trillion bits (1012), and a petabit is a quadrillion bits (1015).
Presently, the big market is the long-haul backbones where the idea is to "set the standards in this market, and then cash in on the peripheral markets like metropolitan-area networks (MANs), which can eventually be a much bigger market," says Bob Solouf, director of business development for Analog Devices. Right now, the MAN market is still small.
The lure of an all-optical network for long-haul telecommunications using MEMS technology is seen in the adaptability advantage that such a network offers (Fig. 1). MEMS devices such as switches and tunable lasers will offer programmable cross-connects and add/drop multiplexers, things not available with present fixed-value components. This will catapult the development of dense wavelength-division multiplexing (DWDM) to the next level. DWDM is driving the demand for greater bandwidth.
One of the costliest components of current DWDM systems is the trans-ponder, which can be replaced by a MEMS tunable laser. Such a laser may be cheaper than the transponder because it can be mass produced using standard IC processing. Yet, producing it in smaller quantities might raise costs. Passive thin-film and arrayed waveguide gratings (AWGs) are presently used in DWDM systems. These are fixed elements that offer no flexibility. MEMS optical devices will change all of this with the ability to program them to adapt to different add/drop and network architecture needs.
Many see the tunable laser as the next hot optical MEMS device after switches. Tunable lasers are now becoming available, although they're rather costly. Given their use in large enough volumes, however, they can be produced less expensively. They would also cost much less than the transponders that they can replace, which have inhibited the growth of DWDM systems.
A System View
While many companies are working on optical switches, they're discovering that there's a lot more to an optical communications network than just a cross-connect switch. Other components such as add/drop multiplexers, variable optical attennuators, tunable filters, tunable lasers, and spectrum analyzers are equally important. Additionally, there are the difficult technical challenges of design simulations as well as packaging, testing, and processing these components into a usable subsystem. In short, how manufacturable a subsystem is becomes the critical factor to its success or failure.
"The real large-volume opportunities will be in the optical component areas, including add/drop multiplexers, line amplifiers, and filters," Grace claims.
When designing an optical switch, one must analyze and take into account many static and dynamic effects caused by several factors—including electrical, mechanical, and thermal—as well as the role of the package and the testing process. This gets more complex with larger cross-connect switches, and few companies have the software expertise to handle these tasks. Trace bits of moisture or dust on a large MEMS cross-connect switch with 64 by 64 ports or higher wreak havoc with the switch's reliability.
"Designing a complex mirror mechanism is only a part of the puzzle," Grace adds. "Interconnects, packaging, and testing will be major product differentiators, and a first-order determinant on who will emerge as a market victor."
Companies that do have a handle on designing a manufacturable MEMS component or subsystem are those that make the simulation software that takes into account all of the different phenomena that come into play. One such company, Coventor (formerly Microcosm Technologies), has been working closely with many of the optical MEMS firms, providing a suite of software tools and design services that cover all of an optical network's MEMS components, both individually and working together in a system.
"In many cases, we're adding IP to what the systems and switch designers are doing," explains Arthur Morris III, optical and RF product manager for Coventor.
Intellisense is another company in this arena. Recently purchased by Corning, it was the first company to come out with a software tool suite for anisotropic silicon etching, a critical step in MEMS manufacturing. Qualified to the ISO 9001 standard, Intellisense considers itself a full-service supplier to optical MEMS companies with expertise in all three segments of optical MEMS: design, development, and manufacturing. The company is focusing its efforts on working with key OEMs, and it is currently linking up all of the simulation software necessary for an entire MEMS manufacturing process.
"Our customers are looking for total solutions that take into consideration MEMS, optics, packaging, control, and software expertise," explains chief operating officer Doug Finke. "And, we've been working with optical MEMS OEMs, supplying all of them."
Some companies like Nanovation believe that the only practical way to make optical MEMS components and subsystems is by using an integrated optical (photonic) approach on a single substrate. They argue that such an approach allows for low-cost integration, packaging, and excellent performance with a moderate tradeoff of switching speeds.
These companies focus on optical waveguide integration implementing semiconductor technology."We have the three core technologies needed to do this: silica on silicon, indium-phosphide (InP), and MEMS," says Andy Mirza, Nanovation's MEMS technology director. "We're really a photonics IC manufacturer."
A Hybrid Process
Nanovation's hybrid process, using what the company terms "nanoshutters," closely mimics the characteristics of the fiber itself, minimizing insertion loss and maximizing fiber coupling to the chip (Fig. 2). Compared with conventional silicon micromachined mirrors that exhibit typical insertion losses of a few dB/km, the process provides less than 0.1-dB/km insertion losses. Moderate switching speeds are attained on the order of less than 10 ms, although such speeds aren't critical for many telecommunications applications.
"With our silica on silicon technique, we can easily make integratable splitters, taps, and separate DWDM signals—all at very low insertion losses," explains Gary Bjorklund, chief technology officer.
When making large core micromachined switch matrices of more than 16 by 16, the interconnect—and by extension, the packaging problem—be-comes extremely challenging. In fact, the packaging/interconnect problem can account for as much as 70% of a system's cost, which prompted Micro-Sound Systems to develop an entirely new approach. The company developed an advanced multilayer liquid-crystal polymer substrate employing UV laser microvia technology. It permits dense low-level interconnects on MEMS devices (level 0 being the lowest and least-complex level) with an aspect ratio of about 20:1.
"This kind of universal substrate easily satisfies the level 0 and 1 interconnect problems that MEMS producers are concerned about," claims Scott Corbett, MicroSound's vice president of R&D and co-founder.
But once you've designed, built, and packaged your MEMS optical circuit, it must be tested to ensure functionality. Testing has to be performed on both a prototype and a production basis. Can the manufactured device work properly, especially if it's mass-produced? Most optical MEMS companies have had to jury-rig their own test setups, often producing less than desirable results. There are no easy answers to this testing dilemma.
But one company, ETEC, believes it has the answer. Its entry into the MEMS test market wasn't so foreign for the company due to its 15 years of experience in the testing business. ETEC has now branched out into the optical MEMS test area with its M/SteP-o test and characterization systems of advanced optical MEMS devices and subsystems. This includes variable optical attennuators, mi-cromirror arrays, and optical cross-connects. Testing can be performed during any stage of the production process, including the wafer, die, and packaging stages.
"You'd be surprised at some of the testing problems we've solved for optical MEMS customers who had heretofore nowhere to turn," says Arthur Holzknecht, vice president of sales and marketing. The company was recently granted an exclusive license for worldwide distribution of the Network Probe Station product line of Umech Technologies LLC, an established leader in MEMS metrology and characterization systems.
What makes the testing problem very challenging is that many techniques are being used to make the core optical MEMS component, the cross-connect switch. In general, these techniques can be broken down into five major categories: movable free-space mirrors, liquid-crystal switches, guided-wave optics, thermo-optic switches, and nonlinear optical switches.
Free-space micromachined mirrors and guided-wave optics are the two most popular methods under pursuit today, with the former offering lower insertion losses and less signal degradation over distances, and, thus, less need to reamplify the signal. At the same time, the latter offers faster switching speeds and inherently lower cost.
Companies such as Lucent, Cronos Integrated Microsystems of JDS Uniphase, XROS of Nortel Networks Inc., Calient Networks, and Transparent Optical are using bulk silicon micromachining to make the mirrors. Others, including C Speed, Optical MicroMachines, Analog Devices, Tellium, and MEMX, are implementing conventional surface micromachining approaches.
Lucent has already been shipping 256-by-256 O-O network switch elements based on its LambdaRouter and expects to start shipping 1024-by-1024 switches this spring. Lucent claims to be the only company doing so commercially. Last June, Calient Networks announced a 256- by 256-port O-O switch called the DiamondWave system. Optical Micro Machines says that it was one of the first to ship 16-by-16 O-O 2D switches and has 32- by 32-port switches in development.
"We see a huge market in MANs for 2D switches," says Anis Hussain, founder of Optical Micro Machines. "In the 3D sector, reliability and robustness will be critical and difficult to solve."
Conrad Burke, vice president of sales and marketing, adds, "We've invested heavily in manufacturing automation, and alignment, testing, sealing, and other factors have been satisfactorily addressed by us. We even make our own automated collimators."
While significant progress has been made in surface micromachining of optical mirrors, the process becomes more challenging when depositing polysilicon films to make relatively large mirrors of a few millimeters. These surfaces tend to suffer from stress, which causes deformations that are difficult to control. It's important to achieve stress-free optical surfaces for optical networking applications. Just a few atoms in a film can change its stress dramatically. This is one reason why many companies are using the bulk micromachining approach.
As mentioned earlier, Nanovation is pursuing the optical waveguide ap-proach. Another firm, Axsun Technologies, is using LIGA (lithography galvanoformung abforemung, a German term for lithography for electroforming and molding) processing. With LIGA, highly parallel X-rays from a synchrotron are used to make 3D structures with very high aspect ratios of about 300:1.
Axsun sees itself as a highly integrated packaging technology house that can take customer chips, modules, and designs and integrate them all into a total hybrid optical network, thanks to its LIGA process. Axsun has been commissioned by the U.S. Department of Energy (DOE), the synchrotron's owner, to offer the first commercial synchrotron beamline for LIGA processing.
The company has already demonstrated its highly integrated optical packaging capability with an optical-channel monitor. Essentially, it's a spectrum analyzer—a key element for an all-optical network—that's ten times smaller than competitive products for only a fraction of their costs (Fig. 3). This 50-GHz device, only 2.5 by 1.7 in. (the size of a business card), has 25-ppm wavelength accuracy. It measures power versus wavelength, optical signal-to-noise ratio (OSNR), channel spacing, and absolute wavelength. Also, it covers the full C or L bands (1529 to 1565 nm and 1570 to 1610 nm, respectively) and continuously calibrates itself.
Agilent Technologies is using a hybrid process that combines CMOS with microfluidic micromachining steps. Based on the company's ink-jet printing technology, fluid-based switches are used. These consist of intersecting silica waveguides, with a trench etched diagonally at each point of intersection. Each trench contains an index-matching fluid that in the default mode allows light to be transmitted through it unimpeded. To switch the light path, bubbles are formed and removed hundreds of times per second by a thermal actuator. The bubbles reflect the light from the input waveguide to the output waveguide. This approach offers the advantage of no moving parts.
More details about switch architectures will be discussed in the second part of this report. Suffice it to say that in the micromachined mirror ap-proach, two main designs are being pursued: 2N configurations, where N input ports and N output ports require 2N controllable mirrors, and N2 configurations, where N input ports and N output ports require N2 controllable mirrors. It should be obvious that the former approach can be more easily scaled to larger port counts (say 1000 times 1000) than the latter approach. But the 2N approach is more difficult to control with larger port counts.
One thing is becoming increasingly clear. For any company to be successful in the optical MEMS market, it must rely on its experience in mass producing such devices cost effectively and reliably. Janusz Bryzek, a noted MEMS pioneer as well as president and CEO of Transparent Optical Inc., cautions, "The adaptation of the automotive MEMS experience is likely to improve the manufacturability and reliability of next-generation optical switches." Bryzek cites the hundreds of millions of automotive MEMS sensors presently used in an environment that's significantly harsher than the communications sector as an example where mastering the manufacturability of optical MEMS switches can be achieved.
That's one reason why Tellium has signed an agreement with Analog Devices to have the latter make MEMS micromirrors for Tellium. Analog Devices, one of the largest suppliers of automotive MEMS sensors, is leveraging its planar micromachining process in which it integrates signal-conditioning electronics on the same chip with optical micromirrors.
Thanks to its acquisition of BCO Technologies in Ireland, Analog Devices is implementing silicon-on-insulator (SOI) technology on its micromachined mirrors. The company feels confident enough about its manufacturing expertise that it supplies reference designs to any customer it works with. Customers can obtain prototype integrated versions of micromachined optical MEMS switches, complete with the signal-conditioning electronics.
Given all of the aforementioned technical challenges, there appears to be enough industry momentum and confidence that the right solutions should be achieved in the next couple of years. Will that happen? Stay tuned.
|Organizations That Contributed To This Report|
Agilent Technologies Corp.
Analog Devices Inc.
Axsun Technologies Inc.
C Speed Corp.
Calient Networks Inc.
Information Gatekeepers Inc.
JDS Uniphase (Cronos
Lucent Technologies Inc.
MicroSound Systems Inc.
Optical Micro Machines Inc.
Roger Grace Associates
Transparent Optical Inc.