Switching technology based on silica-on-silicon thermo-optic integrated lightwave circuits meets switching requirements for market segments such as small OADMs and crossconnects, and for protection and restoration.
All-optical switching is being developed and brought to market in various forms: microelectromechanical systems (MEMS), bubbles, liquid crystals, and planar lightwave circuits (PLCs), to name a few. Each of these is suited to a range of switching matrix sizes, typically ranging from a 4 x 4 switch up to the massively parallel switches envisioned at 1024 x 1024, or even 4000 x 4000. Planar lightwave circuitry, which is feasible in the range of a 1 x 2 up to 64 x 64 switch, has unique characteristics that suggest particular market-segment advantages.
The prevailing consensus for massively parallel switches—which are likely to be satisfied by MEMS technology—is that they are to be found most useful in the long-haul transmission business. But a significant function will be provided by smaller switches in supporting protection and restoration switching, especially in the event that there is partial loss of full switching functionality in larger switches. Thus, the small switch plays a role in front- and back-end protection and restoration.
Switching of wavelength bands, or grouped wavelength channels, between long-haul networks and regional or metro rings/meshes can also be supported by small to medium optical crossconnect switches. Similarly, switching between metro and access rings/meshes can be accomplished with moderate-size switches. Placement for switching applications seems most likely to grow rapidly in the regional and metro and access markets. Environmental roles for small optical switches can include optical frame switching (optical routers), optical circuit switching, optical add/drop multiplexers (OADM), optical crossconnects, and protection/restoration switching (see Fig. 1 and table, p.60).
Integrated optical planar-lightwave-circuit technology is mature: research in the field goes back to the late 1960s, and has explored materials ranging from ferroelectric single crystals to silica-on-silicon (SOS) to polymers. Silica-on-silicon is particularly mature, benefiting from extensive research investments and technology developments in silicon wafer microelectronic integrated circuits (ICs). The advantage of adapting silicon technology is the high expectation—and confidence—in processing and device reliability, low cost, and excellent performance. A number of devices are currently available or coming to market that are based on PLC technology, which will facilitate the realization of all-optical-network management—something that is necessary if DWDM is to meet expectations for traffic demands, cost, and quality of service.
Silica-on-silicon PLC technology enables M x N (M, N = 1, 2...) passive and active splitters and switches, variable optical attenuators (VOAs), and filters such as arrayed waveguide gratings (AWGs). An attractive advantage of the devices coming from SOS technology is that they lend themselves to integration of all-optical solutions. As an example, consider the combination of two AWG PLCs back-to-back, serving as a wavelength demultiplexer and multiplexer. Placing a multiport PLC switch in between results in an OADM subsystem that is entirely solid-state, integrated optics, and environmentally compatible from the packaging standpoint (see Fig. 2).
The basic unit of the thermo-optic planar lightwave switch is a 2 x 2-port Mach-Zehnder interferometer (MZI). The cladding layers above and below the waveguide cores provide mechanical support and optical isolation from the substrate and the upper metalization. The directional coupler is a passive device, whereas the MZI is an active 2 x 2 individual switch element (see Fig. 3). The MZI switching operation is based on the thermo-optic effect that provides analog phase-shifting, enabling variable-output power ratios. From this special property and a unique architecture, the variable-output power control and weighted multicasting are derived. The effect is based on the thermal difference between the reference and active legs, so that the switch is not affected by wafer or packaging temperature.
An input signal at one input port is split evenly by a passive directional coupler (see Fig. 4). One leg of the MZI is heated with a resistive strip (the "on" state), causing the effective optical length to change and shifting the phase of light in one path relative to the other. When recombined in the second coupler, the phase-shifting produces a change in the relative intensities in each output port. These can be varied in an analog fashion, which can be applied to produce weighted splitting and variable optical attenuation, subjects that will be discussed later. As a binary switch, the device is calibrated to produce maximum transmittance at one output when off (negligible heating power), and maximum transmittance at the other output when on (with heating power). The switching speed is determined by the thermal conductivity properties of the PLC and the ability to localize and then dissipate heat from the active leg of the MZI.
The directional-coupler spacing is just a few micrometers, so that waveguide coupling results in the diffusion of light energy from one guide to the other. The percentage crossover is strongly influenced by the difference in index between the core of the waveguide and the surrounding clad silica, by the waveguide dimensions, and the gap between the waveguides. Therefore, unless special care is taken, device performance can be critically dependent on process and photolithographic tolerances.
Historically, thermo-optic switching in silica-on-silicon suffered undesirable insertion loss (~10 dB), high crosstalk (-15 dB), and moderate switching times (~2 ms). By implementing advancements in architecture and resorting to a double-switch design, insertion losses in the 4- to 6-dB range have been demonstrated, as well as crosstalk rejection and extinction ratios of 55 to 75 dB. Switching time is still in the <2-ms range, but advances in switch design, including software control and processing methods, suggest that the switching speed can be reduced considerably, possibly to a few hundreds of microseconds.
The double-switch architecture provides variable output control (attenuation). It also allows for a splitting of the optical signal into multiple output channels—enabling multicasting, broadcasting, and access to the signal for optical-power monitoring (see Fig. 5). The VOA capability is relevant to signal equalization in the event that several output ports are recombined and passed through an optical amplifier. Such features are unique to PLC technology: weighting, multicasting, and VOA cannot be provided by the internal functions of MEMS, bubbles, mechanical, or LCD switches. External devices are required in these cases.
The analog basis of the PLC MZI enables this enhanced functionality within a single module. By controlling both the binary ("on-off") and analog operations, this optical switch design realizes a capability for intelligent switching functionality. As a part of a comprehensive routing and switching system, it enables optical crossconnects, wavelength routing, optical add/drop multiplexing, and protection/restoration switching. As an added benefit, the design of the switch is symmetric, meaning that if multicasting is possible and easy to implement, then so is multiplexing—the combining of different wavelength channels on the same output fiber.
Contributors to this article include Nahum Izhaky, Reuven Duer, Neil Berns, Eran Tal, Shirly Vinikman, and Yosi Shani, all at Lynx Photonic Networks Ltd., 13 Hamelaha St., Afek Ind. Park, Rosh HaAyin, 48091, Israel.
Jeff Schoenwald is chief physicist, US operations, at Lynx Photonic Networks Inc., 26775 Malibu Hills Road, Calabasas Hills, CA 91301, 818 226-4050; he can be reached at firstname.lastname@example.org.
SWITCH MATRIX REQUIREMENTS
The generic requirements for optical switch performance are driven by quality of service and practical operational and environmental requirements in the field.
Small size. The preferred configuration is for a switch that fits on a printed circuit-board card, which then fits into a slot. The typical card-slot separation is 1.0 in., so that device heights of less than 0.75 in. are imperative. In the case of small-to-medium-sized switches, several switches may fit on a single card. In the case of large switch matrices, multiple card slots may be occupied, or an entire cabinet devoted to the switch matrix.
Matrix size/port count. The size of the matrix depends on the application, whether long-haul, regional, metro, or access.
Non-blocking. A truly non-blocking switch passes optical transmissions with no delay, and is not blocked while other input/output ports are reconfiguring. In conventional optical-electrical-optical systems, the process of detection, regeneration, and retransmission requires delay and often buffering, which leads to undesirable delays.
Scalability. It is desirable for the technological choice to be amenable to architectural expansion; i.e., that the switch size be readily expandable in fabrication, and/or the switch module be configurable to expand total switch-matrix size.
Reliability. The switch technology must meet requirements for operational storage and transport conditions and be robust for a lifetime in excess of ten years. Key environmental requirements are specified by Bellcore/Telcordia standards and recommendations, including performance under varying conditions of humidity, vibration, temperature, shock, and cycling of environmental parameters.
Multicasting and signal-output weighting. Some system requirements call for signal splitting for multiple endpoint users. Not all switch technologies are capable of this function, and require additional external splitters. In addition, the signal level for each user may require different power levels to compensate for transmission distance, or unequal splitting may be advantageous to provide for optical power monitoring. Incorporating these functions is a decided advantage in system flexibility while providing economy of space.
Low power consumption. Standard rack-and-slot cabinet installations require budgeting for heat generation in order to design within thermal "heat footprint" limitations—a serious concern in large switching-center facilities. Most all-optical switches have modest power-consumption ratings.
Low insertion loss. Insertion loss consists of any splitter loss plus excess internal (plus connector) loss. In an N x N switch, any one-to-one connection should consist only of excess loss; a 1 x 2 50/50 connection should show an additional 3-dB loss in each channel. An ideal switch would have 0-dB loss, but connector or splice losses ensure that loss is at least 0.2 to 0.4 dB, and each technology has some internal limitations. Typical minimum insertion loss for N x N switches ranges from 3 to 6 dB.
Low crosstalk. It is important to block leakage of signals from one switch path to any undesignated output port. System users are specifying -40- to -50-dB crosstalk suppression.
Low polarization-dependent loss (PDL). Variations in insertion loss due to variations in the input polarization of the signal must be kept to a minimum so as not to discriminate in switching efficiency. Polarization-dependent loss is typically specified by system users to be below 0.4 to 1.0 dB, depending on application.
Flat passband (wavelength-dependent loss; WDL). A switch should impose no penalty on signals of different wavelengths; i.e., it should be colorblind. Wavelength-dependent loss is the additional variation of loss across the bandwidth specified for the switch. Typical WDL specifications call for a maximum of 0.4 to 0.5 dB for C- or L-band, up to perhaps 1 dB for combined C- and L-band operation.
Low return loss. Back-reflections can cause instabilities in the laser source, adding unacceptable noise on the transmission line, as well as echoes propagating undesirably back through the network. Typically, most back-reflection is suppressed by angle polishing fibers or connectors at the input/output ports of devices, which provide better than -55-dB connector return loss; but care must be taken to design for minimal internal reflections as well. Total return-loss backward-reflection is often specified to be less than -40 dB, and sometimes -50 dB.
Switching time. Most optical-crossbar circuit-switching times are specified between 5 and 10 ms. Since some circuit restoration requirements, like SONET, require full restoration within 50 ms, the more margin that a switch can provide for control overhead, the better. Another consideration is the overshoot and settling time that may occur with some switching technologies. These times must be included in the total switching time, usually quoted as the interval to transition between 10% and 90% of the "on" steady-state signal.
Low cost. Cost is generally proportional to the number of physical ports—and therefore the number of physical channels it can handle. Costs are typically quoted as a cost per port. Thus, an 8 x 8 switch has 16 ports.
Environmental test certification. Generally, all active and passive components must meet relevant standards and requirements for reliability and performance according to Bellcore/Telcordia and IEC standards.
As optical-switch matrix technologies continue to rapidly advance with greater capacities and faster speeds, they all must meet the practical standards for deployment outlined.