With so much global effort being directed into the development of optical crossconnect switching fabrics, why is the holy grail of optically transparent interconnectivity proving to be so elusive?
Mark Zdeblick
Many companies are developing optical crossconnect switches. All of these firms are challenged by two parallel universes:
null
Despite decades of development, no single technology, architecture, or design implementation has emerged as a winning industry standard for optical-switching fabrics for telecommunications optical crossconnects (OXCs). Part of the reason for this is the self-contradictory nature of design goals, in which one design variable limits optimization of another. These contradictory goals necessitate design trade-offs that must balance cost and functionality.
While it is practically impossible to define all switching technology using only a handful of variables, most well-known design approaches can be characterized with consideration to the following variables: mode size, fiber interface, integrated actuation, control system complexity, and the number of switching elements as a function of port count. Designs vary, depending on which variable is to be optimized.
Integrated optic or bulk optic?
Many designers believe that the lowest cost implementation of any complex optical component is within integrated optics, including components such as arrayed waveguides, switch arrays, attenuator arrays, and modulators. Furthermore, many integrated-waveguide technology platforms are now considered mature. Foundries abound, offering waveguide capabilities in silica, polymer, glass and III-V semiconductors. Standard single-mode waveguides perform superbly to confine and direct light, combining multiple functions on one chip. They facilitate easy fiber-alignment and attachment through well-established V-groove techniques. However, discontinuities in waveguides, such as air- or liquid-filled gaps, create considerable losses per switching element that can accumulate with port count to unacceptable levels. One way to minimize losses is to use expanded-mode-size, single-mode waveguides, but these are much harder to fabricate than standard waveguides because of the lower cladding-to-core index ratio and the control thereof in a manufacturing environment.
The primary benefit of waveguides is that they reliably stabilize the light path with respect to the switching elements and input/output ports, even during shock, vibrations, and other environmental influences. Free-space systems, on the other hand, do not fix the light path as successfully, and tolerance stack-up is a significant challenge to overcome.
The use of V-grooves for alignment limits the quality of butt-connections between fiber and switch-fabric waveguide. Collimator-based fiber alignment is the other extreme, used most notably across free-space interconnections. Collimators are expensive and relatively difficult to align, especially in array formats. For example, two-dimensional (2-D) and three-dimensional (3-D) microelectromechanical-systems (MEMS) switches use collimators, while the well-publicized bubble-induced Fourier transform infrared (FTIR) switches use V-groove alignment. Direct fiber attachment using V-grooves is probably the greatest advantage of the latter technology in terms of manufacturing (see Fig. 1).
All switches need an actuation mechanism, whether it's a rotating or flipping mirror, a bubble-induced FTIR, a voltage-induced phase-change, a current-induced refractive index change, or one of a myriad of other emerging and far-flung techniques. An actuator integrated within the switch fabric offers significant advantages. Integrating the actuator with the object to be controlled is not only a fundamental axiom of control theory but often produces significant yield, reliability and cost benefits. On the other hand, integrating the actuator with the switch may limit the performance of the switch in other ways, such as stroke if displacement is involved, or angle if rotation is necessary. Separating the actuator from the optical element, as it is done in a CD drive, increases complexity and part count but is sometimes necessary to get the performance that is required.
Control and system management
While the attention of the optical components community remains fixed on the optical principles underlying fiberoptic switches, the enormity of the electronics design, control, monitoring, and management problems should not be underestimated. The simplest electronics solution consists of a microcontroller for communicating with the next layer of the network, with digital inputs directly controlling the switches and no other drivers in between. Closed-loop requirements for 2N designs (where N represents the number of switching elements), such as temperature control or pointing control, add complexity and cost. Such complexity can be very significant, particularly for N2 architectures, which can require thousands of closed-loop control circuitsone per switching element. This complexity adds to development time and impacts manufacturability. Nevertheless, most optical switches require some form of closed-loop control to get the required performance; those that don't have a significant cost, reliability, and manufacturability advantage.
Finally, minimizing the number of switching elements accrues two key benefits. First, there are fewer switches to control and fewer to fail. Second, the maximum allowable insertion loss per switching element is substantially higher as the number of switching elements decreases. For example, a typical 32 x 32 switch that claims a 7-dB insertion loss requires an average loss per switching element of 0.11 dB, ignoring connection losses. In contrast, the same switch built with a 2N architecture could withstand up to 3.5-dB maximum allowable insertion loss per switching element and produce the same 7-dB overall insertion loss.
A similar argument may be made for crosstalk. Crosstalk refers to light from one channel coupling into another channel. For a 1 x 2 switch this would be the amount of light going into the wrong dark channel. For an 8 x 8 switch the aggregate crosstalk is the amount of light from seven channels that go into the eighth. Ideally there is zero crosstalklight only goes where it is supposed to go. A crosstalk of -60 dB implies 1 µW of light goes down the "dark" output for every 1 W of light into the illuminated channel. For example, in an FTIR-based N2 architecture, each channel contributes to an aggregate crosstalk figure. If there are 32 channels, 31 switch elements could contribute crosstalk to a single output. In that case, to meet a maximum crosstalk specification of -50 dB, each switch element must not have more than roughly -65 dB of crosstalk, not an easy target for an FTIR approach.
FIGURE 2. Two popular architectures, 2N and N2, illustrate the number of switches as a function of port count.
Some N2 architectures, commonly referred to as MEMS 2-D, in which the mirrors flip up in order to cause switching, do not obey the above scaling limitations. The reason for this is that the number of interfaces that the light goes through does not scale as 2N-1; there is, in fact, only one reflection per port and no interfaces to traverse. Nevertheless, a 32 x 32 MEMS 21-D switch still has 1024 switches, compared to only 64 for the 2N architecture, which, for equivalent yield and survivability statistics, is a significant disadvantage (see Fig. 2). For 1000 x 1000 switches, the 2N architecture is a tremendous advantage (2000 versus 1 million switch elements). At this scale, each switch element in the N2 architecture needs to have the performance and manufacturability of complementary metal-oxide semiconductors (CMOS) in order to be viable. In fact, if one thinks of leakage currents through a CMOS transistor, such as Fairchild's 2N7002, as analogous to crosstalk in an optical switch, the performance of a typical CMOS device (-64 dB) is not nearly good enough for a 1000 x 1000 N2 total-internal-reflection architecture (-80 dB per switch element required).
Switching elements totaling twice the number of in/out ports seem to be among the most efficient for large-port-count switches. However this 2N architecture conflicts with most of the other variables: it is free-space instead of waveguided, uses collimators instead of V-grooves, and tight closed-loop control instead of on/off. Such difficulties explain the long development times, relatively high costs, and repeated qualification cycles being experienced by many developers of that technology (see table). From the perspective of cost minimization, ease of manufacturing, and scalability, the ideal switch fabric might have the following characteristics: standard waveguides, twice the number of switching elements as ports, and an integrated actuator that requires no feedback control and is aligned to fibers using V-grooves.
MARK ZDEBLICK is chief technical officer of the switch program at K2 Optronics, 1288 Hammerwood Avenue, Sunnyvale, CA 94089. Tel: (408) 747-5915; e-mail: [email protected].