Optical circuit switching may be instrumental in meeting the cost, energy, and aggregate bandwidth requirements of future data center networks. To be of practical interest, optical switches must deliver the switching speed, port count, and optical transmission required for data centers. These requirements differ from the telecommunications provisioning and protection requirements that have guided the design of conventional optical switches. Our current work takes a comprehensive approach to design optical switches and networks from the ground up to meet the needs of data centers.
The starting point of this work was an assessment of the scaling properties of conventional MEMS cross-connects. These switches have been fabricated with over 1,000 ports and less than 4 dB insertion loss, significantly outperforming guided-wave switches built on III-V and silicon platforms. However, large port count MEMS switches have switching speeds on the order of 10’s to 100’s of milliseconds, much too slow to provision bandwidth for the quickly varying traffic patterns of many data center applications. Using physical optics and electromechanics, we conducted a first-principles analysis of how the response speeds of a set of canonical MEMS tilt-mirror switching devices scale as a function of switch port count, crosstalk, and insertion loss. Our model indicated that the optimal actuator design (parallel plate vs. vertically offset comb) and actuation method (digital vs. analog) changes as a function of switch port count. It also suggested that conventional cross-connect switch architectures do not allow a favorable tradeoff between switching speed and optical transmission or crosstalk. Further, without significant reductions in manufacturing tolerances, increases in electrostatic drive voltage, and/or smaller feature lithographic feature sizes, the speed of conventional cross-connects cannot be substantially improved. Faster switching speeds can be achieved by multistage switch architectures at the expense of higher insertion loss.
Given the scaling limitations of single- and multi-stage cross-connects, we are investigating a novel non-crossbar selector switch architecture and pupil-division switching layout which improve optical switching performance by relaxing the requirement of arbitrary switch configurability. This architecture and switch design enable MEMS beam-steering micromirrors to scale to microsecond response speeds while supporting high port count and low loss switching, and can be used to realize a number of useful interconnection topologies. We have designed, fabricated, and experimentally characterized a proof-of-principle prototype which uses a single comb-driven MEMS mirror to achieve 150 μs switching of 61 ports between 4 pre-programmed interconnection mappings.
Continued work addresses switch scalability, network design, and closed loop control.
Conceptual illustrations of MEMS tilt mirror actuators considered in the cross-connect design study: (a) 1-axis comb, (b) 2-axis in-plane comb, (c) 2-axis hidden comb, (d) 1-axis plate, (e) 2-axis plate with gimbal, (f) 2-axis plate with hidden “crossbar” springs, and (g) 2-axis plate with hidden “post” spring. Plate actuators are shown with ramped electrodes, but were also analyzed with flat electrodes. Partial cross sections have been taken to reveal the actuator structure.
Log-log plot of resonant frequency vs. port count for single-axis (left) and dual-axis (right) devices (see previous figure) arranged in a conventional NxN free-space cross-connect, with insertion loss better than 3dB and crosstalk better than -20dB.
Example optimized cross-connect: (a) Optimal 2-axis in-plane comb device for a 132 port switch, arranged in an array accounting for the asymmetric fill factors and tilt angles. (b) Zemax model of the corresponding system including skew angle, microlenses, and Fourier lens.
Example multi-stage cross-connect: (a) Cross section of three 256 port cross-connects, each with 4x reduced scan angle, interconnected with free-space optics. Ray color indicates different optical paths through the system, depending on mirror states. (b) Detail of the passive free-space interconnection with possible connections.
A crossbar connects any two ports in a single hop through the switch using 1xN switching elements. A selector switch uses 1xk switching elements (k = log2N shown) to select amongst k port mappings. With k = log2N port mappings, data passes through the switch up to log2N times. For example, two hops are required to send from node 0 to 3, with data passing through intermediate node 1.
61-port prototype selector module with four interconnection matchings installed.