Optical transparency in modern networks eliminates costly optical-electrical-optical conversions, significantly reducing latency, power consumption, and hardware complexity while offering significant economic benefits. Optical circuit switches (OCS) build upon these advantages, bringing optical transparency to data center networks and offering a compelling solution for data center architectures. Google has demonstrated significant benefits of using OCS technology throughout their data centers including a 30% cost reduction and 40% power savings in their networks. [1]
The initial MEMS OCS used in the Google network had 136 ports and this low port count required additional changes in the network prior to implementation. These changes included the use of bi-directional optics to effectively double the port count, circulators, higher power transceiver to handle the high loss of the MEMS devices as well as other network changes. There is a significant demand for larger port count optical switches to address the demands of ever larger data centers to accommodate recent HPC and AI computing demands while also minimizing the network changes required for implementation.
In addition to MEMS, technologies to address the desire for large-port-count optical switches include piezo-electric stacks, liquid crystal arrays and robotic fiber systems. Aside from the all-fiber robotic fiber system, all the other technologies include a free-space propagation component where the signal exits the fiber to be switched between individual ports and then focused back into the exit fiber. This highlights a challenge for scaling all these technologies – the fiber collimator array. In contrast, the robotic fiber switch simply relies on existing fiber connector technology and can easily scale to > 10,000 fibers per system while offering a low cost per port. In fact, for robotic fiber systems with large fiber counts, the cost is dominated by the fiber cables and the cost to add automation is a small increment over the cost of passive patch panels.
The Complexity of Large-Array Fiber Collimators in MEMS and Other OCS Technologies
Large-array fiber collimators are critical components in optical switches, as they enable the precise alignment and coupling of light between fibers and the free-space MEMS, piezo or liquid-crystal optical elements. While these technologies offer some performance benefits, the fiber collimator array introduces several engineering challenges:
Precision Alignment and Manufacturing
MEMS switches rely on micro-mirrors to redirect light beams across hundreds of fibers. Similarly 2D piezo electric beam steering technology uses fiber collimator arrays for the free-space propagation segment between the piezo electric beam steering elements. Since single mode fibers have a core diameter of 9 micron, to minimize loss collimators must align fibers to an angular tolerance of <1 milliradian during the free-space propagation volume.
Precisely aligning a large number of fibers within a collimator array becomes increasingly difficult as the scale increases, requiring sophisticated mechanical designs and precise positioning mechanisms.
Optical Loss and Crosstalk
The free-space nature of MEMS and piezo switches introduces susceptibility to alignment errors, which can result in significant insertion loss or crosstalk between channels.
Maintaining consistent beam quality across a large array further complicates design, as any imperfection in collimators can degrade performance.
Thermal, Electrical and Mechanical Stability
Environmental temperature fluctuations, variation in voltage and vibration can cause slight shifts in alignment, impacting beam quality and coupling efficiency, requiring robust environmental management strategies.
Scalability and Cost
As network demands grow, MEMS and piezo switches require larger arrays to accommodate higher fiber counts. Scaling collimator arrays while maintaining low insertion loss and high precision significantly increases production costs.
These arrays are not made in volume today to address hyperscale data center volume demands.
Robotic Patch Panels: A Simpler Alternative
Robotic patch panels offer a straightforward, fiber-based approach to reconfiguring optical networks. These systems use robotic arms to physically move and connect fiber ends, mimicking manual patching but with automation. While not as dynamic or fast as MEMS or piezo OCS, robotic patch panels excel in simplicity and reliability. Since the robot can manipulate any fiber connector including MPO type connectors, fiber counts per system can scale to > 10,000 per rack. Telescent’s use of a patented routing algorithm ensures that reconfiguration can be done without blocking, offering an any-to-any reconfiguration between any ports in the system. Benefits of the robotic fiber system include ease of implementation, significantly lower cost and scalability to > 10,000 fibers per system.
The Telescent Robotic Cross-Connect System
Telescent offers a robotic, all-fiber robotic cross-connect system that addresses the need for optical reconfiguration in data center networks. It comprises short fiber links between ports and a robotic mechanism to reconfigure these ports as needed. The system's scalability to high port counts is due to its patented routing algorithm, which enables fibers to be intricately woven around others without blocking. Initially designed for 1,008 simplex LC ports, the Telescent robotic switch has evolved to accommodate multiple fibers per port through duplex and MPO-style connectors, exceeding 10,000 fibers per rack. The Telescent system has passed NEBS Level 3 certification and has been used in production networks. Both single mode and multimode fiber have been deployed in the Telescent system, allowing use with a range of network designs. While a robotic cross-connect system is slower than other optical cross-connect technologies, the high port density with minimal insertion loss are key benefits for higher-bit rate networks using parallel optics and declining link margins.
Conclusion
Of course, the choice between MEMS optical switches and robotic patch panels hinges on the specific requirements of the application. MEMS and piezo OCS with their fast switching speeds are ideal for high-performance, low-latency networks. However, their reliance on large-array fiber collimators introduces significant engineering challenges that can drive up complexity and cost and limits their ability to scale beyond several hundred fibers per system.
Conversely, robotic patch panels provide a simpler, more cost-effective solution for environments where speed is less critical but reconfigurability of the fiber layer brings a new option for network optimization. While they lack the speed of MEMS switches, their reliability and ease of implementation of robotic systems make them a viable alternative in many use cases. In addition, their low loss, ability to accommodate large fiber densities and low cost make robotic patch panels a very attractive option for a range of hyperscale data center use cases.
[1] [2208.10041] Mission Apollo: Landing Optical Circuit Switching at Datacenter Scale (arxiv.org)