Fast optical circuit switches (OCS) are designed to rapidly reconfigure optical connections without the need for optical-to-electrical conversion.  While these switches can replace electrical switches in some network use cases such as restriping in a leaf-spine network, optical switches have a wide range of applications from managing optical networks to lab automation.  Whatever the use case, the goal is to more efficiently utilize resources by allowing reconfiguration and optimization at the fiber layer.

The ever-increasing demand for computer processing power and GPU usage has led to an exponential increase in data and created a surge in network bandwidth requirements – leading to an explosion in the number of fibers used in data centers and optical networks.  While there is a range of technologies that have been used as fast optical switches, from MEMS to piezo beam steering to liquid crystals, none of these can scale beyond a few hundred ports. In fact, with most of these technologies scaling degrades the performance – both the insertion loss and cost per port increase due to the added complexity in higher port count systems.  

There is one technology that can offer very high port density for optical switching.  The Telescent robotic switch system is a low-loss all-fiber system where a robot picks and weaves fibers around other fibers in the system using a patented routing algorithm.  With the use of high-density connectors such as MPO16 connectors, the Telescent robotic system can handle up to 16,128 fibers in a single rack.  

Combing the speed of fast optical switches with the scale of Telescent’s robotic optical switch allows fast optical switching at scale.  Since the Telescent robotic system has a very low insertion loss, a hybrid approach where fast optical switches are combined with a large scale Telescent robotic switch offers the best of both worlds.  

The robotic optical switch provides a large port count, allowing for widespread connectivity across the data center. This switch can be designed to handle the bulk of the optical connections, providing a high degree of scalability. Meanwhile, the fast small-scale optical switch can be used to rapidly reconfigure a subset of connections, providing the necessary speed and agility.


The Fundamental Challenge: Speed vs. Scale

The optical switching landscape presents a frustrating dichotomy:

Fast optical circuit switches (OCS) using technologies like MEMS (micro-electromechanical systems) or liquid crystal can reconfigure in microseconds or even nanoseconds. However, as the array size increases, costs and insertion loss increase for these technologies.

Large-scale optical switches can handle hundreds or thousands of ports but traditionally rely on mechanical components with reconfiguration times measured in seconds or minutes, far too slow for dynamic workloads.

This leaves data center architects in a difficult position: choose speed or scale, but not both.

The Hybrid Solution: Combining Robotics with Fast Optics

In this hybrid architecture, the robotic optical switch acts as a coarse-grained switch, establishing persistent connections between servers with the ability to provide wide scale reconfiguration. The fast small-scale optical switch then fine-tunes these connections, rapidly reconfiguring them as needed to accommodate changing data center demands locally. This two-tiered approach allows for both scalability and speed, making it an attractive solution for data centers.

Benefits of the Hybrid Approach

This architecture delivers several key advantages:

  • Scalability: The robotic component provides the port count needed for large data centers

  • Speed: The fast optical switches handle dynamic traffic requirements within a segment of the fibers

  • Energy efficiency: Both technologies maintain the inherent power advantages of optical switching over electrical packet switches

  • Future-proofing: The modular nature allows incremental upgrades as technology evolves

Benefits and Future Directions

As data centers continue to evolve and grow, the need for fast and scalable optical switching solutions will only increase. The hybrid approach described here offers a promising solution.  As workloads become increasingly bandwidth-intensive and latency-sensitive, particularly with the growth of AI, real-time analytics, and edge computing, innovative optical architectures will be essential to maintaining sustainable data infrastructure growth.