An optical cross-connect (OXC) is a network device that switches high‐speed optical signals between fiber inputs and outputs without converting them to electronics. In essence, an OXC uses photonic switching fabric to route wavelength channels from any incoming fiber to any outgoing fiber, typically by demultiplexing each WDM signal into individual wavelengths, directing them through a switch matrix, and then re-multiplexing onto output fibers. Because the signals remain in the optical domain (“transparent” switching), OXCs preserve data‐rate and protocol transparency. This all‐optical routing is controlled electronically (often via an SDN controller) to dynamically allocate bandwidth and restore paths without manual patching.
OXCs play a critical role in modern fiber networks by enabling dynamic wavelength routing, improving resilience and capacity utilization. In telecom backbones they serve as large optical switching nodes, and in data-center interconnects they enable high-bandwidth, low-latency routing of optical circuits. For example, OXCs allow a backbone mesh network to reconfigure wavelength paths on demand (e.g. for protection or traffic engineering) without human intervention. Similarly, in data centers OXCs can interconnect multiple switches or DC sites with many wavelengths of 100–400 G signals, supporting cloud and HPC applications.
Telecom backbones: OXCs are used at core and regional nodes to groom and switch entire wavelength channels across the network. They replace manual patch panels and even traditional ROADMs in some networks, providing all-optical one-hop paths. For instance, China’s carriers have deployed OXCs in high-degree backbone hubs (e.g. 32-degree and above) to simplify mesh networks and enable one-hop routing of heavy traffic. In a notable example, China Telecom Sichuan built a new all-optical “cube” backbone with 12 OXC nodes at its core, enabling 32-degree high-capacity grooming and 80% smaller footprint compared to ROADM-based sites. OXCs also support optical-layer network restoration, wavelength protection schemes, and flexible capacity expansion.
Data Center Interconnect (DCI): OXCs enable dynamic provisioning of large circuits between data centers or within large data-center fabrics. They can switch multiple fibers or 1×N/X×N arrays to aggregate many 100 Gb/s+ channels. By using fiber-to-fiber optical switching, DC operators can reconfigure lightpaths faster (milliseconds) than manual patch and with lower latency than OEO switching. Although broad industry deployment in data centers is still emerging, OXCs are a key enabler for scalable DC networks: they allow on-demand allocation of optical channels for cloud workloads and can reduce the number of electronic switches in a DC spine by handling high-capacity flows optically.
Different architectures have been developed for OXCs, each with trade-offs in speed, loss, and scale. The major types include:
Mechanical (Fiber/Prism) Switches: Early OXCs and small switches used mechanical optics. These use motorized prisms, mirrors or movable fiber stages to physically reroute light. For example, a 1×N mechanical switch may tilt a mirror or move a fiber to connect an input to a chosen output. Mechanical switches can have very low insertion loss (typically <1 dB) and high optical quality. Commercial 1×2 opto-mechanical switches have ~0.5 dB typical IL (max ~1 dB) and switching times on the order of ~8 ms. However, their port count is limited (tens of ports at most), moving parts require careful alignment, and frequent reconfiguration may reduce reliability. As a result, purely mechanical OXCs are now mainly seen in niche or legacy applications rather than large backbone nodes.
MEMS-Based Switches: MEMS (micro-electro-mechanical system) switches use tiny moveable mirror arrays to steer light paths. In a typical free-space MEMS OXC, each input fiber is collimated to a 2D array of tiny mirrors, which direct each beam to a chosen output fiberg. MEMS cross-connects support much higher port counts than mechanical switches – nowadays up to thousands of ports. For example, experimental MEMS fabrics have demonstrated over 1000×1000 fibers with mean fiber-to-fiber insertion loss ≈2.1 dB and worst-case ~4.0 dB. A 1296×1296 MEMS switch has been built with ~5.1±1.1 dB insertion loss and ~5 ms switching time. In general, MEMS OXCs offer low loss, broad wavelength transparency, and scalability (scaling roughly with N² mirror elements). Switching times are typically in the millisecond range (a few ms for large arrays), although smaller integrated MEMS devices can achieve sub-ms (µs) reconfiguration at the cost of smaller port count. The main downsides are moving parts (mirror fatigue over many cycles) and somewhat higher cost as size grows. MEMS OXCs have become the industry workhorse for high-degree, all-optical switches in research and some production systems.
Liquid-Crystal (LCoS) Switches: Liquid-crystal-on-silicon (LCoS) is a solid-state SLM technology borrowed from displays. In an LCoS switch, an array of pixels containing liquid crystal molecules imposes phase patterns on incoming light, steering it via diffraction. LCoS devices are often used in wavelength-selective switches (WSS) for ROADMs, but can also form the core of an OXC that operates on wavelengths. LCoS can achieve fine granularity (per-wavelength control) and support advanced features (colorless/flex-grid). However, LCoS-based switches tend to be slower and higher-loss: switching speeds are on the order of tens of milliseconds (liquid crystals take ms to reorient) and typical insertion loss is a few to ~10 dB (for example, 1×10 LCoS WSS designs report ~0.5–1 dB coupling loss plus several dB grating loss). In practice, LCoS OXC implementations are usually realized by combining multiple 1×N WSS modules in a backplane (as described below). The advantage is very high flexibility (wavelength routing, colorless/directionless/add-drop) and support for extended C+L bands; the trade-off is complexity, cost and somewhat higher optical loss compared to MEMS.
Wavelength-Selective Switch (WSS)-Based: An OXC can also be built by interconnecting multiple WSS modules with an optical backplane. Here, each fiber-to-fiber connection is implemented by a wavelength-selective path through 1×N or M×N WSS boards. Such architectures effectively combine LCoS (or MEMS) WSS engines with flexible backplane wiring. For example, a single “add/drop” board may contain a 1×16 LCoS WSS; multiple such boards plus WSS-equipped line cards can realize an M×N OXC. The key feature is granular, per-wavelength routing: each WSS can drop or pass individual wavelengths to any degree, enabling colorless, directionless, contentionless (CDC) operation. Performance-wise, WSS-based OXCs inherit LCoS characteristics (switch times ~10–50 ms, insertion loss several dB per pass). They do allow very high effective port counts by aggregating many WSS ports (e.g. a 32-degree OXC with 32 WSS add/drop slots can groom >1000 wavelengths). For example, modern Chinese OXCs use “TWIN 1×N WSS” boards (LCoS 1×N plus amplifier) that each handle 32 wavelengths per board. In summary, WSS-based designs maximize flexibility (fine granularity, flex-grid, C+L band) at the cost of speed and optical loss compared to simple fiber-switch architectures.
The key metrics for OXC fabrics include switching time, insertion loss (IL), port density (matrix size), crosstalk, and scalability:
Switching Time: Mechanical and MEMS OXCs typically switch in the millisecond range. For instance, a 1×2 opto-mechanical switch has ≲8 ms reconfiguration time, and large MEMS fabrics ~5–10 ms. (Waveguide-integrated MEMS can switch faster, ≪1 ms, but at smaller scale.) LCoS/WSS switches are slower; they reconfigure in tens of milliseconds due to the liquid crystal response. As networks evolve, switching speeds matter mostly for restoration scenarios: ms-scale is adequate for many backbone applications, though cutting-edge research explores microsecond or faster switches using silicon photonics and new materials.
Insertion Loss: All-optical switching incurs IL above 0 dB. A small mechanical switch may have IL ≲0.6–1 dB, whereas large MEMS fabrics see a few dB (e.g. a 1296-port MEMS OXC had ~5.1 dB median IL). Higher port counts generally add loss (each extra mirror or coupling stage). LCoS/WSS devices incur several dB IL per pass (e.g. commercial 1×10 LCoS WSS reports ~5–7 dB IL for typical channels). Crosstalk is also a concern (MEMS OXCs report –30 to –40 dB worst-case crosstalk). OXC designs target IL of a few dB plus minimal polarization-dependent loss.
Port Density & Scalability: MEMS OXCs dominate high-density scaling: free-space MEMS has reached ~1000×1000 switch fabrics, and reports of 240×240 on-chip waveguide MEMS. Mechanical switches top out at tens of ports (e.g. 8×8). LCoS/WSS scales by aggregating many wavelengths rather than ports: a single 1×N WSS might have N≈16–20 outputs, but dozens of such boards yield very high total throughput (e.g. a 32-degree OXC with 32 WSS slots can cross-connect thousands of wavelengths). In practice, today’s commercial OXC units support dozens of fiber degrees: Chinese systems are shipping 16-, 20-, and 32-degree OXCs (one fiber per degree). Future systems will push to 64 or 128 degrees as core hubs grow.
Historically, optical networks began with manual patch panels, where fibers were physically re-plugged to reroute traffic. The first automated OXCs (1980s–90s) were opto-mechanical switches similar to mechanical patch cabinets. They were later supplemented by opaque (OEO) cross-connects (electronic switching at the channel level). The advent of colorless/directionless ROADMs in the 2000s introduced limited all-optical reconfiguration per-wavelength, but ROADMs still required many internal fiber connections. Modern programmable OXCs use an all-optical backplane and electronic control plane (often under SDN) to fully automate fiber connectivity. Compared to manual methods, today’s OXCs allow instantaneous (ms-scale) cross-connections by software, eliminating human error. This evolution—from fixed patch panels to flexible, controllable OEO switches to today’s transparent OXCs—has been driven by advances in MEMS, LCoS and PICs.
OXCs are now being deployed in next-generation networks worldwide. In China, all three major carriers have rolled out OXCs in core and metro backbones. For example, since 2018 they have introduced 16/20/32-degree OXC nodes to replace ROADM sites. In one provincial backbone, China Telecom Sichuan built a full-mesh “optical cube” network with 12 OXC nodes at its core, enabling one-hop transmission of high-value traffic. Core-site capacity is rising: China Telecom’s Taiyuan hub grew to 57 degrees of optical connectivity, exceeding the 32-degree limit of current WSS; this suggests 64- or 128-degree OXCs are needed next. Operators also use OXC for metro hubs (typically 16°–32° for core, 8°–16° for edge).
Globally, OXC adoption is expanding as backbone capacities surge. Commercial systems now support Petabit/s fabrics (e.g. a 40λ×40 Gb/s×40 ports = 2.07 Pb/s MEMS switch). Power-hungry, fiber-intensive ROADMs are giving way to single-cabinet OXCs: one vendor reports a 20-degree OXC needs 1/3 the footprint of a 20-degree ROADM. In data centers, optical circuit switches (a form of OXC) are being trialed for disaggregated computing, though real-world DCI stats are emerging. On the capacity trend line, OXC node counts will track backbone upgrades. For example, NTT and other global carriers are researching 200G+ per-wavelength OXCs, and industry analysts forecast multi-billion-dollar markets driven by SDN-ready, high-degree OXCs.
Switching Time: Mechanical/MEMS ≈1–10 ms. LCoS/WSS ≈10–100 ms. (Sub-ms silicon photonic switches are emerging but have much lower port count.)
Insertion Loss: Mechanical ≲1 dB. Large MEMS ≈2–5 dB (with max ~4–5 dB). LCoS/WSS often 5–10 dB.
Port Count: Mechanical: up to ~8–16 ports. MEMS: hundreds to >1000 (1120×1120 reported). LCoS/WSS: typically 1×N (N≈8–32) per card, aggregate to many wavelengths/ports in a backplane.
Scalability: MEMS offers the largest fabric sizes (up to 1000s of ports). LCoS scales by adding more WSS boards (limited by cost/power). Mechanical does not scale well beyond small N.
Crosstalk/PDL: All types target <–30 dB crosstalk. Moving-mirror designs tend to have very low PDL. WSS may require careful calibration to minimize crosstalk between adjacent wavelengths.
OXC technology continues to evolve. SDN Integration: Modern OXCs are being integrated into software-defined control planes. SDN controllers (e.g. ONOS/TeraFlow) now support optical resources, enabling end-to-end multilayer provisioning across OXCs, ROADMs and packet switches. This orchestration will allow operators to dynamically allocate spectrum and paths with fine granularity.
Space-Division Multiplexing (SDM): As single-fiber bandwidth limits approach, future OXCs will need to switch spatial channels. OXC research is exploring multicore and few-mode fiber fabrics, where each core or mode is treated as a “fiber port.” Early labs have demonstrated cross-connects for multicore fiber (with 7 or 12 cores) using similar MEMS/LCoS techniques. In practice, an SDM-OXC could switch hundreds of cores/modes by stacking multiple switch planes. This is a nascent area, but large carriers (e.g. NTT, Toshiba) are trialing long-haul multicore links, implying SDM-OXCs on the horizon.
Extended Band (C+L+): Current OXC/WSS designs typically cover the C-band (1530–1565 nm). To squeeze more capacity, next-generation systems are expanding to the extended L-band. For example, ZTE reports OXCs that support CE/C+/L++ bands, and M×N WSS modules are being developed with C+L support. Future OXCs will natively handle 150 nm+ of spectrum, boosting per-fiber throughput by ~1.5× or more.
Higher Degrees and Port Counts: Demands for mesh networking are driving development of 64- and 128-degree OXCs. Commercial products are already at 32° per chassis; scaling further will require advances in backplane connectors and WSS technologies (e.g. multi-slotted LCoS blocks).
In summary, OXCs are transitioning from static fiberpatch panels to fully programmable, high-capacity switches. They now form the backbone of ultra-high-speed optical networks, and trends like SDN control, SDM fabrics, and multi-band operation will further enhance their flexibility and capacity in the years ahead.
If you want to know more about us, you can fill out the form to contact us and we will answer your questions at any time.
Shenzhen ADTEK Technology Co., Ltd., founded in 2007, is a national high-tech enterprise that integrates R&D, manufacturing, and sales. As a leading provider of optical connectivity solutions, ADTEK brings over 18 years of design and manufacturing experience, specializing in customized fiber optic connectivity products for domestic and international telecom operators, cloud service providers, equipment manufacturers, and system integrators.