Optical interconnects: free or bonded?
Wavelength division multiplexing sends multiple data streams through fiber by encoding them on different light wavelengths. What if you could multiplex spatially instead, a grid of pixels per pulse? That leads to Free-Space Optics. It uses VCSEL arrays to transmit through air and aims for very low latency. CPO, the pragmatic approach integrating fiber inside processor packages, started volume production in 2025. FSO remains mostly in research labs because micron-level alignment is hard to maintain in real datacenters.
Electrical interconnects degrade beyond about 50cm at 100+ Gbps, and power per bit climbs quickly. As AI workloads push petabytes between accelerators, optical links become increasingly attractive. CPO trades latency for proven reliability, while FSO tries to keep latency low but depends on tight alignment.
The interconnect wall
Training trillion-parameter models requires moving petabytes between accelerators, but copper traces consume watts per Gbps and degrade beyond 50cm at 100+ Gbps rates. Torus topologies can make it worse, adding cable length and latency that builds up across synchronized operations.
Co-Packaged Optics (CPO), integrating fiber-coupled photonics inside processor packages, already ships in volume production. Free-Space Optics (FSO), replacing cables entirely with steerable laser beams transmitted through air, targets lower latency but faces alignment challenges that have kept it in research labs. FSO's challenges begin with alignment.
FSO vs CPO
FSO uses VCSEL arrays to transmit parallel data streams through datacenter air. Lab demos report 1.6 Tb/s at 2.3 pJ/bit with sub-nanosecond latency. Steerable beams could reconfigure links without re-cabling, which is attractive for HPC systems with changing topology.
FSO requires maintaining ±5 µm alignment across meter-scale distances in datacenters subject to thermal expansion, floor vibration, and building settling. A 10 µm misalignment can cut power by over 50% and raise bit error rates. Active alignment systems employ piezoelectric actuators with sub-microsecond response, but every link needs dedicated tracking hardware running continuously.
Indoor atmospheric turbulence from equipment heat dissipation causes beam wander and intensity flickering beyond 1-2 meters. Datacenters are not controlled optical labs, so stability is hard to maintain at scale.
Dense VCSEL arrays generate heat causing thermal crosstalk, wavelength drift, and power fluctuations. Dust and condensation can disrupt beams. Hermetic packaging is often required for reliability. While researchers work on these challenges, most production deployments have chosen the option that trades latency for stability.
See the strain: Alignment Control Loop
The interactive below puts FSO's precision challenge in your hands. Tune the PID controller that drives MEMS mirrors, then inject thermal drift and dust disturbances. Watch alignment error spike beyond ±5 µm and bit error rate collapse. This is why FSO remains experimental: the control loop must fight continuous mechanical chaos while maintaining Gbps data integrity.
CPO integrates silicon photonic transceivers inside processor packages alongside the ASIC die. Electrical signals convert to optical at package edge, then couple into standard fiber. The approach drops I/O power from 10+ pJ/bit to 5-10 pJ/bit and eliminates external transceiver modules.
CPO leverages decades of fiber infrastructure, proven silicon photonics processes, and established packaging techniques. Alignment tolerances relax to ±50 µm, 10x looser than FSO, and hermetic sealing protects optics from contamination.
The cost is latency. CPO transceivers need DSP and FEC to fight fiber dispersion, adding 100-150 ns per hop. For tightly synchronized HPC collectives (AllReduce across thousands of GPUs) or latency-sensitive real-time workloads, these nanoseconds accumulate.
The choice between FSO and CPO represents a fundamental engineering trade-off:
| Metric | Free-Space Optics | Co-Packaged Optics |
|---|---|---|
| Latency | Sub-nanosecond | 100-150 ns (FEC overhead) |
| Power Efficiency | 2.3 pJ/bit | 5-10 pJ/bit |
| Alignment Tolerance | ±5 µm (critical) | ±50 µm (relaxed) |
| Environmental Sensitivity | High (dust, vibration, turbulence) | Low (enclosed fiber) |
| Manufacturing Maturity | Experimental | Production-ready |
Each approach optimizes different constraints. FSO offers very low latency and strong power efficiency but demands mechanical precision and environmental control not yet widely proven at datacenter scale. CPO provides robustness through mature fiber technology at the cost of 100-150 ns fixed latency overhead.
Deployment path
CPO met immediate bandwidth needs using mature fiber technology and shipped in volume by 2025, despite latency overhead. FSO targets specialized applications (supercomputing, quantum interfaces) where the latency gains justify the complexity. MEMS advances, improved packaging, and AI-assisted alignment may enable broader FSO adoption, but datacenter deployments require breakthroughs in alignment stability. Production systems often choose what fits existing infrastructure, while research continues to test whether FSO can overcome its physical challenges.