As data center networks seek greater bandwidth and lower latency, optical switching offers an alternative to relying exclusively on electronic switching. Silicon photonics can support integrated, multiport switching, but insertion loss becomes a significant design constraint as switch fabrics grow. A demonstrated lossless optical switch addresses this challenge by combining a silicon photonics (SiP) routing chip with an indium phosphide (InP) semiconductor optical amplifier gain block.
The experimental 8×8 hybrid design uses the SiP device to route signals and the InP section to restore optical power lost along the path. Its results illustrate both the potential of hybrid integration and the engineering trade-offs involved, including amplifier noise, gain saturation, and packaging between different materials.
Why optical switch insertion loss matters
Silicon photonics is attractive for optical interconnects because it draws on CMOS-compatible fabrication, wafer-scale testing, and a high level of integration. However, the silicon platform cannot provide on-chip light amplification. In a high-radix switch, losses from coupling and on-chip propagation can add up, making it harder to deliver adequate optical power at the output.
InP offers active optical functions, including semiconductor optical amplifiers (SOAs) that provide gain. Yet InP Mach-Zehnder interferometer (MZI) switches can be affected by fabrication variation and crosstalk. The hybrid approach assigns routing to the SiP chip and amplification to InP, bringing together the switching characteristics of one platform and the gain capability of the other.
This arrangement also introduces its own challenges. Coupling light between the SiP and InP chips adds loss, and packaging two different materials increases system complexity. The design must therefore provide enough amplifier gain to offset losses across the complete optical path, not just within the switch chip.
How the 8×8 lossless optical switch works
The demonstrated device pairs an 8×8 thermo-optic Banyan switch, built from MZI switching elements on silicon-on-insulator, with an eight-channel InP gain block. Each channel in the block contains a 1,300 µm-long SOA. Thermal phase shifters control the SiP switch, while the InP amplifiers boost the routed optical signals.
At the interfaces, the chips and fibers are edge-coupled. A trident-waveguide spot-size converter helps match the optical modes between the SiP and InP chips. The InP gain block also uses tapered waveguides, coated facets to reduce reflectivity, and output ports tilted by seven degrees to reduce residual Fabry–Perot effects.
The reported design values show why substantial gain is needed. They include 5.0 dB of input fiber-array coupling loss, 6.7 dB of SiP on-chip loss, 7.0 dB of coupling loss between the SiP switch and InP block, 2.9 dB of InP on-chip loss, and 3.0 dB of output fiber-array coupling loss. The SOA net gain is specified at 24.6 dB. In the reported operating condition, approximately 25 dB of gain enables zero fiber-to-fiber optical loss.
Gain, saturation, and signal quality
Measurements of the gain block showed that fiber-to-fiber loss increased as input optical power rose from 1.5 dBm to 12 dBm. The reason is SOA gain saturation: as the amplifier approaches saturation, it provides less additional gain. At an input power of 1.5 dBm, the reported lossless condition required an SOA bias current of 75 mA.
Amplifier gain must also be evaluated alongside amplified spontaneous emission (ASE), which contributes noise and can reduce optical signal-to-noise ratio (OSNR). In the reported measurements, OSNR initially increased with SOA bias current as net gain rose, then saturated and decreased slightly as the amplifier approached gain saturation. Even at higher bias currents, measured OSNR remained above 28 dB.
What the 12.5 Gb/s transmission test showed
To assess payload transmission, the experiment used a 12.5 Gb/s nonreturn-to-zero pseudorandom binary sequence with a PRBS-31 pattern. The signal was tested through one InP gain-block channel at a time, with its input power set to -17.5 dBm to represent a weakened signal after propagation through the SiP chip. Eye diagrams were used to assess the received signal.
- At 30 mA SOA bias: The received optical power was too low for a clear eye diagram.
- At 50 mA: The gain was sufficient to compensate for system losses, and the measured eye was clear and open.
- At 70 mA: Gain saturation and self-gain modulation appeared, so bit sequences with more ones experienced less gain than sequences with more zeros.
- At up to 90 mA: ASE raised the binary-zero level, but the eye remained open in the reported measurements.
These observations highlight why amplifier current cannot be treated simply as a way to maximize output power. The operating point affects gain, saturation behavior, and noise. System designers need to consider these factors together when setting a power budget and evaluating transmission quality.
Practical implications for optical switch design
The experiment demonstrates lossless fiber-to-fiber operation for an 8×8 SiP/InP hybrid switch and open-eye 12.5 Gb/s transmission under the reported test conditions. It does not remove the underlying losses; instead, it compensates for them with an inline InP SOA array. For engineers evaluating larger optical fabrics, this distinction matters: the gain block, chip-to-chip interface, fiber coupling, and amplifier noise all form part of the switch system.
Hybrid integration offers a route to combining silicon photonics routing with III-V amplification, while making coupling and packaging important design considerations. The practical takeaway is to treat insertion loss and optical signal quality as linked requirements. A gain budget that restores output power must also be assessed for saturation and ASE-related OSNR effects. In this 8×8 demonstration, approximately 25 dB of SOA gain supported zero measured fiber-to-fiber loss while maintaining an OSNR above 28 dB and a clear eye at 12.5 Gb/s.