Fibre-to-chip coupling is the packaging interface that gets light from an optical fibre into a sub-micron photonic waveguide and back out again, and it is now the loss, cost and yield bottleneck standing between co-packaged optics and volume deployment.
Summary
A single-mode optical fibre carries a mode roughly ten micrometres across. A silicon or lithium niobate waveguide confines light to a few hundred nanometres. Getting from one to the other without throwing away most of the power is a mode-matching and mechanical-alignment problem, and it has to be solved at every port, on every die, in a production line. Two families of solution dominate. Edge (facet) coupling uses an inverse taper that expands the on-chip mode until it overlaps the fibre mode; recent work on thin-film lithium niobate at 1064 nm used a bilayer inverse taper to reach a measured 1.9 dB per facet for the TE mode with a 1 dB bandwidth of 1055 to 1085 nm, with simulation projecting 0.48 dB per facet if a lensed fibre with a 2.5 micrometre mode field diameter is used. Grating coupling diffracts light out of the plane of the chip, trading some bandwidth and polarisation tolerance for the ability to couple anywhere on the die surface and to test at wafer level; inverse-designed broadband grating couplers were the optical input path for a monolithic 45 nm CMOS receiver running 32 channels at 32 Gb/s for an aggregate 1.024 Tb/s at 71 fJ/b and 4 Tb/s/mm2, with bit error rate below 1e-12 and no equalisation or forward error correction.
The parameters that decide the technology are: insertion loss per facet (paid twice per link, and multiplied by channel count in wavelength-multiplexed engines); optical bandwidth and polarisation dependence; lateral and vertical alignment tolerance, which sets whether assembly can be passive and machine-paced or must be active and slow; and mechanical and thermal stability of the joint over the product’s life. The 1064 nm TFLN work explicitly characterises polarisation-dependent alignment tolerance in the lateral and vertical directions, which is the parameter a packaging engineer cares about more than peak efficiency.
Evidence base
Open questions
Assessment drafted 2026-08-31 from up to 11 KB sources using the technology-scorecard framework; scores are a draft read pending review.