Adiabatic coupler
A waveguide coupler that transfers or splits power by changing its geometry slowly along its length, so light follows one local mode of the structure throughout. Broadband and tolerant of fabrication errors, at the cost of length.
Two waveguides placed side by side support two supermodes, one symmetric and one antisymmetric, whose shapes depend on how the guides differ. A directional coupler launches light into both supermodes at once and relies on their beating to move power from one guide to the other, so the split depends on the length and on the difference between the two supermode indices, both of which change with wavelength and with small fabrication errors. An adiabatic coupler works the other way: it excites only one supermode and then changes the structure so slowly that the light stays in that mode as the mode itself changes shape. Where the light ends up is decided by the geometry at the far end, not by an interference condition.
A common layout starts with two guides of different widths far apart, so that the supermodes are the individual guides' modes. The guides are brought together and their widths are tapered through the point where they match; the supermode that began in the wider guide ends in the other guide, completing a full power transfer, or, if the taper stops at equal widths and the guides separate symmetrically, the light divides equally into a 3 dB split. The same principle, a slow transformation of a single mode, underlies spot-size converters and inverse tapers, the mode evolution in polarization splitter-rotators, and the tapered couplers that move light between a silicon waveguide and a III-V layer in heterogeneous lasers.
"Slowly" has a precise meaning. Power leaks from the intended supermode into the other one at a rate set by how fast the structure changes compared with the beat length between the two supermodes, so the transition must be long where the two supermode indices are close together, typically at the point where the guides match. Designs therefore vary the taper rate along the device, changing quickly where the supermodes are well separated and slowly near the crossing point, to reach a given leakage in the shortest length.
The payoff is tolerance. Because the result does not depend on an exact coupling length, an adiabatic coupler holds its split ratio over a wide wavelength range and across the width variations of a real wafer, where a directional coupler's ratio can drift by tens of percent. The price is size: adiabatic couplers are typically several times longer than a directional coupler for the same function, and longer structures accumulate more propagation loss and occupy more chip area. They are chosen where broadband or fabrication-tolerant splitting matters more than footprint, for example in switch matrices, broadband interferometers and the 3 dB couplers of polarization-diversity circuits.
References: Y. Shani, C. H. Henry, R. C. Kistler, R. F. Kazarinov, K. J. Orlowsky, IEEE J. Quantum Electron. 27, 556 (1991); T. A. Ramadan, R. Scarmozzino, R. M. Osgood, J. Lightwave Technol. 16, 277 (1998).