Chapter 16Edge Couplers and Spot-Size Converters
Coupling at the chip edge is broadband and low-loss, at the price of a prepared facet and tight alignment.
- estimate butt-coupling loss from mode overlap;
- explain how an inverse taper expands the mode, and choose a tip width;
- describe multi-stage, SiN, polymer, SWG and suspended couplers;
- handle facet reflections with angled waveguides, coatings and index matching;
- choose fibres for edge coupling, and compare edge and grating couplers.
How narrow must the taper tip be to match a 3 µm or a 10 µm spot?
16.1Edge-coupling basics
An edge coupler brings the waveguide to a polished or etched facet at the chip edge and couples it end-on, or butt-couples it, to a fibre or a laser. There is no grating and no resonance, so the coupling is broadband and can work for both polarisations. Its efficiency is the overlap of the two modes at the facet (Chapter 3); for Gaussian-like modes of 1/e² radii \(w_x, w_y\) and a fibre of radius \(w_f\),
The trouble is the size mismatch. A 500 × 220 nm silicon strip has a mode about 0.4 µm across; SMF-28’s is 10.4 µm. femwell gives an overlap of 1 % (−20 dB), and even a lensed fibre focused to 2.5 µm reaches only 12 % (−9 dB). Some kind of spot-size converter is needed to expand the waveguide mode before the facet.
16.2Inverse tapers
The obvious converter widens the waveguide, but a wide silicon waveguide confines light more tightly and becomes multimode. The trick, which seems backwards at first, is to narrow it [68]. As a silicon strip is tapered towards a tip of a hundred or two nanometres, it becomes too small to hold the mode. The effective index falls towards the cladding’s (1.444), the evanescent decay constant \(\gamma = k_0\sqrt{n_\mathrm{eff}^2 - n_\mathrm{ox}^2}\) shrinks, and the mode spreads out into the oxide cladding, which acts as a larger, weakly guiding waveguide. If the taper is long enough to be adiabatic (Chapter 8), the light follows the fundamental mode from 0.4 µm to several micrometres with little loss.
Table 16.1 shows how fast the mode grows, from full-vectorial femwell solutions of 220 nm silicon tips in oxide.
Table 16.1. Mode of a 220 nm silicon tip in oxide at 1550 nm (femwell), and its coupling to three fibres. MFD: 1/e² intensity diameter from the second moments, horizontal × vertical. Overlaps use the full mode field; “below 2 µm” is the fraction of the mode’s power more than 2 µm below the core, where a standard BOX ends. Computed by figsrc/ch16_taper.py.
| Tip | Pol. | neff | MFD (µm) | Lensed 2.5 µm | UHNA 4 µm | SMF-28 10.4 µm | Below 2 µm |
|---|---|---|---|---|---|---|---|
| 130 nm | TE | 1.4445 | 11.8 × 14.3 ≈ | 33 % | 52 % | 90 % | 21.0 % |
| 140 nm | TE | 1.4452 | 9.1 × 9.9 | 49 % | 70 % | 89 % | 13.8 % |
| 150 nm | TE | 1.4466 | 6.6 × 6.7 | 66 % | 84 % | 77 % | 8.0 % |
| 160 nm | TE | 1.4490 | 4.7 × 4.9 | 80 % | 91 % | 60 % | 4.2 % |
| 170 nm | TE | 1.4528 | 3.7 × 3.6 | 89 % | 89 % | 44 % | 2.0 % |
| 180 nm | TE | 1.4586 | 2.9 × 2.8 | 92 % | 82 % | 32 % | 0.8 % |
| 200 nm | TE | 1.4785 | 1.9 × 1.9 | 86 % | 62 % | 17 % | 0.1 % |
| 250 nm | TE | 1.6018 | 1.0 × 0.9 | 48 % | 25 % | 4 % | 0.0 % |
| 500 nm | TE | 2.4459 | 0.4 × 0.4 | 12 % | 5 % | 1 % | 0.0 % |
| 90 nm | TM | 1.4457 | 8.3 × 7.9 | 57 % | 77 % | 84 % | 10.8 % |
| 100 nm | TM | 1.4470 | 6.2 × 6.1 | 70 % | 87 % | 73 % | 6.9 % |
| 120 nm | TM | 1.4514 | 4.0 × 3.9 | 87 % | 90 % | 49 % | 2.5 % |
| 140 nm | TM | 1.4585 | 2.8 × 2.9 | 92 % | 82 % | 32 % | 0.8 % |
| 160 nm | TM | 1.4685 | 2.2 × 2.2 | 90 % | 71 % | 22 % | 0.3 % |
| 200 nm | TM | 1.4962 | 1.5 × 1.6 | 78 % | 51 % | 12 % | 0.0 % |
| 500 nm | TM | 1.7706 | 0.8 × 0.7 | 34 % | 16 % | 3 % | 0.0 % |
A 3 µm spot. Table 16.1 points to a TE tip of about 180 nm. The mode is 2.9 × 2.8 µm and couples 92 % (−0.36 dB) to a 2.5 µm lensed fibre, with less than 1 % of it more than 2 µm below the core. A UHNA fibre (4 µm) is best matched near 160 nm: 91 % (−0.42 dB), but 4 % of the mode already reaches beyond 2 µm.
A 10 µm spot. SMF-28 needs a TE tip of about 140 nm. femwell gives a 9.1 × 9.9 µm mode and 89 % overlap, but 14 % of the power lies more than 2 µm below the core. On a 2 µm BOX that light leaks into the substrate, so a single silicon taper cannot serve SMF-28 directly.
TM reaches the same spots at narrower tips: 2.8 µm at 140 nm, 4 µm at 120 nm. Even the best overlaps are about 92 %, not 100 %, because the taper’s mode is not Gaussian: its exponential tails cannot match a Gaussian fibre mode exactly.
Three things limit how far a single silicon taper can go:
- Substrate leakage. A large mode in a 2 µm BOX reaches the silicon substrate, whose index (3.48) is above the mode’s, and leaks into it (Chapter 7). Couplers for large spots use a thicker BOX, remove the substrate under the taper, or hand the light to a second, larger waveguide (Section 16.3).
- Polarisation. Below about 220 nm the tip is narrower than it is tall, so TE light (field across the narrow width) escapes sooner than TM. The same tip gives two very different spots, and a polarisation-independent coupler needs a square tip or a second stage.
- Fabrication. Near the best tip width the spot changes by about 0.1 µm per nanometre of width, and tips of 100–200 nm are near the smallest features a foundry prints reliably. The tip width varies across a wafer, and so does the coupling.
Try it. Narrow the tip and watch the mode grow in the to-scale facet view; choose a fibre, and move it sideways to see the tolerance.
16.3Multi-stage couplers and overlays
To reach a 10 µm spot without leaking, the silicon taper hands the light to a second, lower-index waveguide that is large enough to guide the expanded mode on its own:
- Polymer or oxide overlays. A waveguide of polymer or silicon oxynitride, a few micrometres across, is built over the silicon taper. The light moves into it as the silicon narrows, and the overlay carries it to the facet with a mode matched to a fibre. This was one of the first low-loss converters for silicon wires, at 0.8 dB per conversion [69].
- Silicon nitride stages. Most foundry platforms put a SiN layer above the silicon. A silicon taper couples into a SiN waveguide, and a SiN taper, with its lower index contrast and therefore larger modes from wider, easier tips, expands the light further; multi-tip “trident” designs shape the spot in both directions.
- SWG metamaterial couplers. Segmenting the taper into a subwavelength grating lowers its effective index gradually (Chapter 13), so the mode expands with features that are easier to print. Such couplers have reached high efficiency for both polarisations over more than 100 nm [70].
- Suspended couplers. Etching away the substrate under the taper, leaving a cantilever of oxide, removes the leakage path and lets the mode grow large in a symmetric cladding.
These designs and their trade-offs are reviewed by Marchetti et al. [66].
16.4Facets
The facet must be optically flat at the coupler, and there are three ways to make it:
- Dicing and polishing. The wafer is diced, and the edge is polished in a few steps down to a fine grit. It gives excellent facets, but one chip at a time.
- Deep etching. A deep trench etched through the BOX and into the substrate defines the facet lithographically, at wafer scale and at any position. The chip is then diced in the trench, and the couplers can be tested on the wafer by lowering a fibre into the trench.
- Cleaving along a crystal plane, common for III-V lasers, which are on their own substrates.
At the facet, part of the light reflects. With the mode in the oxide, an oxide–air facet reflects \(R = ((1.444 - 1)/(1.444 + 1))^2 = 3.3\ \%\) (−14.8 dB); a bare silicon strip would reflect 31 %. Reflections cause fringes in measurements and feedback into lasers (Chapter 15), and are reduced three ways: index-matching gel or epoxy between facet and fibre, an anti-reflection coating, or an angled waveguide that meets the facet at a few degrees, so that the reflected beam misses the waveguide’s mode.
Tilt the waveguide by 8° at the facet. The reflected beam comes back at 16° to the waveguide axis, and couples back into a mode of radius \(w\) with efficiency \(\exp[-(\pi n w\,\alpha/\lambda)^2]\) for a tilt \(\alpha\) inside the oxide (Chapter 15). For a 3 µm spot (\(w = 1.5\) µm), \(\pi \times 1.444 \times 1.5 \times 0.279/1.55 = 1.23\), so the reflected light is suppressed by a further \(e^{-1.50}\), or 6.5 dB: −21.3 dB in all. The beam leaves the chip at \(\sin^{-1}(1.444\sin 8°) = 11.6°\) in air, so the fibre must be tilted to match. The larger the spot, the more an angle helps.
16.5Fibres at the edge
The fibre is chosen to match the spot the chip can make:
- Lensed fibres focus SMF-28’s light to a 2–3 µm waist some micrometres in front of the lens. They match a single silicon taper well but tolerate only about ±0.6 µm of offset (Chapter 17), so they suit the lab more than a package.
- High-NA fibres (UHNA) have a small core and a 3–4 µm mode; a short piece, spliced to SMF-28 with a heat-diffused core, makes an adiabatic fibre-side converter.
- Cleaved SMF-28 needs a 10 µm spot on the chip, which only multi-stage or suspended couplers provide, but it then tolerates a few micrometres of offset.
- Fibre arrays at 127 µm pitch serve many edge couplers at once, and V-grooves etched into the chip itself can hold fibres passively aligned to the couplers.
16.6Edge against grating couplers
Table 16.2. Edge and grating couplers compared. Typical values; see Chapters 15–17 [66].
| Grating coupler | Edge coupler | |
|---|---|---|
| Loss per coupler | 2–5 dB; below 1 dB with bottom reflectors | 0.5–2 dB |
| 1 dB bandwidth | 30–40 nm | more than 100 nm |
| Polarisation | one (two with a 2D grating) | both possible |
| 1 dB lateral tolerance | about ±2.4 µm | ±0.6 µm (lensed fibre) to ±2 µm (large spot) |
| Wafer-scale testing | yes, anywhere on the wafer | only with etched facets and trenches |
| Position on the chip | anywhere | at a prepared edge only |
| Fabrication | shallow etch or SWG; one coupler at a time to optimise | narrow tips, often a second layer, facet preparation |
| Packaging | an angled fibre array above the chip | in-plane fibres or lasers; compact, but a facet to prepare |
Grating couplers win for testing and prototyping; edge couplers win for low loss and wide bandwidth, which is why products, and anything with a laser attached, often use them. Many chips carry both: gratings for wafer-scale test, edges for the package.
Edge couplers for the transceiver of the reference link: a two-stage Si-to-SiN converter matched to a UHNA fibre array, with angled waveguides at the facet, so that the four channels and both directions see low, flat loss. Grating couplers stay on the chip for wafer-scale test.
Reading
- [66] Riccardo Marchetti, Cosimo Lacava, Lee Carroll, Kamil Gradkowski and Paolo Minzioni. “Coupling strategies for silicon photonics integrated chips.” Photonics Research, 7(2):201–239, 2019. Open access.
- [68] Vilson R. Almeida, Roberto R. Panepucci and Michal Lipson. “Nanotaper for compact mode conversion.” Optics Letters, 28(15):1302–1304, 2003.
- [69] T. Shoji, T. Tsuchizawa, T. Watanabe, K. Yamada and H. Morita. “Low loss mode size converter from 0.3 µm square Si wire waveguides to singlemode fibres.” Electronics Letters, 38(25):1669–1670, 2002.
- [70] Pavel Cheben, Jens H. Schmid, Shurui Wang, Dan-Xia Xu, Martin Vachon, Siegfried Janz, Jean Lapointe, Yves Painchaud and Marie-Josée Picard. “Broadband polarization independent nanophotonic coupler for silicon waveguides with ultra-high efficiency.” Optics Express, 23(17):22553–22563, 2015. Open access.
- [1] Lukas Chrostowski and Michael Hochberg. Silicon Photonics Design: From Devices to Systems. Cambridge University Press, 2015. https://www.cambridge.org/core/books/silicon-photonics-design/BF3CF13E8542BCE67FD2BBC7104ECEAB
- Butt-coupling a 500 × 220 nm strip to SMF-28 loses about 20 dB: the mode must be expanded first.
- An inverse taper pushes the mode into the cladding. For TE, a 180 nm tip gives a 2.9 µm spot (92 % to a lensed fibre), 160 nm a 4.7 µm spot (UHNA fibre), and 140 nm about 10 µm.
- Large spots leak through a 2 µm BOX; SMF-sized spots need SiN or polymer stages, SWG or suspended couplers.
- TE and TM expand at different tip widths, and the tip width is a critical, fabrication-sensitive dimension.
- Facet reflections (3.3 % oxide–air) are cut by index matching, coatings and angled waveguides.
- Edge couplers offer low loss, more than 100 nm of bandwidth and both polarisations; grating couplers offer wafer-scale testing and relaxed alignment.