Optical packaging: TO-cans, boxes, lenses, isolators
A laser die is a fleck of semiconductor a few hundred micrometres long; a fibre core is 9 µm across. Getting light from one into the other, keeping it there for twenty years and stopping reflections from coming back is the job of optical packaging — the metal cans, lenses, isolators and receptacles that make up a TOSA or ROSA. Packaging decides most of a module's cost, a good part of its reach and nearly all of its long-term reliability.
Package styles
| Package | What it is | Where used | Cost / performance |
|---|---|---|---|
| TO-can (TO-46, TO-56 and variants) | hermetic metal header with the die on a submount, sealed by a cap with a glass window or a ball lens | almost all SFP/SFP+ optics up to 25G, PON, 1G | cheapest hermetic package; ~10 GHz-class bandwidth limits it near 25–28 Gb/s |
| Box / mini-flat TOSA | ceramic or metal box with RF feed-through, room for TEC, thermistor, isolator and a lens | cooled DFB/EML: DWDM SFP+, 10G ER/ZR, 100G LR4 lanes, 400G FR4/LR4 | more expensive; 50 GHz-class feed-throughs for PAM4 |
| Butterfly (14-pin) | classic telecom laser package with TEC and fibre pigtail | legacy XFP/300-pin DWDM, high-power pumps, ITLA lasers | large, costly — replaced by box TOSAs in pluggables |
| Chip-on-board (COB) | bare dies on the PCB under a lens array, no hermetic package | AOC, SR4/SR8, many DR4/FR4 | lowest cost at volume; relies on moulding/coating for protection — Chip-on-board |
| Silicon-photonics assembly | a photonic chip with modulators, waveguides and detectors; laser attached or flip-chipped; fibre array coupled at edge or through gratings | 100G DR/FR, 400G DR4, coherent | one alignment for many channels — Lasers: SiPh |
Inside a TO-can TOSA
┌── cap with ball lens ──┐
│ ○ lens │ → light to the receptacle / fibre stub
│ ▄ laser die on │
│ AlN submount │ ◄── monitor photodiode behind the laser (back-facet light)
│ ▪ thermistor (opt.) │
├─── header (TO-46) ─────┤
│ pins: laser +/−, MPD, │
│ thermistor, GND │
└────────────────────────┘
- Submount (AlN or Si) spreads heat and carries the wire-bonded die.
- Monitor photodiode (MPD) catches the back-facet emission — a fixed fraction of the output — and feeds the APC loop; DDM Tx power is derived from it, which is why it is an estimate (Accuracy & limits).
- Lens — a ball lens in the cap or a separate aspheric lens focuses the diverging beam (laser output diverges at 20–40°) onto the fibre end; coupling efficiency is typically 30–70 %, i.e. 1.5–5 dB of loss between die and fibre.
- Hermetic seal — resistance welding of the cap and glass-to-metal pin seals keep moisture off the facets; leak-tested with helium in production (Manufacturing & testing).
The ROSA mirrors this: photodiode die on the header, often with the TIA inside the same can to keep the high-impedance node short, a lens in the cap, and for APD receivers a high-voltage bias pin.
Isolators
Light reflected back into a laser — from a connector, a dirty end-face or the fibre itself — destabilises DFB and EML lasers: it raises noise (RIN), causes mode hopping and wavelength jitter and can shift the DDM Tx reading. An optical isolator (a Faraday rotator between two polarisers, 25–35 dB isolation, 0.3–0.5 dB loss) sits between laser and fibre in every 1310/1550 nm module of any reach. VCSEL-based SR modules usually omit it: multi-mode links are less reflection-sensitive and every tenth of a dB counts at 850 nm. Low-cost PON ONUs sometimes omit it too — one reason their behaviour on reflective plant varies (PON problems).
Fibre interface: receptacle or pigtail
| Interface | Construction | Notes |
|---|---|---|
| LC receptacle (SFP, SFP+, QSFP LR4/CWDM4 duplex) | a zirconia split sleeve aligns the patch cord's ferrule with a fibre stub or a lensed port inside the module | the end-face you clean is the stub's face; damage there is permanent |
| SC receptacle (PON, GBIC, 1G) | same with a 2.5 mm sleeve; PON uses APC (8° angle, green) | UPC in an APC port costs decibels and reflects into the OLT (Physical mismatches) |
| MPO/MT receptacle (SR4, DR4, PSM4, SR8) | a lens array couples 4/8/12 beams to an MT ferrule with guide pins; APC for single-mode DR4/DR8 | polarity and pinning matter as much as cleanliness |
| Pigtail | fibre permanently attached to the OSA, terminated on a connector outside | industrial/legacy; no internal stub |
| No connector (AOC, DAC) | fibre ribbon or twinax glued in | Cable internals |
Connector families and polish codes: Connectors & fibre.
Alignment: where the money goes
The die-to-fibre alignment tolerance is about ±1 µm for single-mode. Manufacturers use active alignment — powering the laser while a robot moves the lens or fibre to maximise coupled power, then fixing it by laser welding or UV epoxy. It takes seconds per part and is the cost floor of a single-mode TOSA; multi-mode (50 µm core) tolerates ±10 µm and can be aligned passively, which is why 850 nm optics are cheap and why COB and lens arrays work at 850 nm first (Chip-on-board).
Multi-channel packaging
- Parallel (SR4, DR4): four or eight lasers on one submount or as separate chips, one lens array, one MT ferrule.
- WDM (LR4, CWDM4, FR4): four TOSAs (or one 4-channel array) feeding an internal multiplexer, then a single-fibre port — the filter block is part of the package (WDM inside the module).
- BOSA (BiDi, PON): a laser and a photodiode share one port through a 45° WDM filter; three ports in a PON triplexer.
Packaging and reliability
| Failure | Cause | Signature |
|---|---|---|
| Facet degradation | moisture (lost hermeticity), contamination | bias creeps, power falls — Tx bias & ageing |
| Coupling drift | epoxy creep, thermal cycling, mechanical shock | Tx power slowly down, bias up (APC compensates) |
| Stub end-face damage | scratched by a bad patch cord ferrule or dirt ground in | permanent 1–3 dB loss on that side; cleaning does not restore |
| Isolator/lens fogging | contamination in a non-hermetic design | wavelength-dependent loss, higher RIN |
| Cracked TO-can weld | thermal shock, drop | sudden death or moisture ingress over months |
In CodingBox
Packaging is invisible in the memory map, but its consequences are not: coupling loss shows as low Tx power with normal bias, facet damage as a bias trend, and a damaged stub as a low Rx that cleaning cannot fix. The DDM log over the module's life, together with the bench inspection scope, tells which one you have.