How modules are made and tested

A transceiver's datasheet promises twenty years at 70 °C, a wavelength within a nanometre and a DDM Tx within 2 dB of truth. Whether it delivers depends on a factory process that the buyer never sees: chip screening, sub-micron alignment, calibration on a test station, burn-in and a final coding step that writes the identity you later read on the bench. Knowing the process explains why good modules cost what they do, why cheap ones fail at temperature extremes, and what "calibration constants" and "coding" physically mean.

From wafer to module

laser / PD wafers ──► chip test & screening ──► TOSA / ROSA assembly ──► OSA test
                                                                             │
PCB assembly (SMT: driver, TIA, MCU, DSP, EEPROM) ──► board test ────────────┤
                                                                             ▼
                              module assembly ──► calibration & tuning ──► burn-in ──► final test ──► coding & label ──► QA sample ──► ship

Chips

Optical sub-assembly

Die attach on a submount, wire bonding, then the step that sets the price: active alignment — with the laser powered, a lens and/or fibre stub are moved in sub-micron steps to maximise coupled power, then fixed by laser welding or UV epoxy. Hermetic packages are cap-welded and leak-tested; the finished OSA is measured for launch power, extinction ratio, eye shape and, on receivers, responsivity and sensitivity (Optical packaging). COB engines replace this with pick-and-place of bare dies under a moulded lens array (Chip-on-board).

Calibration and tuning

On a test station with a temperature chamber, an optical power meter, a BER tester and a reference receiver, the firmware is taught what its raw numbers mean:

This is where the DDM accuracy of ±2–3 dB comes from — and why a module calibrated only at room temperature reads wrong at 65 °C (Accuracy & limits, Calibration).

Tests a module must pass

Qualification of the design (not every unit) follows Telcordia GR-468-CORE: temperature cycling, damp heat, mechanical shock and vibration, accelerated ageing on a sample to project the lifetime that the datasheet then quotes.

Burn-in and screening

Finished modules run for hours to days at elevated temperature under traffic; DDM is logged; units whose bias, power or errors drift are pulled. This step is expensive and is the first thing skipped by low-cost producers — which is why field failure statistics of cheap optics look fine for a year and then spike (Failures).

Coding: the last step

Only now is the identity written: vendor name, part number, serial number, date code, compliance codes, lengths, wavelength, options, checksums — either the manufacturer's own or the brand of the customer the batch is made for. Contract manufacturers produce one physical module and code it under many names; OEM-branded modules get vendor signatures in the vendor area. This is the same operation CodingBox performs on the bench (EEPROM recoding, Vendor fields) — the factory simply does it with the calibration data already in place.

What separates grades

None of this is visible in the identity bytes; it shows in DDM behaviour over temperature and in failure rates after a year.

In CodingBox

Incoming inspection on the bench repeats a small part of the factory's final test: read identity and checksums, verify DDM values are live and plausible against typical values, record the baseline in the code database. A batch whose Tx power or bias is identical across units, or whose temperature never changes, failed the factory step it claims to have passed.