The lab certifies a design. The factory line has to certify every unit. We’ve watched devices pass final inspection and fail in a customer’s home because the line didn’t test what the certification lab does. End-of-line (EOL) testing is that last gate - it has to catch what the lab catches, plus the failure modes the lab never sees.

Must-test: certification-critical items

Test	Pass criterion	Skip risk
Commissioning loop	5/5 cycles, zero failures	Intermittent field failures
RF calibration	Within 1.5 dB of cert values	Poor range, marginal commissioning
DAC chain validation	100% per unit, no sampling	Batch commissioning failure
OTA update test	Completes + recommissions	Field update failures

Commissioning loop details

Run a full commissioning cycle against a reference controller, decommission, repeat five times. Pass is zero failures. A device that fails one in five cycles on the line will fail in the field. Use a Matter-certified controller (Apple HomePod, Google Nest Hub, or a chip-tool instance), not your development controller.

DAC chain validation

One script, every device, no sampling. Read the DAC from secure storage, validate the chain against the CSA PAA trust store , confirm the VID/PID matches the CD and the firmware basic information cluster. Fold it into the provisioning script’s output.

OTA update test per batch

Trigger an OTA on a sample from each production batch, confirm it completes and the device recommissions. The lab will do this; so should you. Budget 5-10 minutes per batch.

Can-skip: what the lab doesn’t test

Full cluster attribute walk. The lab tests mandatory clusters. You don’t need to read every optional attribute on every unit.
Multi-controller stress testing. The lab validates against declared controllers. Production tests against one reference controller.
Interop with every border router vendor. Test against one Thread border router or one Wi-Fi AP. Save the full interop matrix for pre-production validation.

RF calibration specifics

Thread and Wi-Fi devices both need calibration. Skip it and you ship devices with poor range and unreliable commissioning - a failure that doesn’t show up on a bench test with the device 30cm from the controller. (Calibration is production best practice and radio-conformance-driven; it’s not a Matter-attestation requirement - the spec doesn’t check your radio calibration.)

Thread devices (Silicon Labs EFR32)

Use the RAIL test library to calibrate the HF crystal oscillator and PA ramp. Skipping crystal calibration on a 40 ppm crystal on a temperature-uncontrolled factory floor produces devices that fail to associate with a border router at 20 meters. The RAIL library supports automated factory calibration - integrate it into your test fixture.

Wi-Fi devices (Espressif ESP32-C6)

Calibrate TX power per band (2.4 GHz) and validate using a spectrum analyzer or a golden-node RSSI comparison setup. Know the tooling split: esptool.py flashes the RF-test firmware and espefuse burns the calibration eFuses, but the per-unit RF calibration itself is a runtime PHY procedure (calibration data lives in NVS) - integrate it into the test fixture’s boot sequence rather than treating it as a flashing step.

Nordic nRF53 (Thread + BLE)

Use Nordic’s Direct Test Mode (DTM) for BLE RF testing - automated TX power and sensitivity measurement over UART (the nRF5340 DTM sample implements Bluetooth Core Vol 6 Part F), suitable for production test fixtures. Note: DTM is a Bluetooth test mode - for the 802.15.4/Thread radio, use the radio_test sample instead.

Time budget

Budget 30-45 seconds per unit for full RF calibration on a fixture equipped with a controlled RF test chamber. This includes Thread and BLE calibration for dual-radio devices.

Test fixture design

A production test fixture for Matter devices needs:

Controlled RF environment - shielded enclosure or calibrated test chamber to ensure repeatable measurements
Production interface - SWD, UART, or production BLE service for communicating with the device under test
Pass/fail indicator - green/red LED visible to the operator, driven by the test script’s exit code
Serial number tracking - fixture reads or receives the device’s serial number and logs results against it
Automated commissioning - reference controller integrated into the fixture, not a manual step

FAQ

How many commissioning cycles should end-of-line testing run?

Five complete commission-decommission cycles per unit. This catches intermittent failures that a single cycle misses. If a device passes 4/5 cycles, it still fails - intermittent commissioning issues are the leading cause of field returns for Matter devices.

What RF calibration equipment do I need?

A shielded test chamber or Faraday cage, a calibrated RF power meter or spectrum analyzer, and a golden-node reference device for RSSI comparison. For Thread devices, the Silicon Labs RAIL test library automates calibration. For Wi-Fi, Espressif’s production calibration tools handle TX power per band.

Can I skip RF calibration if my devices pass commissioning on the bench?

No. Bench commissioning at 30cm doesn’t validate RF at the distances and interference of a real home. Devices that skip calibration fail at range - weak signal, dropped connections, long commissioning times - rather than failing outright. The symptom is units that won’t commission at 10+ meters or through walls.

Part of the Shipping Matter series. See also: Factory Flashing at Scale , DAC Provisioning Pipeline .

← All posts