Burn-In and ATE: How to Read a Power Supply Test Record

What ATE Proves and What Burn-In Proves

Automated electrical test is a measurement. It puts a unit at defined operating points and compares what it does against a limit file derived from the datasheet. It answers the question "does this specific unit meet its specification, right now, at these conditions?" It is a snapshot, it is per-unit, and it is only as good as the limits it is testing against.

Burn-in is a screen. It operates a unit under elevated stress for a specified cycle so that defects which would have failed in the first days of field service fail in the factory instead. It answers a different question: "has this unit survived the early-life failure band?" It cannot prove a unit meets its efficiency specification, and a well-designed ATE program cannot tell you whether a marginal solder joint will open up after forty thermal cycles.

Manufacturers who only do one of the two are telling you something. ATE alone produces units that pass on the bench and fail in week one, because infant mortality is a time-and-stress phenomenon that a five-minute measurement cannot reach. Burn-in alone produces units that are thermally stable and out of specification, because nothing measured the curve.

The ATE Sequence, Unit by Unit

The defensible standard is 100% test, not sample test. On our lines every unit — not a sample from the lot — goes across automated test equipment, and the sequence verifies the numbers on the datasheet against the unit standing in front of the fixture:

• Efficiency at the gate points. A sweep across 10%, 50% and 100% load compared against the 80 PLUS Titanium floors of 90%, 96% and 91%, measured at both nominal and low line voltage so the curve is characterised rather than spot-checked.
• Output regulation on every rail, including the standby rail, at load points that bracket the expected operating band.
• Ripple and noise on the main output, measured with a defined bandwidth and probe technique so the number is comparable between units and between shipments.
• Protection trip points: OVP, UVP, OCP, OTP and fan-fault, each driven to trip and recorded as a measured value rather than a pass/fail flag.
• Signalling: power-good and PS-kill behaviour, including recovery after a kill command, because hot-swap sequencing in a multi-module array depends on it.
• PMBus 1.2 readback: the voltage, current, temperature and fan-speed registers are read across the interface before the unit is released, which verifies the digital path as well as the analogue one.

Recording measured values rather than pass/fail flags is what turns ATE from a gate into a data source. A lot where every unit's fan current sits 8% above the previous lot is a process signal, and it is invisible if the record only says "pass".

Burn-In Chamber Profiles

Infant-mortality failures cluster in the first hours of stressed operation, which is exactly when you do not want them appearing in a customer rack mid-production-run. Burn-in compresses that window. Our aging rooms run production units under elevated temperature and controlled load for a specified cycle, with monitoring across the population so a drift cluster triggers a process review upstream at SMT or DIP rather than a rework loop downstream.

The profile itself is a specification item, which is why our published production lead time of 6–10 weeks is stated as depending on the cycle you specify. These are the variables to agree, and what each one is actually for:

Profile variable	What to agree	What it screens for
Chamber temperature	Elevated above the 40°C MTBF reference point, held within a stated tolerance and logged	Accelerates marginal joints and component defects without pushing parts past their own ratings
Load level	Typically 80–100% of rated output, applied through a real electronic load rather than a resistive dummy	Exercises magnetics, rectifiers and output capacitors at genuine electrical stress
Duration	Specified per project, and stated in the purchase specification rather than assumed	Reaches far enough into the early-life band to convert field returns into scrap
Power cycling	Whether thermal and inrush cycling is included inside the cycle	Connector, solder and relay defects that steady-state operation leaves untouched
Population monitoring	Continuous per-unit output, temperature and fan telemetry across the batch	Turns pass/fail screening into a trend you can act on before the next lot is built
Exit gate	Cool-down followed by a re-test at ATE before packing	Catches anything the aging cycle degraded — the whole point of screening

The exit gate is the item most often missing. A burn-in cycle that is not followed by a re-test is a delay, not a screen: it ages the product and then ships it without asking whether the aging changed anything. On our lines units exit burn-in into a cool-down and re-test gate at ATE before they are allowed to move to packing.

What the Test Record Should Contain

A test report that says "passed" is not evidence. What you want is a per-serial record that lets you reconstruct the unit's history months later, and that means it should carry:

• Unit serial number, build date and manufacturing lot, with the work order it was built against.
• ATE station identity, test program revision and limit-file revision, with the timestamp of each test pass.
• Measured efficiency at 10%, 50% and 100% load at line and low line, next to the 90/96/91% Titanium gate values.
• Regulation, ripple and noise results per rail, as values rather than verdicts.
• Measured protection trip points for OVP, UVP, OCP, OTP and fan-fault.
• PMBus readback results and the firmware version reported by the unit.
• Burn-in batch identifier, chamber setpoint, load level, cycle duration and the monitoring data for the batch the unit ran in.
• Post-aging re-test result, which is the record that proves the screen happened before shipment rather than after.
• Component lot linkage — at minimum the capacitor and fan date codes — so a field event can be scoped to a date range instead of a shipment.
• Outgoing quality release, including serial continuity and label and rating accuracy for the carton.

That chain is what makes an RMA investigable. When a unit comes back from the field, the serial number should lead to the component lots, the ATE results at every gate, the aging batch data and the release record. If a supplier cannot produce that linkage, failure analysis is guesswork, and so is any claim about how many other units share the same exposure.

How to Audit a Supplier's Line

Factory audits are frequently scheduled as tours. Run them as verifications instead. The four stations below are what a visitor should walk through, in order, and each has a question attached that a tour will not answer on its own.

• Assembly. Confirm solder-paste inspection precedes reflow and AOI follows placement, and that through-hole joints in the power stage and heatsink torque are checked as specified items rather than by eye. Ask to see the line-side check that confirms golden-finger dimensions and mechanical fit of the 73.5 × 40 × 185 mm envelope.
• ATE. Ask for the test program and limit-file revision number, then ask to see a unit deliberately set up to fail and watch what happens to it. Confirm that 100% of units are tested rather than a sample, and that calibration certificates for the power analyser and electronic load are current.
• Burn-in. Look at what the load actually is. A rack of resistors is not a load bank in any meaningful sense: it draws current but does not exercise the regulation loop, and it cannot produce the telemetry a monitored screen depends on. Confirm the chamber's setpoint log and ask what triggers an alarm during a run.
• Re-test and safety. Verify the post-aging gate exists as a physical station with its own records, not as a note in a procedure. Then check the compliance lab's scope against the certification matrix you are buying: IEC/UL/EN 62368-1 with CB, CE, UKCA, FCC Part 15 Subpart B, RoHS and REACH.
• Traceability. Pick a serial number at random from finished goods and ask them to walk it back. If the round trip takes an afternoon, the system is a filing cabinet. If it takes minutes, you are looking at a production record.

One more audit item that costs nothing: ask whether the samples you receive carry the same records as production. First articles that arrive with a full per-serial record and production units that arrive with a carton label are a common and expensive gap, and it usually reflects a screening step that quietly disappeared when volumes rose.

Ordering Test Data With Your First Article

Put the deliverables in the purchase order, not in a follow-up email. Specify 100% ATE with measured values at the Titanium gate points, the burn-in profile and its duration, the post-aging re-test, and a per-serial record in a machine-readable format so your own quality system can trend it. Add failure-analysis turnaround and the component-lot linkage depth you need for a field campaign, and require the firmware version in the shipment documentation.

Then connect the record to the reliability claim. Our platforms are engineered to a 250,000-hour MTBF target at 40°C per Telcordia SR-332, with continuous output from −5°C to +55°C ambient, and the test chain is what makes that a statement about shipped product rather than about a design file. Every production order can include burn-in screening records and ATE test reports, samples of our production 550 W–2400 W range ship in 2–4 weeks with full test data, and buyers are welcome to audit the line — the stations above are exactly what a visit walks through.

Related Reading