HOME / BLOG / BURN-IN TEST PROGRAMME
QUALITY · SEP 19, 2026
What a 100% Burn-In Programme for Rack Power Involves
Every power vendor claims burn-in. Far fewer can tell you the chamber setpoints, the load level, the exit criterion, and what happens to a unit that fails at hour two. This is the programme as it runs in our aging room, written so a buyer can audit it line by line — and the report structure you should insist on receiving with the shipment.
WHY BURN-IN EXISTS
It is a screen for infant mortality, not a courtesy.
Defects that only appear under stress and time
A cold functional test passes a unit that will fail in a rack three weeks later: a marginal solder joint on a high-current path, a connector with insufficient normal force, a controller that drifts as it heats, a fan that stalls at a specific duty cycle. None of those show up on a bench test at ambient with a light load. They show up when the unit runs at rated output, hot, for hours — which is what the aging room is for.
Why the coverage is 100% rather than sampled
Sampling answers a statistics question about a lot. Burn-in answers a contractual one about a unit: the serial in crate 14 either ran at full load or it did not. That is why the rule on our floor is that no unit skips the aging room, why CRPS supplies and BBU modules both pass through it, and why a lot cannot ship on the strength of its sample — the exit criterion is zero failures at the unit level, not an acceptable quality level.
CHAMBER PROFILE
The profile is a program decision — get it in writing.
There is no universal burn-in recipe, because a 5.5 kW Titanium shelf and a 51.2V LFP battery module stress differently. What must be identical across every program is that each of these parameters is defined, recorded and reproducible. Ask for them in the RFQ reply; a vendor who cannot state them is not running a programme, they are running a room.
| Profile parameter | Why it is a program decision | Evidence it must produce |
|---|---|---|
| Soak temperature | Set by the components with the least thermal margin — electrolytic capacitors and controller ICs, not the power stage | Chamber setpoint and logged air temperature per slot |
| Load level | Full rated output screens current-path defects; partial load hides them | Per-unit output current and voltage trace |
| Duration and soak time | Long enough for thermal equilibrium plus the drift window the failure modes need to appear | Start and end timestamps per serial |
| Power cycling | Turn-on inrush and thermal cycling find joints a steady soak never touches | Cycle count and per-cycle outcome |
| Telemetry polling | PMBus, SNMP or CAN health must be checked while hot, not only at ambient | Poll log with fault and warning events |
| Exit criterion | Zero failures at full load — a de-rated "pass with observation" is not an exit criterion | Signed exit record keyed to the serial |
LOAD BANKING
What "full load" actually means on the bench.
Bank type decides what you screen
Resistive banks are simple and cheap, but their current falls as the supply's voltage sags — so a resistive bank under-tests a unit whose regulation is drifting. Programmable electronic loads hold constant current or constant power regardless of voltage, which is the only way to keep a unit at true rated output for hours and see the failure that hides behind a light bench load. For multi-rail supplies the load matters per rail: a 54V shelf tested only on its main output never stresses its auxiliary rails.
Where the heat goes
Load banking is a thermal engineering problem before it is an electrical one. Forty 5.5 kW modules at full load is 220 kW of output; at 96% Titanium efficiency those units shed roughly 9 kW into the chamber as heat, which the aging room's air handling has to remove continuously. If the room cannot hold its setpoint, the profile quietly becomes something else — which is why chamber capacity, airflow and the temperature log matter as much as the load racks.
Regenerative loading is the economics question
Every kilowatt burned in a load bank is a kilowatt bought from the utility and then thrown away as heat. Regenerative electronic loads push the energy back to the mains instead, which changes the electricity bill of a 100% burn-in programme substantially at scale. When you audit a factory, this is a fair question: not whether they burn in, but whether the burn-in room is designed to do it for years without becoming a cost argument that quietly shortens the duration.
FAILURE ESCAPES
Where defects get past a test programme.
Escape rate is the only honest measure of a quality system. It is the count of defects that reached a customer despite every gate — and each escape has a specific gate that should have caught it.
| Failure mode | Gate that catches it | What an escape looks like on site |
|---|---|---|
| Voided BGA or high-current joint | X-Ray on sampled lots plus 100% of battery welds | Intermittent dropout under thermal cycling, often weeks after install |
| Wrong or reversed component | 100% AOI against IPC-A-610 Class III libraries | Board works in one orientation of the rack, fails in the other |
| Insufficient paste deposit | In-line 3D SPI on 100% of deposits | Weak joint that survives functional test and opens under load |
| Dielectric weakness | 100% hi-pot to IEC 62368-1 limits with ground-bond check | Safety event at the customer, not a performance complaint |
| Metering drift on a PDU | End-of-line metering check against the ±1% billing-grade claim | Billing disputes that surface months later, not a failed unit |
| Telemetry fault under heat | Functional test with PMBus or SNMP polling, repeated hot during burn-in | A rack that is invisible to DCIM exactly when it matters |
Measure it two ways
Internally, escape shows up as first-pass yield and as the Pareto of burn-in failures by station: a spike in one failure mode after a component lot changes tells you more than the aggregate number. Externally, it shows up as dead-on-arrival rate and early-life returns at the customer site. A programme that reports only its burn-in pass rate is reporting half the picture.
What a good number looks like
Through a documented 18,000-piece CRPS ramp, 100% burn-in coverage held first-pass yield at 99.2%. The number is not the achievement — the coverage rule behind it is. Yield measured over the units that were screened says nothing about the units that were not, which is precisely why 100% coverage and a zero-failure exit criterion matter more than any headline percentage.
THE TEST REPORT
What should be in the dossier you receive.
A burn-in certificate with a signature and no data is a receipt, not evidence. Here is the structure we assemble per shipment, and the structure worth writing into your purchase order.
| Report section | Contents | Keyed to |
|---|---|---|
| Unit identity | Model, revision, serial range, firmware and configuration version | Serial number |
| Incoming material record | IQC verdict on critical components and cell lots | Component lot |
| Process record | SPI deposit results, AOI verdict, X-Ray images on sampled and battery-weld units | Panel and lot |
| Safety tests | Hi-pot dielectric withstand and ground-bond resistance against IEC 62368-1 limits, with leakage readings | Serial number |
| Functional test | Rail voltages under load, protection trip points, telemetry polling, metering accuracy check | Serial number |
| Burn-in record | Chamber setpoints, load level, duration, cycle count, exit verdict, captured trace | Serial number and chamber slot |
| Rework history | Any failure, the diagnosis, the rework performed and the retest result | Serial number |
| Lot summary | Units in, first-pass yield, failure Pareto, corrective actions raised | Lot number |
| Certificates | ISO 9001 system certificate and CE / RoHS declarations, shared under NDA with qualified buyers | Program |
AUDITING THE PROGRAMME
Six things to do on a factory visit.
- Pick a serial and walk it backwards. Hand the quality team a number from a finished crate and ask for its SPI, AOI, X-Ray, hi-pot, functional and burn-in records. A working system answers in minutes; a filing system answers next week.
- Stand at the aging room exit. Ask what unlocks the door — on our floor the answer is a pass verdict from the end-of-line fixture, because the fixture sequence is enforced by the station, so no unit can reach the room untested.
- Ask which units skip burn-in. The only correct answer is none. If the answer is "the low-wattage ones" or "by lot", the coverage claim on the website is not the coverage on the floor.
- Ask what happens to a failure at hour two. Failures must lock the serial into a rework loop with the failure log attached and a retest before the unit returns to the line — not a manual note and a shrug.
- Check the chamber log, not the certificate. Ask to see the last week of temperature and load traces for a specific slot. Certificates are generated afterwards; traces are recorded during.
- Schedule third-party inspection and use it. SGS, BV, TUV or your own agency at pack-out, plus a live video walk-through of the burn-in exit for remote programs. Per-serial QC records travel with the shipment either way.
RELATED SYSTEMS
The hardware behind this brief.
CRPS Server PSU
550W–5.5kW+ modules, 96% Titanium, 54V rails — hi-pot, functional test and 100% full-load burn-in on every serial.
Explore CRPS →BBU Battery Backup
51.2V LFP modules with X-Ray on every battery weld, formation cycling, capacity grading and UN38.3 export documentation.
Explore BBU →Quality & Testing
The seven-gate sequence — IQC, SPI, AOI, X-Ray, hi-pot, functional test, 100% burn-in — and the QC dossier that ships with it.
See the gates →
TEST COVERAGE
