Why Power Supplies Fail in the Field

Key points

• The electrolytic capacitor usually dies before the semiconductor does
• Humidity failures need H3TRB data, not HTRB — they are different failure modes
• Latent defects pass final test and appear at 200–500 temperature cycles
• An unnotified die source change is the most common cause of an unreproducible failure

Failure is rarely about the headline rating

Ask a design engineer why a power supply failed and the answer is usually “the MOSFET was too small”. Ask a service engineer and the answer is usually different: a capacitor dried out, a solder joint cracked, a track corroded, or a replacement part behaved differently from the original.

The six causes below account for most field failures in switching power supplies from 5 W to 3 kW. None of them is predicted by the current or voltage rating on the front page of a datasheet.

1. Electrolytic capacitor wear

The output capacitor and the primary bulk capacitor set the service life of most power supplies. Electrolytic capacitor life roughly halves for every 10 °C rise in core temperature, and the core runs hotter than the case.

A design that puts a 105 °C-rated capacitor next to a 3 W snubber resistor and assumes 50,000 hours has done the arithmetic on ambient rather than on core temperature. The cheap fix is a higher-temperature grade or a longer-life series; the expensive fix is a warranty campaign.

What predicts it: ripple current against the capacitor’s rating, core temperature measurement inside the enclosure, and a life calculation at the real ambient — not at 25 °C.

2. Thermal cycling of joints and bonds

Every power cycle expands and contracts the die, the die attach, the wire bond and the mould compound at different rates. Over thousands of cycles, the weakest interface cracks.

In a semiconductor, this shows up as die attach voiding or wire-bond lift-off — both invisible at final test. In a power supply, it shows up as a joint that goes open after two years of daily use.

What predicts it: temperature cycling data (TCT) with the cycle count and temperature extremes stated, and a delta-T calculation for your actual load profile. A part cycled from −40 to 125 °C for 500 cycles is not evidence for a product that cycles 0 to 80 °C ten times a day for ten years.

3. Humidity-driven leakage and corrosion

Moisture plus bias plus ionic contamination equals corrosion and surface leakage, usually at the passivation edge or between package pins. The failure is gradual: leakage rises, a parameter drifts, a protection circuit trips.

This is the dominant failure mode for equipment in humid, non-airconditioned enclosures — telecom cabinets, outdoor lighting, solar inverters, agricultural equipment.

What predicts it: H3TRB data (85 °C / 85% RH with bias), plus conformal coating and enclosure design. HTRB does not predict it; the two tests stress different mechanisms.

4. Overvoltage from leakage inductance

Turn-off di/dt multiplied by loop inductance produces a voltage spike on top of the bus voltage. A 650 V device in a 375 V bus with a slow clamp can see 700 V at full load and low temperature, and the failure occurs on a cold morning when the design was validated at room temperature.

What predicts it: measured drain-source waveform at worst-case load, lowest temperature and highest input voltage, with the actual production layout rather than the lab prototype. Snubber design and layout matter more than the device’s BVDSS.

5. Under-rated inrush and surge

Bridge rectifiers and input fuses are commonly sized on average current while the real stress is the repetitive inrush into the bulk capacitor at every power-on. A 2 A bridge feeding a 1 A load can be at its thermal limit because the RMS current through it is 2 to 2.5 times the DC output.

What predicts it: RMS current measurement through the bridge, inrush waveform, and the number of power cycles per day. IFSM is a single-event rating and says nothing about repetition.

6. Unnotified process or die source change

A supplier changes die source, mould compound or assembly site to protect margin or lead time. The part number stays the same. Behaviour changes slightly — threshold voltage, switching speed, body diode recovery. A design that was marginal now fails, and the failure cannot be reproduced because the original sample is gone.

What predicts it: a documented PCN process, lot traceability from foundry wafer to carton, and a supplier willing to put both in writing. This is a procurement question rather than an engineering one, and it is the reason second-source qualification exists.

What evidence to demand, by application

Application	Minimum reliability evidence	Additional
Consumer adapter, indoor	HTRB, temperature cycling	Solderability, lot test record
Industrial power supply	HTRB, TCT, thermal shock	H3TRB if non-condensing is not guaranteed
Outdoor lighting, telecom	H3TRB, TCT	Thermal shock, isolation test
Solar, energy storage	H3TRB, TCT, 175 °C storage	PCN process, 20-year derating calculation
Welding, motor drive	TCT, IFSM data, avalanche	Soft recovery characterisation

The uncomfortable conclusion

For most power supply failures, the semiconductor is the suspect and the capacitor, the layout or the process change is the cause. Buying a device with a bigger rating does not fix a humidity failure, a cracked joint or a clamp that was never tuned.

The productive approach is to name the failure mode first, then ask the supplier for the test that predicts it. Our reliability testing page lists the nine test types we run in-house and what each one does and does not prove; the traceability page describes what a lot record can and cannot tell you after a field return.

This article describes general manufacturing practice for machined and fabricated components. It is not a substitute for a drawing-specific engineering review — tolerances, material selection and process capability depend on the individual part.