20 % failure rate down to 0 %: how a forgotten 24-hour bake can save your PCBs

20 % failure rate. Out of ideas. Everything tried.

The phone call

It is evening when the head of engineering at a foreign brushless motor controller manufacturer with its own in-house production calls. The next morning he has to face a crisis meeting with the board. About two thirds of the defective devices fail during the first burn-in, one third in the field at the customer. Annual volume in the hundreds of thousands.

At that volume that is 20,000 failures per year: material, assembly, test, shipping, return material authorisation (RMA) logistics, reputation damage. In this specific case they had reinforced via structures, screened component lots, compared suppliers, profiled the reflow oven. Weeks of work, small improvements, no breakthrough. And the deadline for the next morning’s meeting is closing in.

The obvious hypothesis and why it was wrong

The design is compact. The aluminium heatsink is shaped as a frame around the board, and the PCB is bonded into that frame with thermal adhesive over the entire contact area, no screws, no clamps. The base material of the PCB is FR4, the common glass fibre epoxy laminate. FR4 itself is a poor thermal conductor (about 0.3 W/mK versus roughly 200 W/mK for aluminium), so heat removal has to go through the bonded contact surface to the frame. High current density, fast switching edges, tight thermal coupling. On top of that, fine structures sit in the current sensing area: small vias and tight trace spacing around the shunts and current sensor paths, where accuracy fights against layout density. The piquant detail: these fine vias sit right in the centre of the board, exactly where the thermomechanical forces of an all-around bonded frame would presumably converge. This is the safety critical zone: without precise current measurement, neither field oriented control nor overcurrent shutdown works. Even small leakage currents or drifting insulation resistances destabilise the control loop or trigger false shutdowns.

The obvious suspicion: thermomechanical stress. Different coefficients of thermal expansion between FR4 (~14 ppm/K in plane, 300 ppm/K in Z direction above the glass transition temperature T_g) and the aluminium frame (~24 ppm/K), with adhesive in between. With every thermal cycle the materials pull against each other. Crack formation, micro fractures at the vias, solder fatigue, the usual textbook.

The first countermeasure was the textbook answer: reinforced vias. Larger drill diameters in the critical area, thicker copper plating, in places filled and capped vias under the power pads. Textbook answer to textbook hypothesis. It moved the needle, but not far. The failure rate dropped by a few percentage points and stayed firmly in the double digits. That was the hint that the diagnosis could not be right.

We worked with this hypothesis for the first few hours. Until we sorted the failure statistics by time of failure.

What the statistics actually showed

On closer inspection it was striking that a significant share of the defects occurred before the first burn-in, right after reflow in the electrical end of line test. Thermomechanical aging cannot have damaged a fresh board. Something was happening during the assembly process itself.

That shifted the suspicion abruptly: the problem was not in operation, it was in processing.

The decisive finding: several suppliers, same rate

What made the analysis particularly revealing: the customer was sourcing their boards, as is common at that volume, from several PCB houses in parallel. Different plants, different countries, different processes, different surface finish batches, different supply chains.

The natural expectation would be: one supplier has worse quality, the failures cluster there. That is how you normally do it. Sort by source, find the culprit, swap suppliers, problem solved.

We did exactly that. The result was:

All suppliers showed practically the same failure rate, around 20 %, with only minor scatter between them.

That was the real eye opener. When three or four independent PCB manufacturers with different processes all come in at similar numbers, the problem is highly unlikely to sit at the PCB manufacturer. It sits in something common to all deliveries, and that is usually what happens after the PCB delivery: transport, storage, assembly preparation. In other words: the process on their own assembly line.

That insight flipped the case in less than half an hour of diagnostic discussion. “We are looking for the bad supplier” became “we have a systemic process defect in our own house”.

The real cause: moisture in the laminate

FR4 is hygroscopic. The glass fibre epoxy composite pulls water out of ordinary shop air, through open drill edges, through vias, through micro cracks in the solder mask. At 23 °C and 50 % RH, an unprotected board reaches its saturation level in two to three days. The Würth application note on the physics of moisture in PCBs (German) lays out the diffusion mechanics and the typical saturation curves cleanly, including the fact that drying out takes orders of magnitude longer than soaking up.

What that means in reflow is a simple physics calculation:

• At ~100 °C, trapped water starts to expand.
• At 220 to 260 °C in lead-free reflow it generates more than 25 bar of vapour pressure in enclosed cavities.
• When the laminate can no longer hold the pressure, micro delaminations form, the popcorn effect.

The consequences appear partly immediately (blistering, visible delamination, solder bridges from shifted components) and partly weeks or months later in the field:

• Conductive Anodic Filament (CAF): copper ions migrate along the glass fibre to resin interface through residual water, driven by DC bias, and slowly form a conductive path. Result: gradual insulation resistance decay, intermittent shorts, eventually total failure.
• Crack growth at via barrels: micro delamination grows further under temperature cycles.
• Corrosion under the solder mask: trapped moisture attacks copper traces.

This explains both failure phases: the immediate defects at burn-in (delamination, popcorn) and the later field failures (CAF, crack growth).

The fix: trivial and effective

Before each assembly run we introduced a drying regime:

• 24 hours at 105 °C in a convection oven.
• After drying: assembly within max. 4 hours, or storage in a dry cabinet at ≤ 10 % RH.
• Boards delivered in sealed MBB (Moisture Barrier Bag) with desiccant from the manufacturer.

Result: failure rate dropped from 20 % to practically 0 %.

Not 19 %. Not 5 %. To zero point something. The entire phenomenon, both burn-in failures and field returns, disappeared as soon as the boards entered the assembly system dry.

Sealed does not mean dry

A sealed MBB pack does not automatically mean dry boards. At many fabs the bare boards sit between final inspection and shipping for hours or days in normal shop air before they get bagged. If they are already saturated at that point, the bag only preserves the moisture content in the laminate. At the assembler the board then goes straight into the reflow oven without a drying run. In regions with high uncontrolled shop floor humidity that saturation window shortens further.

Practical consequence: a Humidity Indicator Card (HIC) in the bag is mandatory, not a comfort feature. At goods receipt, read the HIC, log the value, and treat any delivery above the threshold like unpackaged stock: dry cabinet, drying run before assembly.

Why this gets overlooked so often

The recommendation is in IPC-1601 (“Printed Board Handling and Storage Guidelines”) and has been state of the art for years. Still, this is what happens in reality:

1. No requirement, no drying. PCB manufacturers typically do not specify drying, and assemblers do not bake what they are not required to. It works without in most cases, and a drying run costs a full day of oven occupancy.

2. Failures appear with delay. If a third of the defective boards only fail in the field weeks later, the link to the assembly batch is statistically blurred. The correlation to board storage time even more so.

3. Other hypotheses sound sexier. Saying “thermomechanical stress” sounds like deep expertise. Asking “did you bake the boards?” sounds like cleaning staff. That shifts the diagnostic priority.

What IPC-1601 actually recommends

The standard scales drying with storage time and packaging state:

Storage time since manufacture	Packaging state	Recommended bake
< 2 months, MBB sealed	at < 30 °C / < 60 % RH	1 h at 120 ± 5 °C
2 to 6 months	normal shop floor	2 h at 120 ± 5 °C
6 to 12 months	normal shop floor	4 h at 120 ± 5 °C

Important: the standard gives a lower bound. In practice, critical designs (multilayer ≥ 6 layers, dense via patterns, compact power designs) benefit from significantly longer drying. Our regime of 24 h at 105 °C is conservative but low risk: longer drying at lower temperature reduces thermal pre-stress and is compatible with most surface finishes (HASL, ENIG, Immersion Tin).

OSP exception: Organic Solderability Preservatives degrade above 105 °C. Anyone using OSP coated boards has to keep drying shorter (≤ 2 h) or at lower temperature, or better switch to a different finish.

Recommended procedure

• Procurement: explicitly order MBB packaging with a Humidity Indicator Card (HIC) and desiccant, require the manufacturing date on every carton.
• Goods receipt: read and log the HIC, route any conspicuous lot directly into the dry cabinet.
• Before assembly: apply the IPC-1601 drying time, max 4 hours between oven and reflow, return leftovers to MBB or dry cabinet.

What this case means for other manufacturers

If you have a compact power design, multilayer, with thermal bonding, and your failure rate is above 5 %, then before you investigate thermomechanical stress, electromagnetic compatibility (EMC), or component quality, clarify this question first:

“Are our boards dried per IPC-1601 before assembly, and can we prove it with an oven log?”

As long as the delivered quality is good, there is no need to dictate the process to the assembler. When problems appear, though, it pays to ask for the logs — drying oven, reflow profile, end of line test. Written specifications on drying, storage and test depth help in such cases, but they make production more expensive. A pragmatic middle ground: trust, request oven logs as spot checks, and when anomalies show up, walk through the process together with the assembler rather than dictating individual steps from the outside.

Why an outside view often helps

The team had collected the statistics and taken countermeasures, but was already deep in the forest seeing only the trees. In such situations it can help to bring in an external, unburdened person. In this case it was enough to sort the existing failure data by a different axis: by time of failure instead of by supplier. The table already existed, it had simply never been read through that lens.

Sources and further reading

• IPC-1601A — Printed Board Handling and Storage Guidelines (IPC, 2014)
• Würth Elektronik — Application Note: Physics of Moisture in PCBs (German)
• Sierra Circuits — IPC-1601A: PCB Handling and Storage Guidelines
• PCBSync — PCB Baking: Complete Guide to Temperature, Time & IPC-1601 Standards
• ZVEI — Recommendations/Comments on IPC-1601 (2021)
• Saturn Electronics — PCB Pre-Baking and Delamination
• ResearchGate — Analysis of Catastrophic Field Failures Due to CAF Formation
• Isola — Conductive Anodic Filament Growth Failure (Tarun Amla)
• IPC-9691 — Guideline for the Use of the IST Test for Conductor and Insulator Reliability
• IPC/JEDEC J-STD-033 — Handling, Packing, Shipping and Use of Moisture/Reflow Sensitive Surface Mount Devices

Keywords: IPC-1601, PCB baking requirement, FR4 moisture, popcorn effect, Conductive Anodic Filament, CAF, MBB packaging, reduce PCB failure rate, reflow delamination, assembly failure consulting, electronics manufacturing Switzerland, swissEmbedded consulting