Thermal Stress Screening and Burn-In for Assembled Boards

Some failures appear in the first few hours of operation rather than after years in the field. Thermal stress screening exists to force those early failures to happen in the factory, where they can be analysed and corrected, instead of at a customer site. It is a deliberate application of stress, and like any stress it must be justified by the failures it prevents.

What Screening Is Meant to Achieve

Manufacturing variation leaves some units with marginal joints, weak interfaces, or contaminated surfaces. These latent defects survive electrical test because they are not yet failures; they need a stimulus to become detectable. Screening supplies that stimulus in a controlled way, and the resulting failures are treated as process information rather than as bad luck.

The purpose is not to prove reliability. Screening removes early failures from a population, which improves the delivered failure rate, but it says nothing about long-term wear-out. Those two aims require different tests, and confusing them leads to profiles that are either too weak to precipitate defects or so severe that they consume useful life. A screen that removes a hundred hours of life to catch one defect is a poor trade, and that judgement belongs in the reliability plan.

Burn-In Versus Thermal Cycling

Burn-in holds the assembly at an elevated temperature, usually with power applied, for a defined period. It precipitates defects driven by temperature and by electrical stress, such as marginal semiconductors and unstable connections. Temperature cycling, by contrast, moves the assembly between temperature extremes, which stresses mechanical interfaces through differential expansion.

Many programmes use both, in sequence or in combination. The choice depends on the failure mechanisms that matter for the product: a unit with large ceramic parts and heavy connectors responds to cycling, while one with marginal semiconductor processes responds to burn-in at temperature. Choosing a test without a failure mechanism in mind wastes time and money.

Designing a Screening Profile

A profile is defined by the temperature extremes, the ramp rate, the dwell at each extreme, the number of cycles, and whether power is applied. The extremes must remain inside the limits of the components and the laminate, and the ramp rate must be slow enough that the assembly reaches a uniform temperature before the dwell begins.

The profile should be developed from the physics rather than copied. A ramp that is too fast stresses the board unevenly and can damage parts that would never fail in service, while a ramp that is too slow adds hours without adding detection. Thermocouples on representative locations confirm what the assembly actually experiences inside the chamber.

Chamber Loading and Airflow

A chamber is only uniform within a defined working volume, and loading affects that uniformity. Dense loading restricts airflow and creates temperature gradients, so units near the front may see a different profile from those at the back. The chamber’s own qualification data should guide how much can be loaded and where.

Fixturing matters as well. Racks that block airflow, cables that carry heat away, and power supplies inside the chamber all change the conditions. Where powered burn-in is used, the heat generated by the boards themselves must be accounted for, since a fully loaded rack at power can be substantially hotter than the chamber setpoint.

Monitoring During the Test

Continuous monitoring is what turns a soak into a test. Functional monitoring during the cycle detects intermittent failures that would disappear once the unit returns to room temperature, and it records the temperature at which a failure occurred. Without monitoring, the test produces only pass or fail after the fact.

Monitoring also protects the chamber. A unit that fails short and draws excessive current can damage the rack or the power supply, so current limits and per-unit fusing are normal. The data collected during screening is valuable beyond the test itself, because failure temperature and time are clues to the mechanism. Retaining a failed unit in its failed condition, rather than repairing it at once, preserves the evidence that analysis needs.

Failure Analysis After Screening

Every failure precipitated by screening should be analysed, at least to the level of identifying the mechanism. Failures that repeat at the same location or in the same component family indicate a process problem, which screening has successfully exposed. This is the return on the investment: the test has paid for itself by finding a defect that would have shipped.

Units that pass must also be considered. A high failure rate during screening is a signal that the process needs attention, while a rate of zero may mean the test is too weak. Tracking the rate over time, and comparing it with field returns, shows whether the screen is doing anything useful.

Cost, Duration and Batch Size

Screening consumes chamber time, labour, and in some cases a portion of the product’s life. The cost per unit falls as batch size rises, but loading density is limited by uniformity. For low-volume, high-value products, screening is often easily justified; for high-volume consumer goods, the cost per unit usually outweighs the benefit.

The duration should be justified by data. A programme that runs because it has always run is not a quality measure, it is a habit. Reviewing the failure rate, the mechanisms found, and the field return data at intervals can show whether the duration can be shortened or the screen removed altogether.

Standards and Customer Requirements

Customers in aerospace, defence, medical, and automotive markets often specify screening in contractual documents, sometimes referencing established standards for temperature range, duration, and monitoring. Where a specification exists, it takes precedence over internal preference, and deviations require documented approval. The record should capture the profile, the equipment, the loading arrangement, and the monitoring results for each lot.

Standards rarely prescribe a single profile, because the right profile depends on the failure mechanisms and the product. They establish the framework and the documentation requirements. The supplier is expected to justify the chosen conditions and to keep records that demonstrate each unit was screened as specified.

Common Mistakes and Limitations

The most common mistake is applying a profile that is severe enough to damage good units, which converts a screening programme into a yield problem. The second is loading the chamber beyond its uniform volume, so that some units receive a milder stress than intended and pass without being tested. Chamber qualification with a data logger at several positions is the only reliable way to know what the working volume really delivers.

Screening also has inherent limits. It detects defects that respond to the applied stress, and it cannot detect a design weakness that only appears under field conditions, nor a defect that requires humidity, vibration, or voltage stress. Where those mechanisms matter, additional screens are needed, each justified by its own failure mode.

FAQ

How long should burn-in last? The duration should follow from the failure mechanisms and the observed failure distribution, not from a familiar number. Many programmes use a period justified by supplier data and by the rate at which failures stop appearing, and shorten it when the data supports that change.

Does screening improve long-term reliability? It removes early failures from the delivered population, which lowers the failure rate seen by the customer. It does not slow wear-out mechanisms or improve the design, so a product with an inherent weakness will still fail eventually, only later than it otherwise would.

Should every product be screened? Rarely. Screening is justified where the cost of a field failure is high, the population contains latent defects that cannot be detected otherwise, and the product value supports the added cost. For high-volume, low-cost assemblies, the economics usually favour process control instead.