In short: Burn-in is a reliability screen that operates chips at elevated voltage and temperature for a fixed period so that weak, defect-bearing parts fail on the test floor instead of in the customer's hands. It attacks the "infant-mortality" region of the bathtub curve, trading a few hours of stress for far lower early-life field returns.
Burn-in is a post-fabrication reliability test in which packaged (or sometimes wafer-level) devices are powered up and exercised under deliberately harsh conditions — typically a raised supply voltage and an oven temperature of 85 °C to 150 °C — for a controlled dwell time ranging from a few hours to over a day. The goal is not to confirm a chip meets its datasheet; that is what functional and parametric ATE testing does. Instead, burn-in exists to accelerate latent defects so that any part carrying a weak oxide, a marginal via, a contaminant particle, or a stressed bond wire breaks down before it ships.
The physical idea is simple. A defect that would take months to kill a device at normal operating conditions can be driven to failure in hours if you push the electric field across gate oxides and the thermally activated diffusion of contaminants. Weak parts die early; healthy parts are essentially unaffected because the applied stress stays within their design margin. After burn-in the lot is re-tested, the failures are culled, and only the survivors move on to final shipment.
Semiconductor reliability over a product's life follows the classic bathtub curve. Failure rate starts high and drops steeply (infant mortality), settles into a long flat valley of low random failures (the useful life), and finally rises again as wear-out mechanisms like electromigration and time-dependent dielectric breakdown accumulate. Burn-in is aimed squarely at the left wall of that bathtub.
Burn-in leans on two well-characterised acceleration levers that let engineers compress a defect's lifetime into a practical test window.
Engineers combine both into an acceleration factor (AF) — the ratio of field lifetime to stress-test lifetime. A high AF means a short oven dwell screens for defects that would otherwise take years to surface. The art lies in choosing conditions aggressive enough to catch the weak units yet gentle enough not to consume meaningful life from good ones.
A typical production burn-in cell loads devices into sockets on a burn-in board (BIB), inserts the boards into a temperature-controlled oven chamber, and drives each part with patterns while monitoring its response.
In static burn-in the device simply sits biased in the oven with fixed voltages applied — cheap, but it only stresses what the DC bias reaches. Dynamic burn-in clocks the part and toggles its inputs so internal nodes switch, exercising far more of the circuit. The most capable variant, monitored (or test-during-burn-in, TDBI), reads each device's outputs live inside the oven, catching failures the instant they occur and sometimes shortening the required dwell.
Burn-in is effective but costly: it consumes oven time, energy, expensive burn-in boards and sockets, and adds cycle time to the supply chain. For high-reliability markets — automotive, aerospace, medical, and data-center silicon — that cost is justified because a single field failure can be catastrophic or very expensive to recall.
For high-volume consumer parts, the industry increasingly supplements or replaces blanket burn-in with smarter, cheaper screens:
The engineering decision is always an economic one: the cost of burn-in versus the cost, and risk, of letting an infant-mortality part escape to the customer.
Real projects and industry tool flows, with certification.
Explore CourseTron Courses →Not quite. Consumer "stress tests" check whether a working system holds up under load. Semiconductor burn-in is a reliability screen at the chip level whose purpose is to make defective units fail so they can be removed, not to benchmark performance.
It varies with the product, the required reliability target, and the acceleration factor achievable at the chosen voltage and temperature. Dwell times commonly run from a few hours up to 24–48 hours; high-reliability or military parts may specify longer.
Minimally, by design. Conditions are chosen so the stress stays inside a healthy part's margin, consuming only a negligible fraction of its useful life while rapidly killing defective units. Over-aggressive burn-in that eats real lifetime is considered a process error.
For many consumer parts, yes — manufacturers substitute IDDQ, statistical outlier detection, and wafer-level screens. For automotive, medical, aerospace, and critical infrastructure silicon, burn-in (or an equivalent qualified screen) is usually mandatory.
Related: related topic · Glossary · all courses.