Silicon burn-in testing is a screen in which chips are powered and operated at high temperature and elevated voltage for hours, so that weak parts fail in the factory instead of in a customer's product. It targets infant mortality, the early failures caused by latent manufacturing defects. Parts that survive burn-in and pass a final electrical test are shipped; parts that fail are scrapped, and their failures are analysed to improve the process.
Key takeaways
- Burn-in screens out early-life (infant mortality) failures; it does not measure or extend the wear-out life of good parts.
- Heat and voltage speed up the chemical and physical processes that turn a latent defect into a hard failure.
- The temperature effect is usually estimated with the Arrhenius model; an illustrative 48 hours at 125 °C is equivalent to several thousand hours at 55 °C.
- Burn-in is costly in time, ovens, sockets and boards, so many consumer products use it only during ramp, while automotive, medical, aerospace and server parts often keep it in production.
- Wafer-level burn-in and known-good-die screening matter more as chiplets and multi-die packages grow, because one bad die can scrap a whole package.
What burn-in testing is
Burn-in test is a type of high-stress testing performed on packaged silicon chips to accelerate the detection of infant mortality failures. Infant mortality failures are those that occur early in a product's life, within the first hours, days or weeks of operation, and are often caused by manufacturing defects such as near-shorts between metal lines, weak gate oxide, voids in vias or improper wire bonding.
During burn-in, packaged chips are subjected to conditions that represent, and exceed, real-world operation. The devices are typically powered on and run at elevated temperatures and voltages for an extended period, often several hours or days. The purpose is to identify devices that are likely to fail in the field due to infant mortality, and to eliminate them before they are assembled into finished products.
Why it works: the bathtub curve
Reliability engineers describe a population of chips with the bathtub curve, a plot of failure rate against time:
| Region | Failure rate | Typical causes | What deals with it |
|---|---|---|---|
| Early life (infant mortality) | High at first, falling quickly | Latent defects from fabrication and assembly | Burn-in and stress screens |
| Useful life | Low and roughly constant | Random events, overstress, rare defects | Good design margins, ESD and latch-up protection |
| Wear-out | Rising | Electromigration, oxide wear-out, hot carrier injection, transistor ageing (BTI) | Design rules and qualification (life tests) |
Burn-in pushes each part through the steep left side of the bathtub before shipment. A defective part fails during the stress; a good part spends a small fraction of its life and leaves the factory already in the flat region.
How a burn-in test is run
- Load: packaged devices are placed in sockets on burn-in boards, often hundreds of devices per board.
- Stress: the boards go into a burn-in oven or chamber. The parts are powered at raised voltage, commonly 10 to 40 per cent above nominal (illustrative range), at a junction or ambient temperature of 125 °C or more.
- Exercise: in static burn-in, inputs are held at fixed levels. In dynamic burn-in, the parts are clocked and toggled with patterns, often from on-chip test logic (scan or built-in self-test), so internal nodes actually switch. Test during burn-in (TDBI) also monitors outputs during the stress.
- Unload and retest: after the set time, parts are removed and given a full electrical test. Parts that now fail, or that drift beyond limits such as leakage current, are rejected.
- Analyse: the failures are sent for failure analysis to find the defect and its process cause.
Heat control is a real engineering problem. A large processor can dissipate tens or hundreds of watts at raised voltage, so modern burn-in systems control temperature per device to avoid thermal runaway on the hottest parts.
Worked example: how much life does burn-in use?
The Arrhenius model estimates how much faster a temperature-driven failure mechanism proceeds at a higher temperature:
AF = exp[ (Ea / k) x (1/T_use - 1/T_stress) ]
Ea activation energy of the failure mechanism (eV)
k Boltzmann constant = 8.617 x 10^-5 eV/K
T_use use temperature in kelvin
T_stress stress temperature in kelvinTake an illustrative activation energy of 0.7 eV, a use temperature of 55 °C (328.15 K) and a stress temperature of 125 °C (398.15 K):
- 1/328.15 - 1/398.15 = 0.000536 per kelvin
- Ea / k = 0.7 / 0.00008617 = 8,123 K
- AF = exp(8,123 x 0.000536) = exp(4.35), about 78
So 48 hours of burn-in at 125 °C is roughly equivalent to 48 x 78, about 3,700 hours (around five months) at 55 °C, for this one mechanism. Raising the stress to 150 °C lifts the factor to about 260. Elevated voltage adds a further acceleration for voltage-driven mechanisms such as oxide breakdown, so the combined factor is larger. All numbers here are illustrative; real activation energies depend on the mechanism and are set by each manufacturer's reliability data.
Why burn-in matters
Burn-in testing is an important step in the manufacturing of silicon chips and integrated circuits, because it improves the reliability of the final product. By identifying and eliminating devices that are likely to fail, burn-in reduces the cost of failures further down the line: a defective chip caught at burn-in costs only the chip, while the same chip found after it has been soldered onto a board, built into a system or shipped to a customer costs far more. It also reduces warranty claims and returns from customers.
For safety-critical and high-reliability markets, early-life failure rate is part of qualification. The automotive standard AEC-Q100 includes an early life failure rate (ELFR) test, and JEDEC publishes methods for temperature, bias and operating life testing (JESD22-A108) and for calculating early life failure rate (JESD74). Military and space parts follow MIL-STD-883, whose burn-in method (Method 1015) is widely cited with 160 hours at 125 °C as a common baseline for one quality class.
Burn-in versus related tests
| Test | Purpose | Applied to |
|---|---|---|
| Burn-in | Screen out weak parts before shipment | Every part (production screen), or samples during ramp |
| HTOL (high temperature operating life) | Qualify the product and process, estimate failure rate over life | Sample lots, typically for 1,000 hours |
| ELFR | Measure the early-life failure rate of a qualified product | Large sample lots |
| IDDQ and voltage stress tests | Catch some of the same defects quickly without ovens | Every part, at wafer sort or final test |
Costs and alternatives
Burn-in is expensive: oven time, burn-in boards and sockets, handling, and a small amount of life consumed in every good part. As processes mature and defect densities fall, manufacturers try to reduce or drop it. Common alternatives and complements include:
- Wafer-level burn-in (WLBI): stressing dies on the wafer before packaging, important for known-good die in chiplets and stacked memory.
- Short high-voltage stress tests at wafer sort that trigger oxide and metal weaknesses in seconds.
- IDDQ testing: measuring quiescent supply current, where an abnormally high value points to a defect.
- Statistical screens, such as part average testing in automotive, which reject parts whose test results are outliers even when they are within specification.
- Sample burn-in: burning in only a sample from each lot and returning to full burn-in if the failure rate rises.
FAQ
Does burn-in damage good chips?
It uses a small part of their life, by design. Conditions are chosen so that good parts lose only a small fraction of their expected lifetime while defective ones fail.
Is burn-in done on every chip?
Not always. It is common for automotive, aerospace, medical, telecom infrastructure and server parts, and during early production of new products. Mature consumer products often rely on other screens.
What is the difference between burn-in and HTOL?
Burn-in is a production screen to remove weak parts. HTOL is a qualification test on samples to show that the design and process meet their reliability target over life.
Your comments will be moderated before it appears here.