Infant Mortality

Infant mortality is the early-life failure pattern where devices carrying latent manufacturing defects fail disproportionately soon after deployment — the first, decreasing-failure-rate region of the classic reliability bathtub curve — and it's almost entirely preventable through screening and burn-in, because the defects causing it already exist at the moment the device ships, just waiting for enough operating stress to actually manifest as a failure.

Latent defects are the entire story behind infant mortality — a device carrying one doesn't necessarily fail its own manufacturing test, but the defect sits there, weak and waiting, until enough real operating stress eventually pushes it into outright failure. A particle-induced short that's marginal rather than fully conductive, a solder joint with slightly insufficient volume, a plating void too small to cause an immediate open — all of these can pass standard electrical test cleanly while harboring a genuine, pre-existing weakness that ordinary operation will eventually expose. The defect isn't created by use; it's revealed by use, which is exactly why infant mortality concentrates so heavily in the earliest hours and weeks of actual operation rather than appearing randomly throughout a product's life.

The specific manufacturing root causes behind infant mortality — contamination, plating defects, incomplete solder joints, residual stress, CMP or lithography errors, assembly issues — span essentially every major process step a device passes through, which is exactly why no single fix eliminates it. Each of these defect categories traces back to a different, specific point in fabrication, assembly, or packaging, meaning reducing infant mortality systematically requires attention across the entire manufacturing flow rather than hardening any single process step alone. This breadth is also why infant mortality rate serves as such a useful composite signal of overall manufacturing process health — a rising infant mortality rate points toward a problem somewhere in the broader process, even before the specific root cause has been pinpointed.

Burn-in deliberately operates devices at elevated stress — higher voltage, higher temperature, or both — specifically to accelerate exactly the failure mechanisms that would otherwise show up slowly during normal customer use, compressing weeks or months of real-world stress exposure into hours of factory testing. A device carrying a latent defect is far more likely to actually fail during this accelerated stress than it would be during the gentler conditions of normal operation, letting the weak device get caught and discarded at the factory rather than shipping to a customer who then experiences the failure firsthand. HTOL (high-temperature operating life) testing is the formal, standardized version of this same accelerated-stress philosophy, used specifically to qualify and verify that a given process's infant mortality rate meets target specifications before volume production begins.

Screening through electrical testing complements burn-in by catching devices whose weakness shows up as a measurable electrical parameter deviation, even without needing the device to actually fail outright under stress. Some latent defects produce subtle shifts in measurable parameters — slightly elevated leakage current, marginal timing margins — that electrical screening can flag directly, without requiring the device to be pushed all the way to catastrophic failure first. Combining burn-in (catching outright failures under accelerated stress) with screening (catching marginal parametric deviations) covers a broader range of latent defect severity than either technique alone would catch.

The cost asymmetry behind infant mortality control is stark and well-understood throughout the industry: a device failing during factory burn-in costs the manufacturer a scrapped part, while the same device failing in the field after shipping costs vastly more — warranty claims, customer dissatisfaction, brand damage, and potentially expensive product recalls. This asymmetry is exactly why manufacturers are willing to accept real burn-in and screening costs — extra test time, extra equipment, extra factory floor space — specifically to shift failures from the expensive field-failure category into the much cheaper factory-scrap category. Getting the burn-in stress level and duration right is itself a genuine engineering tradeoff: too little stress leaves weak devices undetected and shipping to customers anyway; too much stress risks damaging genuinely healthy devices that would otherwise have shipped and performed reliably for their full intended lifetime.