Good Die in a Bad Neighborhood
Good die in a bad neighborhood (GDBN) is a wafer-sort screening rule that bins out a die that electrically passed its own test if too many of the dies immediately around it failed. A single passing die surrounded by failures is treated as a latent reliability risk rather than a confirmed good part, because the same local wafer event that killed its neighbors — a particle cluster, a process excursion, an edge-of-field lithography error — may have partially damaged it too, just not enough to show up on today's test:
The rule is a radius and a count, not a guess. A fab defines GDBN as a ring of adjacent dies around the candidate — commonly the 8 immediate neighbors, sometimes extended to a 2-die radius for the strictest grades — and sets a fail-count threshold inside that ring, for example rejecting any passing die with 3 or more failed neighbors; tighten the ring or lower the threshold and more marginal die get screened out at the cost of shipping fewer good parts, so the threshold itself is a tuned yield-versus-reliability trade, not a fixed law of physics.
The rule exists because electrical test cannot see everything, especially at the margins reliability grading cares about. A die can pass every functional and parametric limit today and still carry a sub-threshold defect — a thinned metal line, a partially open via, a weakened dielectric spot — that only shows up as a field failure after months of thermal cycling; GDBN trades a small amount of shippable yield for a large reduction in that kind of latent escape, which is exactly the trade an automotive or aerospace customer is paying for.
Automotive-grade parts are where this rule is strictest and most explicit. Suppliers building to AEC-Q100 reliability grades — NXP, Infineon, Texas Instruments, STMicroelectronics among them — apply GDBN alongside part average testing as a standard spatial outlier screen on every wafer, and testers from Advantest and Teradyne carry built-in wafer-map rule engines so the neighbor-count check runs inline during wafer sort instead of as a separate offline pass; a part destined for a consumer product with a looser reliability target may skip GDBN entirely and ship the same die that an automotive line would have rejected.
Read a GDBN reject through a *neighborhood* lens rather than a *this-die* lens — the part failed not because of what the tester measured on it, but because of what the tester measured all around it, which is the whole point: the defect you can see on a wafer map is often a better predictor of hidden failure than the test result on the one die in question.