Failure Analysis Assistant

Failure analysis on a single failed device is a one-shot game. The sample is precious, the tools — focused ion beams, electron microscopes, emission cameras — are slow and expensive, and most of the techniques are destructive: cut in the wrong place and the evidence is gone. So the defining skill of a good FA engineer isn't operating the microscope; it's forming the right hypothesis from the electrical signature before any physical work begins.

That signature is surprisingly diagnostic. An open that develops over time and accelerates with temperature whispers electromigration. A sudden, localized, catastrophic short shouts ESD or overstress. Leakage that grows with field-stress time points at gate-oxide breakdown. An intermittent failure that comes and goes with temperature implicates the package, not the silicon. Each mechanism leaves a fingerprint in how the part fails and how that failure behaves, and reading it narrows seven candidates to one or two before a sample is mounted.

The first fork is always catastrophic versus wear-out. A sudden, complete failure is localized — you hunt the damage site with emission microscopy and image the burn. A gradual, time-dependent failure is a degradation — you characterize its time and temperature dependence and often need stress data, then cross-section to find the void or the broken oxide. The two demand different mindsets and different tools, and getting the fork right sets the whole investigation.

This assistant encodes those mappings: select what you observe, and it ranks the mechanisms and names the technique to start with — emission microscopy for a hot leakage path, acoustic microscopy for delamination, EDX for a particle, cross-section for a void. It's a starting hypothesis and a plan, not a verdict; the physical analysis confirms it. For the population-level context that often points the way, use the Wafer Map Analyzer and the Root Cause Finder.