Wafer Bow Stress Measurement
# Wafer Bow Stress Measurement
Every stress number used to predict via stress migration, chip-package interaction, or low-k TDDB has to come from somewhere, and the most common way fabs actually measure thin-film stress doesn't touch the film directly at all. It measures how much the whole wafer curves. Deposit a stressed film on one side of an otherwise flat wafer, and the wafer bends — tensile films pull the wafer into a slight bowl, compressive films push it into a slight dome — and the amount of that curvature, measured optically across the whole wafer, converts directly into the film's own stress through a single, well-established equation.
Stoney's equation is the bridge between a measurable bend and an otherwise invisible internal stress. A thin, stressed film deposited on a much thicker substrate acts like a bimetallic strip: the film wants to change length, the substrate resists, and the compromise is a slight curvature of the whole assembly rather than either layer doing what it would on its own. Stoney's equation relates that curvature directly to film stress through the substrate's known thickness and elastic properties and the film's thickness — measure the wafer's radius of curvature before and after the film deposition, plug in the known substrate and film thicknesses, and the equation returns the film's average stress in units of pressure, without ever needing to physically probe the film or destroy the wafer to get there.
Optical scanning measures the curvature without ever touching the wafer, which is exactly why this method became the production-line standard. A laser or multi-beam optical scanner sweeps across the wafer surface, measuring the reflected beam's angle at many points to reconstruct the wafer's surface curvature map directly, before and after each deposition step. Because the technique is fully non-contact and takes seconds per wafer, fabs run it routinely as an in-line process-control check after every stress-sensitive deposition — nitride, oxide, or metal films that are known stress sources — rather than reserving it for occasional destructive failure analysis. Equipment from suppliers like KLA and Frontier Semiconductor built entire tool categories around exactly this measurement, since it's one of the few stress-characterization methods fast and non-destructive enough to run on every wafer in a lot rather than a destructive sample.
The number this method returns is an average over the whole film, which is both its main strength and its main limitation. Stoney's equation assumes a uniform film and a simple curvature response, and it returns one stress value representing the film averaged across the entire wafer — it can't localize stress to one particular structure, line, or via the way a more targeted technique like Raman spectroscopy or X-ray diffraction mapping can, both of which measure stress at specific, small spots rather than a whole-wafer average. That tradeoff is exactly why wafer bow measurement and point-specific techniques complement each other in a reliability engineer's toolkit rather than replacing one another: bow measurement screens every wafer, cheaply and fast, for whether a given deposition process is drifting out of its expected stress window at all, while targeted spot techniques get reserved for root-causing a specific structure once bow measurement or a device failure has already flagged that something is wrong.
Via stress migration and chip-package interaction both describe what happens once a stress gradient exists inside a finished interconnect stack. Wafer bow measurement is the upstream step that actually quantifies the stress in the first place, during process development and ongoing production monitoring — the number a process engineer checks before a new film recipe ever reaches a wafer carrying real devices, let alone a packaged part undergoing CPI qualification.