Overlay Fingerprint

An overlay fingerprint is the systematic, repeatable pattern an overlay error map forms across a wafer or across exposure fields — decomposed into named components like translation, rotation, magnification, and distortion, with whatever doesn't fit any of those patterns left over as residual, non-correctable error. Separating a fingerprint's systematic components from pure randomness is exactly what makes targeted correction possible at all: you can't correct what you haven't first identified as a repeatable pattern rather than noise.

Interfield components describe how overlay systematically varies from one exposure field to the next across the whole wafer — translation, rotation, and magnification being the three most basic, intuitive terms in that decomposition. Translation is a uniform shift applied equally everywhere on the wafer; rotation is a systematic angular offset that grows with distance from the wafer's center; magnification describes a uniform scaling difference between the two layers, as if one pattern were printed slightly larger or smaller than the other across the whole wafer. Higher-order interfield terms — trapezoidal distortion and beyond — capture more complex wafer-level patterns that these three simplest terms don't fully describe on their own.

Intrafield components capture overlay variation within a single exposure field — distinct from interfield variation, and tied specifically to the scanner's own lens characteristics rather than to anything happening at the wafer level. A scanner's projection lens doesn't image every point within its exposure field with perfectly identical fidelity; small, repeatable lens distortions mean overlay can genuinely vary from one position to another within the same field, following a pattern describable by third-order, fifth-order, and higher polynomial terms. This distinction between interfield (wafer-level) and intrafield (within-field) variation matters because the root causes, and therefore the correction strategies, are completely different for each.

Per-exposure corrections are what a scanner actually applies in real time, field by field, based on the fingerprint model derived from prior measurement — this is the step where identifying a systematic pattern actually translates into reduced overlay on the wafer currently being exposed. Once interfield and intrafield systematic components have been characterized from measurement data, the scanner can apply compensating adjustments to each individual exposure as it happens, correcting for exactly the systematic error the fingerprint model predicts at that specific field location. This is precisely why these components get called "correctables" — they're not just descriptive categories, they're the specific terms a scanner's control system can actually act on.

The non-correctable residual is whatever remains after every systematic term the fingerprint model can describe has already been identified and corrected — the genuine, irreducible overlay floor rather than a modeling shortfall. This residual represents random, unpatterned error that doesn't follow any of the systematic forms (translation, rotation, magnification, lens distortion) the fingerprint decomposition searches for, meaning no amount of additional fingerprint modeling sophistication can reduce it further. Distinguishing genuine non-correctable residual from a systematic pattern the current model simply hasn't captured yet is an ongoing, real analytical challenge — improving fingerprint models over time is partly about finding systematic structure that was previously misclassified as pure residual.

Fingerprints genuinely change lot to lot, with cassette slot position, and with process conditions, which is exactly why real-time correction based on current measurement data is essential rather than a single fixed correction being set once and reused indefinitely. A fingerprint characterized from one lot's measurement data doesn't necessarily describe the next lot's overlay behavior perfectly, because the specific sources of systematic error (thermal history, process-tool variation, wafer handling) shift somewhat from lot to lot and even from wafer to wafer within a cassette. This is also why tool matching between different scanners is its own distinct challenge: two scanners, even nominally identical models, develop their own somewhat different fingerprints over time, and running the same process across multiple scanners without accounting for each tool's own fingerprint risks introducing exactly the overlay variation careful, fingerprint-aware correction is supposed to eliminate.