Every lesson so far has assumed a pattern can be measured well enough to know whether it printed correctly. This lesson is about how. Two quantities matter most in lithography: overlay, how well one layer's pattern lines up with the layer below it, and CD, how close a printed feature's width came to its target. Both are measured on every lot, and both get a dedicated metrology toolset and a dedicated statistical method for mapping out, in advance, how much dose and focus error a layer can tolerate before it fails.
Reading time: about 8 minutes.
6.1 Overlay
Overlay is the registration error between the current layer's pattern and a previous layer already on the wafer, measured in nanometers, typically in x and y separately. Lesson 5's LELE scheme showed why overlay matters directly: two masks for the same layer must register to each other; every layer also has to register to the layers below and above it, for example a contact must land on the gate or diffusion it is meant to connect to, not drift off the edge of it.
Overlay is measured with dedicated targets, box-in-box or advanced imaging targets printed in the scribe lines between dies, that let a metrology tool directly measure the offset between the current layer's printed target and the previous layer's target at the same location. Overlay error has both a systematic component (consistent across the wafer, for example a uniform magnification mismatch between two exposure passes) and a random component (field to field or die to die), and the overlay budget for a layer is allocated between the two, because they are corrected differently: a systematic error can often be compensated by adjusting the scanner's exposure model for the next lot, while random error is reduced only by improving process and tool stability itself.
6.2 CD-SEM and scatterometry
CD is measured two main ways:
- CD-SEM (critical-dimension scanning electron microscope): images the actual printed feature at high magnification and measures its width directly from the image. CD-SEM is a direct, intuitive measurement, but it is comparatively slow, since each site is imaged individually, and the electron beam itself can slightly damage sensitive resist.
- Scatterometry (optical critical dimension, OCD): shines light onto a periodic test structure and measures the diffracted light's intensity and polarization pattern, then fits that measured signal against a library of simulated signals for many candidate structure shapes to infer the actual printed dimensions. Scatterometry is much faster than CD-SEM and non-destructive, but it measures an averaged structure over a small periodic array, not a single isolated feature, and it depends on the accuracy of the simulated model library it is fit against.
Most fabs use both: scatterometry for fast, high-volume monitoring across many sites per wafer, and CD-SEM for detailed characterization, process development, and investigating anything scatterometry flags as out of spec.
6.3 The focus-exposure matrix and Bossung curves
Before a layer's recipe is released to production, lithography engineers characterize how CD responds to dose and focus error together, using a focus-exposure matrix (FEM): a special wafer exposed with every field intentionally set to a different combination of dose and focus offset, stepping across a grid of both. Measuring CD at every field on this wafer and plotting CD against focus, as a separate curve for each dose, produces a family of curves called Bossung curves.
The process window is the region of dose and focus (traced as the overlapping area of the Bossung family that stays within CD spec, usually drawn as an ellipse or rectangle) inside which the layer is guaranteed to print correctly. A wider process window is more robust to normal process drift, scanner dose repeatability, wafer non-flatness and the depth-of-focus limits lesson 3 covered; a layer with a razor-thin process window is the layer most likely to need the resolution-enhancement techniques of lesson 4 and the multiple-patterning schemes of lesson 5, specifically to widen it back out.
6.4 Worked example: reading a process-window report
Assumptions: two illustrative layers from the same FEM characterization run, both targeting the same nominal CD, with Bossung data already fit.
- Layer A's process window spans a wide focus range at every qualified dose, with all Bossung curves staying comfortably between spec limits across that range.
- Layer B's process window is narrow: only a small combined range of dose and focus keeps every curve within spec, and a small scanner dose drift that Layer A would absorb easily pushes Layer B's CD out of spec.
- A lithography engineer seeing this comparison would flag Layer B as the layer needing attention first, whether that means adding SRAFs, tightening dose control specifically on that layer, or revisiting whether it needs a resolution-enhancement technique Layer A does not.
This is the practical reason FEM characterization happens before a recipe goes to production, not after: it finds the layer's actual tolerance for process variation while there is still time to widen it, rather than discovering the tolerance was too tight once yield data starts coming back low (lesson 7 of the Fab Equipment & Yield path covers yield loss from exactly this kind of process-window excursion).
6.5 Interview questions
What is overlay, and why does it have both a systematic and a random component? Overlay is the registration error between the current layer's pattern and a previous layer's. The systematic component (for example a uniform magnification mismatch) is consistent across the wafer and can often be corrected by adjusting the scanner's exposure model; the random component varies field to field or die to die and is reduced only by improving process and tool stability.
Compare CD-SEM and scatterometry. CD-SEM images and directly measures an actual printed feature, is slow and can slightly damage resist; scatterometry fits a diffracted optical signal from a periodic structure against a simulated model library, is fast and non-destructive, but measures an averaged periodic structure and depends on model accuracy.
What is a focus-exposure matrix, and what does it produce? A test wafer exposed with every field set to a different dose/focus combination, used to measure CD across that grid and produce Bossung curves, CD plotted against focus for each dose.
How is a process window defined from Bossung curves? As the region of dose and focus where every dose's CD-vs-focus curve stays within the CD spec limits; outside that region, at least one qualified dose prints out of spec somewhere in the focus range.
Why would a layer with a narrow process window be a priority for resolution enhancement? A narrow window means ordinary process drift, scanner dose repeatability, or wafer non-flatness can push CD out of spec; RET techniques like SRAFs or phase-shift masks, and multiple patterning, are specifically used to widen the process window back out for such layers.
Key takeaways
- Overlay measures registration between layers; CD measures printed feature width; both are tracked on every lot.
- CD-SEM directly images features but is slow; scatterometry fits a fast optical signal to a simulated model library but measures averaged periodic structures.
- A focus-exposure matrix characterizes CD across a dose/focus grid, producing Bossung curves.
- The process window is the dose/focus region where every curve stays within CD spec; a narrow window is fragile to ordinary process variation.
- RET and multiple patterning are, among other things, tools for widening a layer's process window, not just for resolving a finer pitch.