Process Window
A process window is the region of process-parameter space — dose and focus, in lithography's most common case — where a manufacturing step still produces results within specification. Everywhere inside that region is a working chip; everywhere outside it is scrap.
In lithography, the exposure-defocus (E-D) window is the most-used version of this idea. Plot every combination of exposure dose and focus offset that still prints a feature within specification, and the result is a roughly elliptical region on a dose-vs-defocus chart — exactly what the diagram above shows. The window's horizontal extent at the nominal dose is the depth of focus (see our Depth of Focus article for why DOF shrinks quadratically with NA); its vertical extent at best focus is the exposure latitude, the percentage dose variation a process can absorb and still print correctly. A bigger window means a more forgiving, more manufacturable process; a vanishingly small one — as EUV at advanced nodes increasingly is — means there's almost no margin for the dose and focus errors that real scanners, real wafers, and real environments inevitably introduce.
Mask error enhancement factor (MEEF) determines how much of that window gets eaten by the mask itself. At MEEF = 1, a 1nm error on the mask shows up as exactly 1nm of error on the wafer — ideal. Near the resolution limit, MEEF routinely runs 3–4x or higher, meaning the mask's own manufacturing tolerance has to be tightened proportionally just to keep the same wafer-level error budget. A process window computed purely from optics and resist, without accounting for MEEF, systematically overstates how robust the real process actually is.
Process capability (Cp and Cpk) is how a process window's robustness gets reported as a single number — see our dedicated Cpk article for the full breakdown of centered-vs-off-center capability, the standard thresholds (Cpk ≥ 1.33 for production, ≥ 1.67 for critical parameters), and why Cpk can never exceed Cp. The short version as it applies here: a wide process window with a process mean sitting right at its center is a high-Cpk, low-risk process; the same window with the mean drifted toward one edge is a lower-Cpk process running the same nominal recipe with far less real margin for error.
The window shrinks from multiple directions simultaneously as nodes advance, which is why margins get tight faster than any single factor would suggest. Smaller features need tighter depth of focus. Higher NA, while improving resolution, degrades DOF quadratically (the trade-off covered in detail in our Depth of Focus article). MEEF climbs as features approach the resolution limit, amplifying whatever mask error budget remains. And the scanner's own focus budget — wafer flatness, topography, lens aberrations — doesn't shrink just because the window did. Every one of these pressures compounds in the same direction, which is exactly why process-window engineering, not any single optical specification, is what ultimately decides whether a given resolution target is actually manufacturable in volume.
In practice, a real process has to satisfy several overlapping windows at once, not just the lithography window alone. The pattern has to print correctly (litho window), survive the etch step that follows (etch selectivity and profile-angle windows), and meet overlay alignment to the layers below and above it — and the usable process is the intersection of all of them, not any one window considered in isolation. A layer with a generous lithography window but an incompatible etch process window is just as unmanufacturable as one with a tight lithography window alone.