There is one lever every process engineer reaches for first: dose. More light means more photons per feature, which means smaller relative shot noise and a thinner defect tail. It works — but it is never free, and the shape of the trade-off decides how much room you actually have to work in. That room is the process window.
More light, fewer defects
Raising the dose raises the mean photon count μ in every feature. Since a feature fails when its count falls below a threshold, a higher mean pushes the threshold further out into the safe part of the distribution, and the failure tail collapses. Sweep the dose across a realistic range and you trace the classic dose-response curve: defect probability falling, often across many orders of magnitude, as dose climbs.
Notice how steep the curve is. Over a span of a few millijoules the probability can fall by several decades. That steepness cuts both ways: it means a modest dose increase buys a large defect reduction, but it also means a modest dose error — drift, non-uniformity across the wafer — can undo that gain just as fast. The width of the dose range over which the pattern stays within spec is the dose latitude, and it is one of the two axes (with focus) that define the process window.
Inverting the question
Engineers rarely ask "what is my defect rate at this dose." They ask the inverse: "what dose do I need to hit my defect target?" Because the dose-response curve is monotonic, that question has a clean answer — find the dose where the curve crosses the target line. It turns a physics model into an actionable setpoint.
The inversion also exposes when dose is the wrong lever. If the curve never reaches your target even at the top of a realistic dose range, no amount of extra light will save the feature — it is too small, absorbs too little, or the threshold sits too high. At that point the answer is a design or materials change, not a hotter exposure.
Every extra millijoule costs throughput — wafers per hour — and adds thermal and resist-damage load. The defect curve tells you what dose buys; the economics of the fab set the ceiling on what dose you can actually spend. The real process window lives where those two meet.
From a curve to a real process
This single-feature dose-response is exactly the dimension explored by the Stage 1 Monte-Carlo engine in the SECD-CRF repository, whose dose-sensitivity study sweeps dose and confirms the defect rate falls monotonically — a check against well-established EUV physics. The closed-form calculator here is the fast cousin of that simulation for a single feature. The full project extends it to whole layouts, where different geometries have different dose-response curves and the process window has to hold for all of them at once.
EUV Dose–Defect Calculator
Set a feature and a target defect rate and read off the dose you need — and see when the target is simply out of reach by dose alone.
Dose is where the physics of shot noise and the defect tail meet the economics of the fab. Understanding the curve is what lets you spend photons where they actually buy yield.