Mask Inspection
Mask inspection scans a finished photomask's entire pattern at high resolution and compares it against either the original design database or an identical neighboring die, flagging any deviation that could print as a defect on every single wafer the mask subsequently exposes. Because one undetected defect doesn't stay contained to one die the way a wafer-level defect would, mask inspection has to operate at a sensitivity level most other inspection processes in the fab never need to reach — near-perfect detection, not merely good detection.
Die-to-database inspection compares the mask's actual imaged pattern directly against the original design intent, making it the only inspection mode that can catch an error on a mask with no repeated pattern to compare against. A reticle containing a single, unique die layout — which is the normal case for most advanced logic masks — has no identical neighboring pattern on the same mask to compare against, so the only way to confirm the mask matches its intended design is comparing the actual imaged pattern directly against the design database itself. This comparison catches any deviation from intent, whether it originated in mask writing, mask etch, or contamination that landed after the fact, regardless of whether that deviation happens to repeat anywhere else on the mask.
Die-to-die inspection exploits repeated patterns on the mask itself, comparing identical dies against each other rather than against an external reference database — a faster, simpler comparison that works specifically because any genuine design feature appears identically in both locations. When a mask contains multiple copies of the same die (common for memory or certain repeated-structure masks), any real difference between two supposedly identical locations has to be a defect, not a design feature, since the design itself specifies identical patterns at both sites. This comparison mode doesn't require loading and aligning against the full design database, which can make it computationally faster for masks where it's actually applicable.
Inspection wavelength directly limits achievable sensitivity, which is exactly why EUV masks need their own dedicated inspection approach rather than simply reusing conventional DUV inspection tools. DUV inspection wavelengths — 193 or 248 nanometers — deliver the finest resolution available for conventional, transmissive DUV masks, consistent with the general principle that shorter inspection wavelengths resolve finer defects. But EUV masks are reflective rather than transmissive, and more importantly, a defect that looks completely invisible or harmless under DUV inspection light can still behave very differently, and potentially print, under actual 13.5-nanometer EUV exposure — which is exactly why EUV masks require actinic inspection, using EUV light itself to confirm how a potential defect will actually behave under the real exposure conditions the mask will experience in production.
Sensitivity requirements for mask inspection tighten with every advancing process node, because the same physical-sized defect that could be safely ignored at a relaxed node becomes fatal at a tighter one. A particle or pattern deviation of a given physical size represents a shrinking fraction of a feature's total dimension as that feature itself shrinks generation over generation — meaning a defect that printed as a harmless, sub-specification deviation at an older node can cross into genuinely fatal territory at a newer, tighter one, even though the defect's own physical size never changed at all. This relentless tightening is exactly why mask inspection tool sensitivity has had to improve continuously, generation after generation, just to keep pace with node scaling rather than falling behind it.
Mask inspection's near-zero tolerance for missed defects, combined with the sheer resolution and throughput demands of scanning an entire mask at the sensitivity modern nodes require, makes inspection a genuinely significant fraction of total mask manufacturing time and cost. Every mask has to be scanned in full, at whatever sensitivity its target process node demands, before it can proceed to qualification and production use — there's no practical way to inspect only part of a mask and extrapolate confidence about the rest, because a single missed defect anywhere on the pattern defeats the entire purpose of inspecting at all. This exhaustive, zero-compromise requirement is exactly why inspection tool capability and throughput remain a continuous area of investment for mask-shop equipment, directly alongside writer and metrology tool development.