Semiconductor Metrology

Semiconductor metrology is the measurement discipline that gives a fab direct, immediate evidence that its lithography, etch, and deposition steps actually hit the geometry and layer registration a chip design requires — rather than discovering a process problem only when a finished die fails electrical test, hundreds of process steps and roughly three to four months after the wafer entered the fab. That timing gap is the entire reason metrology exists as a dedicated discipline rather than an afterthought: electrical test is the only measurement that directly confirms a chip works, but by the time a wafer reaches it, every lot behind it in the fab has likely inherited the same undetected drift, so metrology's job is to catch a process excursion inline, step by step, while there is still a chance to stop it before it compounds across hundreds of wafers.

Without inline metrology
process drift is invisible until electrical test, 3–4 months and hundreds of steps later
Inline metrology catches drift at the step where it happenedthree disciplines, each checking a different failure mode, run throughout the flow
CD-SEM + OCD scatterometry
is the feature the right size and shape?
Overlay metrology
is this layer registered to the one below it?
Film/composition metrology
is the film the right thickness and material?
Drift caught and corrected within the same lot
SPC alarms an engineer, or APC adjusts the next run automatically

The economics behind this timing gap are stark enough to justify metrology's entire cost. A single wafer can carry hundreds of individual dies, each one representing accumulated material, tool time, and process steps across a journey of 60–100+ patterned layers; a defect that survives undetected to electrical test doesn't just kill the one lot it was found in, it typically means every lot processed on the same tool during the same drift window carries a correlated risk of the identical defect. KLA, the dominant process-control vendor, has put the economics plainly in its own investor materials for years: catching a yield-limiting excursion at the metrology step where it occurred costs a tool recipe adjustment, while catching the same excursion at final test costs the fully-loaded value of every wafer that passed through the fab during the undetected drift window — which is why metrology and inspection tools, despite producing no product themselves, typically consume 10–12% of a leading-edge fab's total equipment capital budget.

Metrology's Place in the Fab Flow inline measurement at every step vs. one verdict at the very end Deposition step N Film metrology tap ellipsometry / XRF checks before next step Lithography step N+1 Overlay metrology tap diffraction-based AIM targets, sub-nm budget Etch step N+2 CD-SEM / OCD tap critical-dimension check before the lot moves on ... ~60-100+ more layers 3-4 months Electrical test the only direct pass/fail — far too late to prevent loss Inline taps catch drift at the step it happened a correction applied here stops before the next hundred lots inherit it electrical test alone would only reveal the damage after it already compounded Metrology + inspection ≈ 10–12% of leading-edge fab capex equipment that manufactures nothing, yet decides which lots are worth continuing to process Three disciplines, explored in the rows that follow this one CD-SEM vs. OCD mechanism trade-off · coordinated SPC across disciplines closed-loop APC feeding corrections back into the process automatically

Metrology's position in the flow is also what distinguishes it from inspection, a related but genuinely different discipline it is frequently confused with. Inspection tools, also led by KLA alongside Applied Materials' defect-review line, scan a wafer surface broadly looking for the presence of a discrete defect — a particle, a void, a pattern-collapse event — without necessarily quantifying it against a specification; metrology, by contrast, measures a continuous physical quantity (a critical dimension, an overlay offset, a film thickness) against a numerical target and control limit. A fab needs both: inspection finds the needle in the haystack that metrology's targeted, high-precision measurement would never randomly land on, while metrology provides the continuous, quantitative signal that statistical process control and closed-loop automation depend on to function at all.

Question metrology answersTechniqueWhere it sits in this topic
Is the feature the right size and shape?CD-SEM (direct imaging) + OCD scatterometry (model-fit)Covered in the next two rows
Is this layer registered to the one beneath it?Diffraction-based overlay (DBO) on AIM targetsCovered in the taxonomy row
Is the film the right thickness and composition?Spectroscopic ellipsometry, XRF, XRRCovered in the taxonomy row
Did a drift in one parameter also move another?Coordinated SPC across disciplinesCovered in the SPC row
Can the process correct itself automatically?Closed-loop APC, run-to-runCovered in the final row

None of the individual measurement techniques explored in the rest of this topic matters without this framing: metrology exists to compress the gap between a process step going wrong and someone finding out, from months down to minutes, because every day that gap stays open is another day of wafers inheriting a defect nobody has caught yet. The specific physics — electron-beam imaging, diffraction, ellipsometry — is the engineering detail; the economic logic that justifies spending 10–12% of a fab's capital budget on equipment that makes nothing is the reason this discipline exists at all, and it is the thread that ties together every row that follows.

Take semiconductor metrology further

Ask the copilot about this term, or have our engineers assess it against your process.