Measuring What You Can't See
Modern transistors have features measured in single-digit nanometers. The gate length of a 3nm-node transistor is actually around 12-16nm in physical dimension (node names stopped corresponding to actual feature sizes years ago). Measuring structures this small accurately and repeatedly, across billions of features on a 300mm wafer, in the middle of a high-volume production line — that's the job of semiconductor metrology.
It's not glamorous work. But if you can't measure it, you can't control it. And if you can't control it, yield goes to zero.
Critical Dimension Measurement
CD-SEM
Critical Dimension Scanning Electron Microscopy (CD-SEM) has been the workhorse of semiconductor metrology for decades. A focused electron beam scans across features, and the secondary electrons emitted from the sample surface create an image. The intensity profile across a line or space gives you the critical dimension.
Modern CD-SEMs from Hitachi High-Tech and ASML (the former Hermes Microvision business) can measure features below 10nm with sub-nanometer precision. But there's a catch — the electron beam itself can damage the sample. EUV photoresists are particularly sensitive to e-beam damage, which means you can't just blast them with electrons and expect accurate measurements. Low-voltage techniques (below 500V landing energy) help, but measurement precision suffers.
See also: 5G and mmWave Chip Design: RF Front-End, Beamforming, and Po.
Scatterometry (OCD)
Optical Critical Dimension measurement (OCD), also called scatterometry, works by shining polarized light on a periodic grating structure and analyzing the reflected light's polarization change. By comparing the measured optical signature to a library of simulated profiles, you can extract the grating's line width, height, sidewall angle, and other geometric parameters.
The key advantage of OCD: it's fast and non-destructive. A single measurement takes seconds and averages over thousands of features in the measurement spot, giving excellent statistical precision. The downside is that it measures gratings, not actual device structures, so you need well-designed metrology targets that correlate with the real devices. Tools from KLA (SpectraFilm and SpectraShape families) and Onto Innovation dominate this space.
Film Thickness and Composition
Ellipsometry
Spectroscopic ellipsometry measures the thickness and optical properties of thin films by analyzing how polarized light changes as it reflects from the film stack. It's incredibly sensitive — it can detect sub-angstrom thickness variations. Most fabs have dozens of ellipsometry tools monitoring film depositions at virtually every process step.
This connects to the ideas in EUV Lithography Explained: How Extreme Ultraviolet Light Sha.
XRF and XPS
X-ray fluorescence (XRF) identifies elemental composition by exciting atoms with X-rays and measuring the characteristic fluorescence they emit. X-ray photoelectron spectroscopy (XPS) goes further, revealing chemical bonding states in addition to composition. These are critical for monitoring high-k gate dielectric composition, metal gate work function layers, and barrier metals in interconnects.
Defect Inspection
Wafer Inspection
KLA's 29xx-series brightfield inspection tools and their Surfscan systems for unpatterned wafers are the industry standard. Brightfield inspection compares adjacent dies on the wafer — if something appears in one die but not its neighbors, it's flagged as a defect. At advanced nodes, the minimum detectable defect size is around 10-15nm, though sensitivity depends heavily on the background pattern complexity.
Darkfield inspection (KLA's 8 Series and competitors) captures scattered light from defects on patterned wafers. It's faster than brightfield and better for certain defect types, particularly particles on the wafer surface.
See also: Việc Làm Bán Dẫn Việt Nam 2026: Lương & Tuyển Dụng.
E-beam Inspection
E-beam inspection tools image the wafer with electron beams, detecting electrical defects that optical tools miss — like opens and shorts in buried interconnect layers. These tools are slow (hours per wafer for full-wafer inspection) but catch defects that nothing else can see. They're used strategically rather than on every wafer.
Overlay Measurement
Overlay metrology ensures that each lithography layer aligns correctly to the previous one. At advanced nodes, overlay requirements are below 2nm — meaning layer-to-layer alignment must be controlled to a precision smaller than a dozen atoms. ASML's YieldStar and KLA's Archer tools measure overlay using dedicated targets printed alongside the device patterns.
The move to EUV has added new overlay challenges. EUV masks have different heating and distortion characteristics than DUV masks, and the pellicle (protective membrane over the mask) can introduce additional distortion. Getting overlay right at 3nm and below requires feed-forward corrections computed from metrology data and fed into the scanner in real time.
The Data Explosion
A modern fab generates terabytes of metrology data daily. Every wafer gets measured at multiple steps, each generating thousands of data points. Making sense of this flood requires advanced analytics — machine learning models that correlate metrology signatures with final device performance, identifying process excursions before they impact yield. The companies that can extract actionable intelligence from metrology data fastest have a real competitive advantage. It's become as much a data science problem as a measurement physics problem.