Metrology

Metrology is the science of measurement. In semiconductors it means precisely measuring and inspecting wafers during manufacturing to catch defects and keep yields high. With 400-600 process steps per chip, metrology is essential to making the advanced AI processors and memory inside data centers.

What metrology means (plain version)

Metrology is simply **the science of measurement** — defining units, building instruments, and checking that a measurement is accurate. In everyday life it underpins everything from fuel pumps to medical scanners. In the technology world, the word almost always refers to **semiconductor metrology**: the tools and techniques used to measure features on a silicon wafer during chip manufacturing. These features are now just a few nanometers wide (a human hair is roughly 80,000 nm), so you cannot eyeball them — you need specialized machines that measure dimensions, film thickness, and layer alignment, and inspect for tiny defects.

How it works

Making a chip takes **400 to 600 process steps** over one to two months. If a defect creeps in early, every later step is wasted, so manufacturers insert measurement and inspection checkpoints throughout the line. **Metrology** measures *attributes* — critical dimension (CD, how wide a feature is), film thickness, and **overlay** (how precisely one patterned layer lines up with the one beneath it). **Inspection** hunts for *defects* — particles, scratches, or pattern errors. The main techniques are **optical** methods like scatterometry and optical CD (fast, non-destructive, good for repeating structures) and **e-beam** methods like the CD-SEM (slower but higher resolution for the finest features). The data feeds back into the process: run-to-run control and advanced process control (APC) automatically nudge recipes to keep wafers inside tolerance.

Why it matters for AI and data centers

AI accelerators (GPUs) and **high-bandwidth memory (HBM)** are pushing chipmaking to its limits. EUV lithography at leading-edge nodes leaves no margin for error — a tiny overlay or dimension drift can ruin a multi-thousand-dollar wafer. Just as important, AI chips increasingly rely on **advanced packaging** (stacking and bonding many dies together, as in 2.5D/3D and HBM stacks), which creates entirely new things to measure: bump height, warpage, bond alignment. Without metrology to keep yields high, the supply of AI silicon would be slower and far more expensive. This is why measurement and inspection sit at the heart of the AI hardware build-out.

Where it sits in the supply chain

Metrology and inspection are part of the **wafer fab equipment (WFE)** market — the machines fabs buy to make chips — within a category often called **process control**. It complements, rather than competes with, the big deposition, etch, and lithography tool makers. Every fab run by TSMC, Samsung, Intel, SK Hynix, or Micron is studded with these tools at critical steps. A related niche is **test** — for example probe cards that electrically touch each die on a wafer to check it works before packaging. As chips move from pure transistor scaling toward photonics (co-packaged optics) and chiplets, metrology is expanding into these new domains too.

Who the key players are

The market is concentrated. **KLA Corporation (KLAC)** is the dominant force, holding roughly half of the broader process-control market and an estimated 80%+ share in reticle (photomask) inspection — by far the largest player. **Onto Innovation (ONTO)** is a focused specialist whose Dragonfly inspection and metrology platforms have become a chokepoint for **AI packaging and HBM**; it secured a 240M+ USD purchase agreement with an HBM maker and is guiding to 30%+ advanced-packaging growth. **Camtek (CAMT)** leads in inspection and 3D metrology for packaging and bought FormFactor's FRT metrology unit in 2023. **FormFactor (FORM)** specializes in probe cards for wafer test. Other large names include Applied Materials, ASML, Hitachi High-Tech, Lasertec, Nova, and Bruker. *(These are examples for context, not investment advice.)*

What is changing now

Three shifts are reshaping metrology in 2025-2026. First, **advanced packaging and HBM** are the fastest-growing demand driver, pulling spending toward inspection of stacked, bonded, and bumped dies rather than just flat wafers. Second, **AI is being applied inside the tools** — machine learning speeds up defect classification and lets optical tools approximate measurements that used to require slow e-beam imaging. Third, the overall **semiconductor metrology and inspection market** is growing steadily; industry estimates put it in the rough range of 8-10+ billion USD in 2025, expanding mid-single-digit percent annually through the mid-2030s. The common thread is that smaller features and more complex 3D structures mean more — and more difficult — things to measure.

Frequently asked

What is metrology in simple terms?

It is the science of measurement — making sure that when you measure something, the result is accurate and traceable to a standard. In the chip industry it specifically means precisely measuring features on silicon wafers (their size, thickness, and alignment) and inspecting them for defects during manufacturing.

What is the difference between metrology and inspection?

Metrology measures attributes you care about — how wide a feature is (critical dimension), how thick a film is, or how well two layers line up (overlay). Inspection looks for defects you do not want — particles, scratches, or pattern errors. Many tool vendors and fabs do both, and the terms are often used together.

Why is metrology important for AI chips?

Advanced AI processors and high-bandwidth memory are made at the very edge of what is physically possible, with 400-600 steps and nanometer tolerances. Metrology catches drift and defects early so yields stay high; without it, AI silicon would be slower to produce and far more expensive.

What are the main metrology stocks?

The most-tracked names are KLA (KLAC), the dominant process-control leader, Onto Innovation (ONTO), Camtek (CAMT), and FormFactor (FORM). Larger diversified equipment firms like Applied Materials and ASML also participate. This is informational context, not a recommendation.

What is the difference between optical and e-beam metrology?

Optical methods (such as scatterometry and optical CD) use light, are fast and non-destructive, and work well on repeating structures. E-beam methods (like the CD-SEM) use a focused electron beam for much higher resolution on the finest features, but are slower. Fabs use both, choosing the right tool for each step.

Related companies