Silicon Almanac.All explainers
Drag the scene to turn it · scroll to continue

Process control inspection

Finding the defect.

A chip can die from one speck a few nanometres wide. Somewhere on a 300 mm wafer, a machine has to find it, and fast enough to stop the next lot from getting it too.

NIST put the scale this way: a 10 nm defect on a 300 mm wafer is like one foot in 3,000 miles. This page is about the machines that look, and the choices about where to look. Scroll to go on.

~3 nmkiller particles, by the IRDS roadmap
~1,000×e-beam is slower than optical
$18 bn+spent on inspection and metrology, 2025

NIST (2013); IRDS 2024 Yield Enhancement, on particles in ultrapure water; Semiconductor Engineering (2016); Yole Group (2026).

01 Zoom

One foot in 3,000 miles.

Zoom in from the wafer to a single defect, five steps down. By the last one, a speck of material bridges two lines that were meant to stay apart.

300 mm

The images are schematic. The comparison is Rick Silver's, of NIST: "One foot in 3,000 miles is identical a 10 nm defect over a 300 mm wafer." The IRDS roadmap puts the critical particle size below 5 nm, and as small as 3 nm for the most critical particles.

02 Two ways to look

Fast, or sharp.

Inspection finds where defects are; metrology measures whether the pattern came out the right size. Both face the same choice. Light is fast and sweeps a whole wafer, but blurs below a few tens of nanometres. An electron beam sees the smallest defects, but crawls.

Finds defects down to-
Speed-

Figures as reported by Semiconductor Engineering (2015, 2016), quoting industry experts: e-beam inspection "1,000 times (or more) slower than optical". So fabs use light to find candidates, then electrons to look closely at the few that matter. As a KLA column put it: "You can't fix what you can't find."

03 Closing the gap

More beams at once.

If one electron beam is too slow, use many. Multibeam inspection tools split the work across a grid of beams that scan in parallel. Others make a single beam brighter and faster.

Throughput, as claimed-
Compared witha single-beam tool

ASML's figures for its HMI eScan 1000 (nine beams, 2020) and eScan 1100 (25 beams). Applied Materials took the other route: a cold field emission source it says images up to 10× faster. Vendor claims, labelled as such.

04 You cannot look everywhere

Measure a few, trust the rest.

Fabs do not measure every point on every wafer. They sample. A typical metrology plan, as Hitachi High-Tech describes it: one or two wafers from each lot of 25, five to twenty dies on each, ten to a hundred points on each die.

1
10
30
Points measured per lot0
Wafers touched0

Ranges from Hitachi High-Tech's metrology primer; the totals are arithmetic. KLA engineers have written that the best way to balance cost is usually to adjust the lot sampling rate.

05 Why it pays

Every lot until someone looks.

When a tool drifts, every lot through it is at risk until an inspection catches it. Here a tool goes wrong at the red lot; blue gates are inspections. Space them out and more lots are made wrong before anyone knows.

every 8 lots
Lots at risk0
Wafers at risk0

Arithmetic for this example, 25 wafers to a lot. KLA's columns make the point in words: "Finding the excursion sooner limits the lots at risk." Yole puts inspection and metrology at over $18 billion in 2025, about 14% of fab equipment spending. KLA's own slides, as Investing.com reported them, claim 58% of the process control market.

06 Where to go next

Look first, then fix.

Every good die on a wafer is one where nothing went wrong, often because a drift was caught before it got there.