Computer Vision Inspection

Wafer inspection at production line speed. False positives below 0.1%.

Lenspathio's computer vision system runs defect detection on every die, every wafer, every shift — without slowing your throughput.

INSPECTION LOG — Third Shift / Tool T47-B
Timestamp Wafer ID Die Defect Type Conf. FP
03:17:42W-29A-0471X12/Y08PASS
03:17:43W-29A-0471X13/Y08Particle0.94No
03:17:44W-29A-0471X13/Y09PASS
03:17:46W-29A-0472X04/Y14Scratch0.97No
03:17:47W-29A-0472X05/Y14PASS
03:17:48W-29A-0472X22/Y03Edge Chip0.88Yes
03:17:49W-29A-0473X11/Y11PASS
03:17:51W-29A-0473X19/Y06Particle0.91No
Extreme close-up of a semiconductor wafer surface with UV-lit inspection lighting showing die grid pattern
<0.1%
False Positive Rate
120
Wafers / Hour
7nm+
Process Node Support
<8ms
Per-die Latency
The Challenge

Third-shift fatigue creates yield loss you can't afford

Human visual inspection degrades 40% on overnight shifts. A single missed scratch defect on a 300mm wafer can scrap $80,000 in product. Lenspathio runs at full line speed — it doesn't fatigue.

At 7nm and below, defect classification precision becomes non-negotiable. Static threshold methods break down. Our per-recipe adaptive calibration stays stable across tool lifetimes.

False-Positive Control Methodology
SHIFT COMPARISON — Defect Escape Rate
Recipe: 7nm Logic Gate Layer — Tool T22
Shift Human Lenspathio
First 1.8% 0.07%
Second 2.3% 0.08%
Third 3.9% 0.09%
FPR consistent across all shifts vs 117% human degradation
Platform Capabilities

Built for the production line, not the lab

Three integrated subsystems cover every inspection requirement from defect classification to fab equipment communication.

Defect Detection

Convolutional models trained on semiconductor defect libraries — scratches, particles, edge chips, pattern deformations. Classifies and maps to die coordinates.

23 defect types

Line-Speed Throughput

120 wafers/hour without inspection bottleneck. Parallel inference pipeline processes each die frame in under 8ms. No queue backup.

Throughput architecture

SECS/GEM Integration

Native SECS/GEM host communication. Pushes defect maps directly to your fab MES without middleware. Qualifies in standard ATC test framework.

Integration guide
The 0.1% Commitment
<0.1%
False Positive Rate

False positives are yield loss disguised as quality

Every false positive pulls a good wafer for re-inspection. At scale, a 2% false-positive rate on 500 wafers/day wastes 10 engineer-hours. Lenspathio's adaptive threshold calibration keeps false positives below 0.1% — verified per recipe per tool.

How We Control False Positives
Abstract visualization comparing false positive rates in wafer inspection — conceptual data visualization in indigo and violet tones
Process Node Compatibility

From 28nm mature node to 3nm leading edge

Inspection sensitivity scales with node — we tune model resolution and threshold profiles per recipe. Pre-qualified recipes ship for common process nodes.

28nm 16nm 10nm 7nm 5nm 3nm

Custom recipe qualification available for non-standard nodes and memory / logic hybrid architectures.

From the Fab Floor

Engineers who have run evaluations

Names and fab identifiers withheld per NDA. Role and facility type provided for context.

"We were seeing 2.4% false positives with our legacy AOI. Lenspathio brought it under 0.08% within three recipe cycles."

"Third-shift inspection parity was the requirement. We validated it against first-shift baseline — detection rates matched within 0.3%."

Technical Resources

Engineering documentation for your evaluation

White Paper

False-Positive Rate Methodology in Automated Wafer Inspection

46-page methodology report covering adaptive threshold calibration, per-recipe validation protocol, and benchmark comparison against human inspectors.

View White Paper
Application Note

SECS/GEM Integration Checklist for 300mm Fab Environments

Step-by-step host communication configuration, ATC qualification procedure, and MES push verification.

View Application Note

Evaluate Lenspathio on your wafer set

We run pilots on customer-provided wafers with your process recipes. Typical evaluation timeline: 3–4 weeks from wafer receipt to full report.