Built by engineers who ran the inspection tools
In 2019, Diana Muller was running process integration at a 300mm logic fab, managing recipe qualification for the 16nm-to-10nm node migration. The inspection system they were using — a threshold-based AOI from a major OEM — was generating false-positive rates above 2% on gate layer inspection. Every day, yield engineers were reviewing wafers that had nothing wrong with them. On third shift, the escape rate climbed further because operators couldn't keep up with the review queue. The defects that actually mattered were drowning in nuisance flags. Diana spent three years trying to tune static thresholds into acceptable behavior. They never got below 1.8%. In 2022, she left to build a system that solved the problem differently.
The 0.1% standard
Our technical goal is simple: false positive rates below 0.1%, consistent across every shift, every recipe, every tool. Not as a marketing number — as an engineering contract with every facility that runs our system.
We achieve this through per-recipe adaptive calibration rather than static thresholds. It's more complex to build and maintain — but it's the only approach that stays stable as tools age and process conditions drift.
Lenspathio is not a general-purpose machine vision system adapted for semiconductor use. The model architecture, training data, calibration logic, and deployment model were designed from the ground up for wafer-level defect classification in production fab environments. If you need flexible vision for unstructured inspection scenarios, we are the wrong tool.
Evaluation-first commercialization
We don't sign contracts before you've seen the data on your wafers. Every engagement starts with a 3–4 week on-site evaluation where we run blind classification against your gold standard defect maps. You see the false-positive rate before you commit.
This slows down our sales process. It's the right approach in a precision manufacturing context — engineers making capital deployment decisions need to see performance data, not just claims.
Engineering background, not startup background
The founding team comes from semiconductor equipment and industrial computer vision. We've run inspection tools, written process qualifications, and sat through ATC cycles.
Diana Muller
12 years at semiconductor equipment companies. Previously ran process integration at a 300mm logic fab. MS in Electrical Engineering, optical sensing systems focus.
Marcus Holt
10 years in industrial computer vision. Developed CNN-based defect classification for PCB and solar cell manufacturing. PhD in Computer Vision, University of Colorado Boulder.
Start with an evaluation
3–4 week pilot on your wafer set. Your process recipes, your defect maps, your baseline to compare against.