Anonymized case study. No customer or product is identified.
The problem
Inspecting a patterned reticle is already difficult. Adding a thin transparent membrane (a pellicle) over it introduces an entirely new set of challenges.
Challenges
- Transparency and unwanted reflections.
- Contamination and surface particles on multiple planes.
- High throughput requirements.
The two-plane problem
The core difficulty: a particle on the membrane and a particle on the pattern below look nearly identical in a single image, but only one of them matters in the same way. The two surfaces sit a few millimeters apart, which turns out to be the key.
- Focus separation - with a moderate-NA optic, the depth of field can be made smaller than the plane separation. Focus on the membrane and its particles are sharp while the pattern blurs into background; focus on the pattern and the membrane contamination becomes a soft haze. Two focused captures classify which plane a particle lives on.
- Lighting separation - dark-field grazing light skims the membrane and lights up its particles brilliantly while barely reaching the pattern underneath. Combined with a bright-field image of the pattern plane, each surface gets its own dedicated view.
Neither trick is exotic. The work was in making both run fast enough and proving the classification was right on real contamination, not lab dust.
What made it hard
The inspection challenge extended well beyond optics. To get it right you also had to understand:
- The process flow
- How the part is handled
- Cleanroom constraints
- Production expectations
Handling dominated everything. The part cannot be touched, barely tolerates being moved, and every transfer is a contamination opportunity - so the inspection had to come to the part, integrate with existing handling, and never become the riskiest step in its own process.
Lessons learned
The optical problem was solvable. The systems problem - handling, cleanliness, and process flow - was where most of the real risk lived.
When evaluating any inspection task on high-value parts, I now estimate the handling risk before the imaging difficulty. If inspecting the part is more dangerous than not inspecting it, the imaging quality is irrelevant.