// Showcase

Fiducial detection on mirror, glass and hologram

Detecting features from 5 µm to within ±1 µm, on polished nickel, clear film, polycarbonate and a hologram that throws back a different colour for every angle.

Photonics & Laser Semiconductors & Microfabrication

01Starting point

Before every imprint, a step-and-repeat machine needs to know where the substrate, the tool and the most recently created structure are, and that is what fiducial mark detection answers. The requirement was ±1 µm accuracy and ±0.25 µm repeatability on features as small as 5 µm. Standard machine vision on its own, if it were not for four materials that behave completely differently: polished nickel is a mirror, polyester film is transparent, polycarbonate also reflects off its back surface, and a hologram changes its appearance with every viewing angle. And all of that on a moving 200 kg head whose own axes are only accurate to ±5 µm, so the vision system has to be five times better than the mechanics carrying it.

02Approach

We packed two magnifications onto one head: 1x at 5 µm per pixel to find the mark across a large field of view, and 10x at 0.5 µm per pixel to measure its centre. Telecentric, plan-apochromatic optics keep the measurement position from shifting even as the working distance changes with substrate thickness from 0.5 to 5 mm. Each channel switches independently between ring and coaxial light, so four materials become four recipes instead of four different hardware setups. A custom-built controller keeps everything within the 21 W PoE+ power budget, even though the illumination alone can draw up to 17 W.

03Result

Measurement happens in the same coordinate system as printing, right in the machine rather than on an external measuring machine. That saves unclamping and reclamping, which is itself a source of error. We did not check the whole thing against ourselves, but against an externally measured fiducial grid: ±1.0 µm mean deviation and ±0.25 µm standard deviation, both requirements met. Every measurement also leaves the machine as an inspection report with a target-actual comparison, recipe and comment, not as a screenshot.

What is inside

A complete machine vision head: two monochrome GigE cameras with Power over Ethernet, telecentric optics in two channels, our own illumination controller, the detection algorithms plus the image database behind them, and evaluation against an externally measured reference grid. Add to that the inspection function with target-actual reporting, and an extra confocal distance sensor on the same head for height and thickness.

What happened next

One cable per camera, data and power together, means fewer failure modes in the cable carrier, which travels over 1600 × 1400 mm² and hundreds of thousands of cycles. The measurements from this head have since become the basis for the placement analysis of the whole machine.

Get in touch

We are ready for new challenges

You bring the challenge. We develop the solution. From the first idea through concept and development to the validated machine, we guide your project.