// Showcase

Sub-micron wafer alignment: why two marks are not enough

Six wafers, two mark types, one machine. In one configuration placement was under 1 µm, in the other 5 to 6 µm. The difference was not in the mechanics, but in the number of measurement points.

Photonics & Laser Semiconductors & Microfabrication

Sub-micron wafer alignment: why two marks are not enough

Photonics & Laser Semiconductors & Microfabrication

01Starting point

A machine replicates hundreds of microlenses onto a glass wafer in one go, and stacking two such wafers gives you a complete lens assembly. The spec: ±1 µm placement, because the errors from both layers add up. But a spec is, at first, just a promise. So six 8-inch wafers with two different alignment marks ran through two processes, once face-to-face for lens replication, once back-to-back for stacking. The question: where does the machine actually stand, and if it is off target, why?

02Approach

First we checked the marks themselves: 500 images per type, with scatter staying within one pixel, so around ±0.6 µm. Each mark also carries a vernier scale, letting you read off the residual offset directly in the camera image, independent of the machine. Face-to-face: placement after bonding under 1 µm, spot on. Back-to-back: suddenly 5 to 6 µm, even though the machine had measured under 1 µm one step earlier. The reason: with two marks you can see offset and rotation, but not tilt between the surfaces. That takes three points.

03Result

A capacity claim that is actually measured, and for the case that does not hold up, the exact cause instead of a vague number. The fix costs next to nothing: no new mechanics, no rebuild, just a third focusable feature in the mark layout, for example a lens that is already there anyway. That lets you measure the tilt, and the compensation the machine can already do kicks in. What looked like a machine weakness turns into a change to the consumable.

What is inside

A complete measurement report: detection accuracy across 500 images per mark type, the verified replication case on real material, the geometrically sound diagnosis of the stacking error, and a concrete, low-cost correction recommendation. Plus an honest note on what the campaign proves and what it does not: six wafers, one campaign, no repetition across multiple days.

What happened next

The ±1 µm spec for the replication case now counts as verified on real material, backed by six wafers. For the stacking case the fix is ready to go, as soon as the mark layout gets a third focus point. A change to the consumable, not to the machine.

All showcases

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.