// Showcase

Finding the mark: edge finding versus pattern matching

A machine that joins a glass wafer and a tool to ±1 µm first has to know where both parts are.

Photonics & Laser Semiconductors & Microfabrication Sensors & Metrology

Finding the mark: edge finding versus pattern matching

Photonics & Laser Semiconductors & Microfabrication Sensors & Metrology

01Starting point

A machine that joins a glass wafer and a tool to ±1 µm first has to know where both parts are. That is what alignment marks are for, and there are two very different ways to find them. One locates the mark's edges and intersects them; the other is taught the mark as a pattern and searches for it again. The acceptance criteria were clear: ±1 µm in x and y, ±0.5° in angle, each at one standard deviation. The catch: at 0.6 µm per pixel, one micrometre is less than two pixels wide. That makes it a subpixel problem from the very first line of code.

02Approach

We ran both methods against each other on the same marks, hundreds and thousands of times, and logged not just the scatter but the detection rate as well. The result was uncomfortable. Edge finding measures to 0.04 µm in X: one fifteenth of a pixel. But depending on the direction of the light, it found the mark 99 times, 100 times, or only 37 times out of a hundred. Pattern matching returned 1000 out of 1000 in three runs, without a single miss, at 0.18 to 0.64 µm.

03Result

Instead of a winner, there is a division of labour. The robust pattern search carries the process down to a gap of a couple of hundred micrometres, where a few micrometres still cost nothing. Edge finding takes over for the last hundred micrometres, because that is where the angular criterion tightens by three orders of magnitude, and only a line fit across three calibrated cameras gets there. The customer ends up with a machine that processes every part and still delivers the accuracy it was bought for.

What's inside

A measured head-to-head of both methods with detection rate, standard deviation and full range in x, y and angle. Plus a vision acceptance list with 68 numbered test cases that states, for every process phase and every layer stack, which method applies at which tolerance: from a bare wafer on the chuck to tool glass with epoxy in the beam path.

And what happened next

The biggest surprise was how much the lighting weighs: same pattern search, same marks, only a shallow instead of a steep angle of incidence, and the scatter goes up by a factor of three and a half. And it is the angle, not the position, where the two methods are furthest apart: a factor of seven to twenty-five. Illumination geometry has since stopped being a commissioning note here and become part of the acceptance list: with every test case naming its layer stack.

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.