Fiducial points for better pnp precision

[liudr]

I have been using a desktop pnp for a couple of years and one thing that my pnp is bothering me is its inability to automatically identify fiducial points on a board on a panel of boards. I would move its head to one board on the panel, snap a photo of two fiducial points. Then it can automatically adjust with the rest of the boards on the panel, in theory. In reality, it starts to fail after a few boards. I suspect that it was due to the fact that my fiducial points are shiny tin-plated discs from the leadfree HASL process. The downward camera on my pnp has an array of about 4 LEDs so the illumination on the board is less than uniform, which causes the mirror-like fiducial points on each board to reflect light differently from the next, causing comparisons to fail. My thinking is, if I either cover the fid with solder paste (add to cream layer), which is going to be a diffuse scatter of light instead of specular reflection like a mirror, then it would be easier to auto identify the fid pts. Or maybe I should use silk screen or exposed copper instead. What do you all do when it comes to making fid pts and using pnp to auto identify them? Thanks.
BTW, my PCB camera also randomly goes into black-white video but luckily I almost exclusively use black PCB so I don't always notice.

[IconicPCB]

What does Your machine manufacturer recommend as the fiducial parameters?

Say 1mm pad on a 3mm solder mask and copper pour clearance?

[liudr]

There wasn't any recommendations. I could adjust the recognition area sizes. I'll reach out to them to see if they would answer any questions.

[selcuk]

You can use a most common fiducial design as recommended here:

https://jlcpcb.com/help/article/48-How-to-add-edge-rails/fiducials-for-PCB-assembly-order

Solder mask opening diameter is 2mm in the dimensions but I think they used 3mm in those images. That is what I usually use as well.

[liudr]

Thanks @selcuk. I read the description. I happen to use jlcpcb myself but have not used their PCBA yet. If I understand correctly, the exposed copper disc with a larger solder mask hole will just result in a nice shiny disc on the board, which is the issue I am having with my pnp. The specular reflection conditions on each board is slightly different, causing recognition to fail. That's what I understand, unless I'm mistaken. Here is what I am using:
You can see that the copper diameter is 0.635mm, less than the 1mm recommendation. Maybe larger discs will have more uniform surfaces? Or maybe I just tick the "cream" box and order a new unframed stencil, 5 single boards, try my pnp recognition at different parts of my pnp's pcb holder to see if there's any improvement? I've been using panels and framed stencils so I'm kind of reluctant to make new stencils, thus wasting existing ones.

[selcuk]

I think p&p machine wants to detect the fiducial among its surrounding copper. You have very small copper free area around the fiducial. When you have a 25mil pad in diameter, you may use a 75mil diameter keep out for copper plane and solder mask. Check the images on the JLCPCB site. The empty area is so wide that you can place the same fiducial pad around it on both left and right without touching outer copper and silk screen.

I'm not familiar with the cream option. I think you can use the same stencils after altering fiducials on the top copper and silk screen layers for the next PCB order.

[liudr]

Thanks. I found an EAGLE library online that has fiducials. Based on my understanding and analyzing how the fiducials were made in that library, I made a new fiducial marker that I believe will work.

The marker has a copper disc with 1mm diameter and a pull-back area about 3mm diameter (showing fiber substrate). So a 1mm diameter filled circle, a 3mm diameter top-restrict disc, and a 3mm diameter top solder resist mask hole is what I have. The ones I was using came from an old sparkfun library. After analyzing that library part, I realized that it was made incorrectly, having a round SMD pad with a default pull-back and a top soler resist mask, which resulted in that concentric disc pattern I posted. I'll order some panels with this new fid design and see if they come out correctly and post an update once I have the panels.