Contributor Post Created with Sketch. Recommended by Ricochet Members Created with Sketch. How to Build a Computer 14: Alignment

 

Last time we saw how you physically expose a panel. That is, how you shoot it with ultraviolet light to get a pattern into the stuff so that you can do things to that pattern later on. Today the plan is to talk about all the ways this can go wrong. We’ll start with the big one: alignment. If you’ll recall the profile of the jumping trace we looked at a couple weeks ago:

Hooray for a well-stocked media library!

See that trace on the top? Suppose you were to shift it over to the right. Eventually, you’d lose contact with your left via and you’ve got a hole in your wire. Busted circuit, sorry, can’t sell that one. Now imagine you’re shifting it forwards or backward; sooner or later you lose contact with your via and again you start making scrap. Or twist it side to side. Or shift it and twist it. Suddenly you’re wondering how they get these things on there at all. Don’t worry, it gets worse. Suppose both the vias and the top trace are aligning to the bottom traces. The vias get printed in an okay spot, but a little south of where they ought to be. Still in tolerance. The trace gets placed in its own okay spot, but a little north of where it ought to be. See where I’m going with this? The compounding of the two errors is enough to, again, cost you money. The problems compound when you have a second phototool on the bottom to align as well.

To print these things, and not be scrapping them left and right, one has to align your layers very carefully together. And when I say “align your layers” I mean align your phototools such that the image they leave is going to align to the previous layers, which were made with previous phototool images. The humble exposer is therefore responsible for aligning all the layers together. (I’d say it doesn’t get appreciated enough but I’ve had to work with those machines. They’re ornery cusses.)

There are a couple ways that your phototool can be misaligned. The first is a translational error:

The time of Peanut Butter & Jelly is upon us!

The second is a rotational error:

No baseball bats were harmed in the making of this photo.

There are more errors you can have when you consider the third dimension, which we don’t. The third dimension is for people who are showing off. We’ve got enough complications as it is. But let’s say that your phototool is misaligned. How would you know? Let me zoom in on one corner of the one from last week:

I’ll forgive you if you missed it on the first go-around. The larger diamond indicates that this is a bottom phototool.

See that little diamond there? That’s an alignment mark. When we start a web we punch holes in the four corners of each panel. In your autoexposer you’ve got four cameras (nine, actually, but that’s a detail) that look down through those holes looking for the alignment marker in each corner of the phototools. There’s another alignment mark on your top phototool that you have to line up too. Here, take a look:

See? It’s aligned. I don’t know what you’re making a fuss about.

That’s a photo off an actual expose run. The two diamonds are on the phototools and the circle is actually a hole punched in the stainless steel. You’ve got a light down below (red so you don’t accidentally add extra photons) that shines upwards and a camera up top that’s looking down through it. Here, let me draw it out for you:

Rather than four holes in the corner of a panel we’ve got one big one in the middle. Because that’s easier to draw.

That thing that looks like one of the pipes from Mario? that’s a camera. There’s one pointing down (which is used) and one pointing up (one of the ones we’re ignoring.) Also on the camera subassembly you’ve got a red LED light. The one up top is turned off. The one down below is turned on. The blue parallelogram represents a panel, and the circle in the middle of it is your fiducial hole. You can tell where your parts are on the panel in relation to that hole, and you can tell how your phototool is aligned in relation to the diamonds. (Small one on top and a large on the bottom.)

You don’t see blue resist or red lights in the camera; it renders it all in black-and-white. Photoresist covers the hole but it’s still easier to see light through the photoresist than through the stainless steel and photoresist. The computer takes that image and calculates three positions off of it (the centers of the circle, the big diamond, and the small diamond respectively).

Let’s imagine you’re running through this calculation. You know where your big diamond is, your little diamond is, and your circle. You note your big diamond is shifted four to the left. (Four what? The computer isn’t giving you units; you’re supposed to know them already.) Okay, just shift it back four and… not so fast.

Remember how there are four cameras taking four images, one on each corner of the panel? Well, it’s coming back to bite you. This camera is four whatever to the left, but the one in the opposite corner is only three to the left. To make it worse The others are at 3.5 and 1.5 respectively. You can move it, but you’ll never get them all to agree to zero. You just minimize the errors. Here’s an action shot of an exposer aligning:

You’d pay good money for this kind of action at a Vegas Casino. Here you’re getting it for free!

It’s only aligning off the top tool (only small diamonds, no big ones). See the green lines around everything? That’s to let you know where the computer is seeing the edge of the features. If you look closely at the scratch on the lower right one you can see a red dot on that circle; that’s where it expects the edge of the hole to be but it isn’t, because it’s confusing the black background material with the black scratch.

Note the position of the diamonds. The green crosses tell you where the center of a feature is. Those small diamonds are all shifted to the right and a little bit down. That’s a translational error. Relatively easy to take care of. The math is harder when you’re talking about rotational errors. I’ve repented of my earlier intention and I’m not in fact going to go through the equations, but this is how you’d do it.

First, define a coordinate system that encompasses the position of the panel. You can figure out the position of the panel from the location of the four circles. Second, do the same thing for your phototool. Since your tool is going to be twisted these systems don’t line up. It’s your job to make them line up. Draw a line from the upper left corner to the lower right corner on both. All you have to do now is calculate the angle between the phototool line and the panel line, and then rotate your phototool to match the panel.

Okay, you’ve got your calculations complete. You know where you want to move your phototool. (Oh, and by the way, when I say “move your phototool” I mean it’s stuck to this big heavy granite frame. The thing the cameras are mounted to. The cameras are on rails; they’ve got to slide in and out. Wouldn’t do to leave the shadow of a camera imaged into your parts.) Okay, you move the frame into position, and now you’ve got to check that it’s in the right place again.

Run through the math again. Are you within tolerances? You’re still going to have small amounts of error in all directions, but if they’re small enough you can live with it (that is, still be making good parts.) We good? Snap open the shutters, let the UV light burn the photoresist for a specified number of seconds, snap them shutters closed, and move on to the next panel. Lift that granite chunk so it isn’t going to scrape the web as it travels through, advance exactly 308 millimeters to get the next panel in place and start over.

Now here’s where we put the “auto” in autoexposer. If you set it up good it’ll do all this automatically. Advance, lower the frame, adjust, adjust, adjust, expose, raise the frame, and repeat. How long do you think it takes? I mean, when the stars are aligned against you, when you can’t get a good alignment, when there are wrinkles in the panel that screw with your locating, I’ve taken over an hour to shoot a single panel before. And every time it errors it goes “BEEP BEEP BEEP” to let you know that it’s erroring even though you’re standing right there. I was a support guy; the production folks who take care of these machines twelve hours a night? They hear BEEP BEEP BEEPing in their dreams. Okay, but when things are running smoothly? How long does it take to cycle through if it doesn’t error out?

About twenty seconds.

Join us next week when we get to the develop process in “Carbonite Dreams” or “Hey, I Paid Good Money to put Photoresist on this Thing and Now You’re Taking it Off?”

This is part fourteen of my ongoing series on How to Build a Computer, the PB&J way. You may find previous parts under the tag How to Build a Computer. This week’s post has been brought to you by Silver Spring Condiments. Silver Spring makes the finest Beer ‘n’ Brat mustard around. Try it at your local Eau Claire ball game. Silver Spring!

[First] [Previous] [Next]

There are 13 comments.

Become a member to join the conversation. Or sign in if you're already a member.
  1. GLDIII Temporarily Essential Thatcher
    GLDIII Temporarily Essential Joined in the first year of Ricochet Ricochet Charter Member

    Not as succinct as last installment raves audiences….

    – NYT

    • #1
    • September 1, 2018, at 6:22 PM PDT
    • 5 likes
  2. Judge Mental, Secret Chimp Member

    They have to do math? Why not just two knobs, like on an Etch-a-Sketch? One for up and down, the other for right and left; turn them both at the same time and go any direction you want. Look through a spy hole and line it up.

    • #2
    • September 1, 2018, at 6:26 PM PDT
    • 3 likes
  3. Hank Rhody, Badgeless Bandito Contributor

    Judge Mental (View Comment):

    They have to do math? Why not just two knobs, like on an Etch-a-Sketch? One for up and down, the other for right and left; turn them both at the same time and go any direction you want. Look through a spy hole and line it up.

    Once you’ve taught the machine to do the math for you it’s quicker than getting it to shift into position. I think. These are rather stupid computers on those machines.

    • #3
    • September 1, 2018, at 6:37 PM PDT
    • 2 likes
  4. Boss Mongo Member

    Thanks, Hank.

    I’ve thoroughly enjoyed your series. And one lesson is the foundation, easily achieving primacy over all the others: I will never, ever build a computer.

    At least it gives me talking points to sound like I’m smart, though: Ah, well y’see, what you have here is a basic translational misalignment of your PB&J…

    • #4
    • September 1, 2018, at 6:50 PM PDT
    • 7 likes
  5. Gary McVey Contributor
    Gary McVey Joined in the first year of Ricochet Ricochet Charter Member

    Using a computer to align 3D stereoscopic images has many of the same issues. The images can be too far or close side by side, though the eye can compensate for that. Even the slightest vertical misalignment can’t be tolerated, though, so the software is particularly vigilant about that.

    What’s really tough are rotational errors, which creep in because the old slides weren’t mounted in their frames with digital precision. They require tons of calculation to correct, and if you don’t catch them the results can be comic–as long as you’re not tragically close to a deadline. 

    • #5
    • September 1, 2018, at 9:41 PM PDT
    • 4 likes
  6. Arahant Member

    I thought we already had thirteen? Shouldn’t this be fourteen?

    • #6
    • September 1, 2018, at 9:49 PM PDT
    • 3 likes
  7. Judge Mental, Secret Chimp Member

    Arahant (View Comment):

    I thought we already had thirteen? Shouldn’t this be fourteen?

    I agree, two thirteens is totally crazy. What’s next, three fourteens?

    • #7
    • September 1, 2018, at 11:07 PM PDT
    • 4 likes
  8. Arahant Member

    Judge Mental (View Comment):
    What’s next, three fourteens?

    It has happened before. Then the universe will require four fifteens.

    • #8
    • September 1, 2018, at 11:08 PM PDT
    • 4 likes
  9. Hank Rhody, Badgeless Bandito Contributor

    Arahant (View Comment):

    I thought we already had thirteen? Shouldn’t this be fourteen?

    Yeah, had this mostly written up before I went with triskadekaphobia. Forgot to edit the number in the title. I’d go back and fix it now but I’m not a fan of realigning all the pictures all the time. I’ll get it when I edit it for that [Next] link on the bottom.

    • #9
    • September 2, 2018, at 8:43 AM PDT
    • 1 like
  10. OldDanRhody (this comment has … Member

    Judge Mental (View Comment):

    Arahant (View Comment):

    I thought we already had thirteen? Shouldn’t this be fourteen?

    I agree, two thirteens is totally crazy. What’s next, three fourteens?

    We could go with 13-B: It’s been done before, Judge.

    • #10
    • September 2, 2018, at 2:14 PM PDT
    • 1 like
  11. Judge Mental, Secret Chimp Member

    Major Major Major Major (View Comment):

    Judge Mental (View Comment):

    Arahant (View Comment):

    I thought we already had thirteen? Shouldn’t this be fourteen?

    I agree, two thirteens is totally crazy. What’s next, three fourteens?

    We could go with 13-B: It’s been done before, Judge.

    I have no idea what you’re talking about.

    • #11
    • September 2, 2018, at 2:25 PM PDT
    • 5 likes
  12. Percival Thatcher
    Percival Joined in the first year of Ricochet Ricochet Charter Member

    Judge Mental (View Comment):

    Major Major Major Major (View Comment):

    Judge Mental (View Comment):

    Arahant (View Comment):

    I thought we already had thirteen? Shouldn’t this be fourteen?

    I agree, two thirteens is totally crazy. What’s next, three fourteens?

    We could go with 13-B: It’s been done before, Judge.

    I have no idea what you’re talking about.

    Neither do they, so it’s all good.

    • #12
    • September 2, 2018, at 4:32 PM PDT
    • 4 likes
  13. Mack The Mike Coolidge
    Mack The Mike Joined in the first year of Ricochet Ricochet Charter Member

    Those alignment marks suck. Try DBO.

    • #13
    • September 2, 2018, at 6:34 PM PDT
    • Like

Comments are closed because this post is more than six months old. Please write a new post if you would like to continue this conversation.