super-cheap pick-and-place device with ~1 mil accuracy
Moderators: adafruit_support_bill, adafruit

super-cheap pick-and-place device with ~1 mil accuracy

by bootstrap on Tue Jul 27, 2010 9:48 pm

I have an idea how to make an super-cheap pick-and-place machine that is accurate enough to place BGAs and QFNs with 0.40mm ~ 0.50mm ball/pad pitch. But I'm currently sans machine shop, so I'm looking for someone to partner with who can machine a few parts to build a couple prototypes, then maybe sell them. My guess is, we could sell hundreds to thousands per year at about $400 ~ $500, because it lets hobbyists, one-man operations and nano-companies design and reliably assemble prototype PCB with fine-pitch SMT components (BGAs, QFNs, etc).

My goal is to make a device for people who cannot afford or justify an automatic pick-and-place machine. This includes me, at the moment. I don't want a partner driven primarily by profit. I want a partner who wants to do this for himself and/or likes the idea of introducing a product that breaks loose creativity that's been bottled up by the difficulties and expense of fine-pitch SMT assembly. But the idea is to make something radically cheap, not to line your pockets with cash. Profit? Yes, a little. Wealth? Nope. Incidentally, this approach assures we don't have competition from jerks just trying to line their pockets. Once we're done, maybe we'd sell through places like here, pololu, sparkfun and others. Whatever makes it easy to find any "buy cheap".

Rather than file for patents, play "secret", or ask for confidentiality agreements, I'm gonna treat this as an open/community design. I will describe the basics of my idea here in this message, then ask for ideas to make the device cheaper, more flexible, more accurate, more convenient to operate, etc. Let's see if we can come up with something that "blows their minds", and more importantly, unleashes an avalanche of creativity in untold thousands of new projects.

Before I describe the device, let me explain the philosophy of this design. Pretty much any of us with good vision, magnifier or low-power microscope and reasonable dexterity can reliably solder most SMT components with nothing more than a pre-heater beneath the PCB, a hot-air pencil/gun above the PCB, and either manually-applied solderpaste or stencil-guided solderpaste application. This includes all components with visible/exposed contacts (QFPs, TSSOPs, SOICs). I'm even able to place and solder 0201 capacitors with a vacuum pick-up pencil, though it would be nice to make 0402 and smaller devices easier to handle.

But we cannot reliably place fine-pitch components with non-visible/non-exposed contacts on the PCB, including BGAs, QFNs, iLCCs, etc. Therefore, the goal is to make a very inexpensive but extremely precise and very reliable device to place the components we cannot reliably place otherwise. If it can help place other components too, that's fine. But we should not significantly increase the complexity or cost of the device to support components that can be reliably assembled without this device.

Now I'll describe the design. I'm sorry to not include a sketch yet, but I'd rather not bias the conversation with specific implementation details and lock your vision into mine. We'll get to that after a day or two of hearing configurations you envision to accomplish our objectives. For now, I'll describe the configuration fairly specifically, but try not to specify the design more narrowly than is necessary to explain its principles (which is all that matters to make it work).

First, what the design does NOT have - which is partly why the device can be cheap. The design requires no motors, no linear or rotary encoders, no electronics, and no power-supply (maybe/hopefully).

Presumably the device will weight 2~6 kilograms (5~15 pounds), and be about 300mm long by 100mm wide by 200mm high, though the 300mm dimension would need to be longer to support large PCBs.

Okay, now for a description. The device is a "tabletop device" --- it must be placed on a flat, smooth, reasonably level tabletop to operate properly. Imagine you are sitting at the table with the device on the table in front of you. The device is a single linear "track" ~300mm long (left-to-right == X-axis) and ~100mm wide (toward-and-away from you == Y-axis). This linear track has 2 feet on its left end and 2 feet on its right end that raise the bottom of the track ~20mm above the tabletop (up-and-down == Z-axis).

I usually visualize this track as two stainless-steel "guide bars" forming the X-axis, attached to fairly thick, solid, aluminum rectangle bases at both ends, with 2 feet attached to the bottom of each base. Mounted to the "guide bars" is a rectangular aluminum "stage" with 2 linear bearings sliding on each "guide bar".

So far, you should be able to visualize that you can slide this stage freely along the X-axis --- leftward and rightward. I usually visualize this stage being about 100mm long (X-axis) and 50mm wide (Y-axis).

The stage holds two devices. Mounted on the left side of the stage is a 1280x1024 (or better) webcam, pointing straight downward along the Z-axis. Mounted on the right side of the stage is a "vertical slider" (for lack of a better term) that you can move up-and-down ~20mm. In the middle of the "vertical slider" is some kind of suction/vacuum based "component holder". If you place a BGA, QFN or other component on the table beneath the "component holder", then lower the "vertical slider" until it touches the component, then raise the "vertical slider"... the component will be held by the "component holder" and rise up with it.

The webcam side of the "stage" is quite simple and straightforward. However, the "vertical slider" and "component holder" can both be implemented in many different ways (but perform the same functions). The only significant requirement of the "vertical slider" is: the component must not rotate when the slide moves up and down. The "component holder" could be something we provide as standard equipment. But the "component holder" is nothing more than a simple vacuum pick-up-tool that is mounted to the "vertical slider". So I usually imagine we provide some kind of simple adjustable mounting gizmo that lets customers mount their existing vacuum component pick-up-tools, which then function as the "component holder". Of course, we can still make or find a cheap one to provide to customers who have no vacuum pick-up tools. Sometimes I even imagine a big red squeeze bulb on top to generate the vacuum --- talk about cheap! Yet that's all we need for this device.

Okay, that's almost the entire device... believe it or not. The only other component of the system is a flat sheet of plastic/teflon that lies on the table beneath our device, with some holes drilled near the end closest to us. Believe it or not, this is all we need to place SMT components to 1~3 mil precision... assuming you've already figured out the simple procedure that makes it work! Just in case you haven't, I explain that next.

Some of the following elements aren't entirely necessary, and the process can be performed somewhat differently, but the principle remains the same. So keep your minds alert and active, looking for superior ways to implement the necessary process.

First I'll describe the process with a BGA, but the process for QFNs, iLCCs and other "hidden contact" packages is almost identical. You can process small to medium size PCBs when sitting down, but you may want or need to stand up to process large PCBs.

Near the front edge of the teflon sheet are x,y arrays of 10~25-mil diameter holes, plus several square and rectangular holes with dimensions between ~2mm and 64mm. Usually I visualize this 50mm~100mm wide front section of the teflon sheet is actually a separate ~1.6mm thick sheet of aluminum or stainless attached to the teflon sheet, but that's just my visualization (albeit for non-arbitrary reasons).

The hole arrays are for BGAs, and the larger squares/rectangles are for QFNs. There is at least one hole array with 1.27mm pitch, 1.00mm pitch, 0.80mm pitch, 0.75mm pitch, 0.65mm pitch, 0.50mm pitch, etc. Each hole array or has a ~5mm boundary around it without holes or openings.

For convenience, I'll assign a number to each step, so you can discuss them more easily in your replies.

#1: Place PCB on teflon sheet, slide it towards you until the near edge is flush against the metal sheet that has the holes and rectangular cutouts.

#2: Optional: slide the PCB leftward until its left edge if flush against the bar mounted along the left edge of the teflon sheet.

#3: Secure the PCB to the teflon sheet with tape (maybe we provide clips or something more convenient). Some of your are probably thinking "Why not just attach a thin layer of rubber or something less slippery to the top of the teflon sheet?". That's probably a good idea, since we want the sheet to slide on the tabletop easily, but we don't want the PCB to slide around on its top surface.

#4: Move the X-axis "stage" to its extreme left position (where it can't move any further), and lock it in place (if detents aren't sufficiently reliable). Note: though the X-axis "stage" is in the middle of a 300mm long X-axis "track", it can only move 50mm left or right before it reaches its limit (or detent).

#5: Place the BGA component on the hole array that has appropriate pitch for that BGA component. Be sure to place the BGA component so it is justified in the upper-left corner of the appropriate hole array. The easiest and most reliable way to accomplish this is to place it about one pitch unit too far left and high, so the upper-left solder balls have no holes beneath them, and therefore the upper-left corner of the BGA component cannot "drop into" the holes. Then very gently, and without any downward pressure, slide the BGA downward and rightward until you feel and see the BGA package nicely "detent" into the hole array. As long as you are gentle, you will not tear off any BGA solder balls.

#6: Slide the teflon sheet on the tabletop until the center of the pickup device is directly over the center of the BGA component. It is not necessary for the center of the pickup device to be precisely over the center of the BGA component, but it must be close enough to assure the BGA component is securely held by the pickup tool. The rotation angle of the component doesn't really matter, but try to be nice and neat and keep the X-axis of our device (the X-axis "stage") parallel with the horizontal rows of BGA balls.

#7: Apply vacuum to the component pick-up tool (squeeze the big red rubber bulb, hahaha), lower the "vertical slide" until the pick-up tool suction-tip is firmly pressed against the BGA component, then raise the "vertical slide" and verify the BGA component is firmly and securely held to the suction tip.

#8: Move the X-axis "stage" to its extreme right position (where it can't move any further), and lock it in place (if detents are not sufficiently reliable). Note: The X-axis "stage" is designed so that moving the stage to its "extreme left" and "extreme right" positions places the center of the pickup-device and center of the webcam directly over the same point. While we could easily machine the parts to assure this is "exactly true", the process does not require this to be precise to place components precisely. This means the gizmo that holds the vacuum pick-up tool need not hold it exactly in the correct X,Y position.

#9: Look at the LCD screen on your tabletop and hit the space key on your keyboard. Since the webcam is looking directly at the hole array where the BGA was picked-up from, pressing the space key displays the array of holes the BGA was sitting in near the center of your LCD. This step assumes the keyboard and LCD attached to your PC is on the same tabletop, but this is not a requirement, of course. Of course, the webcam must be plugged into the PC, and you must be running software designed to help us perform this process.

#10: Now slide the teflon sheet on the tabletop to position the BGA footprint on the PCB directly beneath the webcam. As you look at your LCD and slide the teflon sheet around, you'll notice the software displays two images simultaneously (each image at roughly half intensity - though each intensity is adjustable). One image is the still image that was captured when you hit the space key, which displays the hole array where you originally placed the BGA component in that hole array. The other image is a realtime display of what the webcam sees --- which is the now pads on the PCB where you want to place this BGA component, plus surrounding area of the PCB (typically 80mm x 64mm or smaller, depending upon what is the largest SMT component you need to support).

#11: Slide the teflon sheet on the tabletop to exactly position the two images over each other, so the holes in the hole array (the still image) exactly superimpose over the BGA pads on the PCB. Make sure the upper-left hole in the hole array is superimposed with the upper-left BGA pad on the PCB - otherwise you'll solder the BGA offset in X and/or Y by one or more pads! You can remember to do this, right? You better!

#12: Somehow secure the teflon sheet to the tabletop. In practice, the teflon sheet should have enough friction that no steps need be taken to secure the teflon sheet to the tabletop. But if your tabletop is too slippery, or you tend to bump into things without noticing, or it makes you feel better... place a couple weights on opposite corners of the teflon sheet, or tape the teflon sheet to the table. Then check that the images are still perfectly aligned on your LCD.

#13: Move the X-axis "stage" to its extreme left position (where it can't move any further), and lock it in place (if detents aren't sufficiently reliable). This positions the BGA component precisely over the BGA pads on the PCB. Do you understand why this is precise, even though we have NO encoders? If not, re-read the whole process until you "get it".

#14: Lower the "vertical slide" until the BGA component is touching the PCB, release the vacuum in the pick-up tool to release the BGA component, then raise the "vertical slide" while watching the BGA component to assure it doesn't move during this process.

That's pretty much the entire process. Of course, I leave out questions like "Do we place all our SMT components to the PCB before we solder them, or do we solder each component after we place it?". The answer to the above is probably, "If you plan to solder each SMT component by hand with a hot-air pencil, then you're probably better off soldering each component immediately after you place it. But if you plan to place the PCB into a reflow oven to perform soldering, then you're probably better off placing all (or many) SMT components (on one side) before you carefully place it into your reflow oven.". Clearly you must be careful not to bump the PCB or components from when you place the first component until after the solder has hardened. Otherwise, the precision is lost.

Now for a few random comments.

A: It is quite simple to make a "limit slide" that surrounds the X-axis "slide". You could then move this "limit slide" ALL the way leftward and rightward on the "guide bars" (300mm or more) to more easily support larger PCBs. Because it surrounds the X-axis slide, the "limit slide" is what limits the X-axis slide to moving exactly 50mm. The only difference is, now that 50mm range can be anywhere along the 300mm "guide bars" instead of "only in the middle".

B: We may want to make the piece of metal that contains the hole arrays and rectangular cutouts slide leftward and rightward to position the desired component template where convenient (which nominally near the center of the X-axis range, unless we provide the adjustment mentioned above in comment A). Alternatively, we can provide each hole-array or rectangular "template" as a separate small (~80mm) square of metal, then provide a way to secure any one of them to the teflon sheet to place parts that correspond to that template.

C: The pick-up tool doesn't necessarily need to be vacuum based, but whatever mechanism is adopted must not move or rotate the component during or after pickup.

D: Components in QFN, iLCC and even 0402 and 0201 packages can be handled in an equivalent manner with the above process (given appropriate "templates").

E: The teflon sheet could be replaced by an X,Y stage. While that would be rather nice in some ways, especially in keeping the PCB and components aligned with the PCB X and Y axes, it would probably add an additional 50% to 100% to the cost. That doesn't seem worthwhile to me, except as an option --- maybe.

I've surely forgotten to mention many ideas, thoughts and alternatives that I've had while dreaming up this device. Sorry about that, but that leaves everyone more unconstrained by my bad (and good) ideas. I hope this makes everyone more crazy and creative with their suggestions. Let the alternatives fly! Perhaps it is good to say things like:

--- a more sturdy version would .....
--- a more precise version would .....
--- a more flexible version would .....
--- a less expensive version would .....

... where your ideas don't apply [equally] to every possible implementation.

Don't feel constrained. Let the ideas fly.

And anyone ready, willing and able to help me make prototypes, please offer. I'm guessing that 50% to 90% of the components can be purchased as standard parts from the kind of supplier who sells small mechanical components and devices like linear bearings, precision shafts, and so forth. But surely not every part will be a standard item, or will not attach to the rest of the components securely and conveniently without some custom machined parts. So I'm looking for someone who can help with this aspect especially. And, unfortunately, my current project is stalled due to my inability to place components precisely, so I'm most interested in someone who can hit the ground running (can start machining up parts as soon as we settle on designs for each part).
bootstrap
 
Posts: 72
Joined: Wed Jun 02, 2010 8:47 pm
Location: phobos

Re: super-cheap pick-and-place device with ~1 mil accuracy

by mikeselectricstuff on Wed Jul 28, 2010 5:08 am

#7: Apply vacuum to the component pick-up tool (squeeze the big red rubber bulb, hahaha),

You need continuous vacuum to deal with small leaks - a bulb type thing is useless for pick/place, like those cheap 'manual' pick/place pens. Fortunately aquarium type pumps can be adapted to provide a cheap vacuum pump.

I think the whole alignment process is overcomplicated - why not align the package based on the image of the body from the top side, i.e. you have a fixed overhead camera, produce a composite image of the stored image of the PCB pads, overlaid on a live image of the top of the package, Add some software smarts to detect and display the package and PCB footprint centroids, and maybe produce an audio feedback representing the deviation between the 2 images so you can align without looking at the screen?
mikeselectricstuff
 
Posts: 164
Joined: Fri Jun 11, 2010 9:21 pm

Re: super-cheap pick-and-place device with ~1 mil accuracy

by bootstrap on Wed Jul 28, 2010 4:45 pm

mikeselectricstuff wrote:
#7: Apply vacuum to the component pick-up tool (squeeze the big red rubber bulb, hahaha),

You need continuous vacuum to deal with small leaks - a bulb type thing is useless for pick/place, like those cheap 'manual' pick/place pens. Fortunately aquarium type pumps can be adapted to provide a cheap vacuum pump.

I think the whole alignment process is overcomplicated - why not align the package based on the image of the body from the top side, i.e. you have a fixed overhead camera, produce a composite image of the stored image of the PCB pads, overlaid on a live image of the top of the package, Add some software smarts to detect and display the package and PCB footprint centroids, and maybe produce an audio feedback representing the deviation between the 2 images so you can align without looking at the screen?

I also suspect the much-to-cheap "red bulb" would not work, unless the seal between the rubber tip/cup was so good that the vacuum did not lead out. The only downside of a cheapo pump is the extra wires and power required, but I suspect that's necessary to assure reliability.

My original thought was to align based upon the top outline of the package. Unfortunately, some of the BGA parts I have are not properly aligned with the solder-balls. For this reason I kept thinking until I came up with a way to assure alignment between the solder-balls and the BGA pads on the PCB. Probably 90% or 95% of components would work the way you say, but unfortunately not all (including some that I have on my PCB - 74AUC16244s and 74AUC32374s).
bootstrap
 
Posts: 72
Joined: Wed Jun 02, 2010 8:47 pm
Location: phobos

Re: super-cheap pick-and-place device with ~1 mil accuracy

by mikeselectricstuff on Wed Jul 28, 2010 7:22 pm

Are they actually inconsistently aligned between ball and package, or just offset?

Another possible issue is the cost of a camera lens that will give a close enough image while leaving enough workin distance between the lens and the subject, while having sufficiently low distortion (although clever software could compensate for the latter).

I wonder if you could maybe so something clever with a mirror..?
mikeselectricstuff
 
Posts: 164
Joined: Fri Jun 11, 2010 9:21 pm

Re: super-cheap pick-and-place device with ~1 mil accuracy

by bootstrap on Thu Jul 29, 2010 12:05 am

mikeselectricstuff wrote:Are they actually inconsistently aligned between ball and package, or just offset?

Another possible issue is the cost of a camera lens that will give a close enough image while leaving enough workin distance between the lens and the subject, while having sufficiently low distortion (although clever software could compensate for the latter).

I wonder if you could maybe so something clever with a mirror..?

Yes, the alignment errors are inconsistent, which sucks, doesn't it?

I don't think distortion will matter, because the distance to the templates and the distance to the PCB will be the same. Thus the distortion will be identical for both targets, which therefore does not introduce error into their alignment.

I never could think up any scheme with mirrors that wasn't terribly complicated and error prone.

I have an alternate, simpler, cheaper approach that doesn't require the hole array or rectangle "templates", but it is a bit gross and kludgy, and doesn't work as well with QFNs and iLCCs.
bootstrap
 
Posts: 72
Joined: Wed Jun 02, 2010 8:47 pm
Location: phobos

Re: super-cheap pick-and-place device with ~1 mil accuracy

by mikeselectricstuff on Thu Jul 29, 2010 4:42 am

webcams are cheap - maybe just have a second upward-facing one to measure the relative alignment of the package alignment to the pads, then use this to correct the placement on the overhead one.
mikeselectricstuff
 
Posts: 164
Joined: Fri Jun 11, 2010 9:21 pm

Re: super-cheap pick-and-place device with ~1 mil accuracy

by bootstrap on Fri Jul 30, 2010 12:00 am

mikeselectricstuff wrote:webcams are cheap - maybe just have a second upward-facing one to measure the relative alignment of the package alignment to the pads, then use this to correct the placement on the overhead one.

I'm not sure I understand the purpose. Are you trying to avoid the hole-array and rectangle templates? Or what? The scheme I described only has 1 webcam and is extremely accurate, so I must be missing something about your intention or line of reasoning. Please explain more completely.
bootstrap
 
Posts: 72
Joined: Wed Jun 02, 2010 8:47 pm
Location: phobos

Re: super-cheap pick-and-place device with ~1 mil accuracy

by mikeselectricstuff on Fri Jul 30, 2010 2:33 am

bootstrap wrote:
mikeselectricstuff wrote:webcams are cheap - maybe just have a second upward-facing one to measure the relative alignment of the package alignment to the pads, then use this to correct the placement on the overhead one.

I'm not sure I understand the purpose. Are you trying to avoid the hole-array and rectangle templates? Or what? The scheme I described only has 1 webcam and is extremely accurate, so I must be missing something about your intention or line of reasoning. Please explain more completely.

Yes - to avoid the need for a template for every different footprint, and any need for accurate mechanics, e.g. to avoid rotation & drift issues between the template and the placement position.The templates will only work on BGAS, not with parts like QFNs which have no bumps.
mikeselectricstuff
 
Posts: 164
Joined: Fri Jun 11, 2010 9:21 pm

Re: super-cheap pick-and-place device with ~1 mil accuracy

by bootstrap on Fri Jul 30, 2010 3:52 am

mikeselectricstuff wrote:
bootstrap wrote:
mikeselectricstuff wrote:webcams are cheap - maybe just have a second upward-facing one to measure the relative alignment of the package alignment to the pads, then use this to correct the placement on the overhead one.

I'm not sure I understand the purpose. Are you trying to avoid the hole-array and rectangle templates? Or what? The scheme I described only has 1 webcam and is extremely accurate, so I must be missing something about your intention or line of reasoning. Please explain more completely.

Yes - to avoid the need for a template for every different footprint, and any need for accurate mechanics, e.g. to avoid rotation & drift issues between the template and the placement position.The templates will only work on BGAS, not with parts like QFNs which have no bumps.

The scheme I mentioned does not require a template for every different footprint. It only requires one template (area) for every pitch. For example, every BGA component with 1mm pitch is handled by a single "hole-array" with ~32 x 32 (or maybe 40 x 40) holes with 1mm pitch. No matter how many solderballs a given 1mm pitch BGA component has, we place it in the upper-left corner of the 1mm "hole-array". Yes, this still requires about 8 templates for BGAs, but not the hundreds or thousands of templates that would be required to support every BGA pattern.

What "rotation and drift issues" between template and placement position? Unless I'm missing something, the only precision that is needed is repeatability of the endpoints in the X-axis stage. When you move the X-axis stage all the way to the right, then to the left, then back to the right - that stage must end up exactly where it was the first time --- relative to the pick-and-place device (but NOT to the templates or PCB). In fact, that's the main beauty and attraction of this scheme (unless I'm missing/mistaking something important)... that nothing needs to be "measured" (by encoders). All that matters is, as long as the "full-left" position is always 50mm (actually, ANY distance) from the "full-right" position - the device places the component exactly. Yes, the component must not rotate on the vacuum pick-up tool, but why would that happen? Even pick-and-place machines with fancy encoders and upward-looking cameras move the component (real fast) between the upward-looking camera and final placement position. If THAT doesn't cause "rotate or drift" problems, why would anything about the configuration of this device?

However, I don't think the above is the kind of "rotation or drift" you mean. To attempt to expose the source of the misunderstanding (either you or me), let me make a statement that seems perfectly correct to me, but may seem totally insane to you. If this is what happens, we need to focus on this example more carefully to figure out who is making a mental mistake.

I claim the following. In step #9 we take a still image of the template where the BGA component was located a few seconds before. At this point, we know the BGA component is exactly 50mm to the right (along the X-axis) from the center of the still image we just took. We know this because that's exactly how far the X-axis stage moves when we move it to its "extreme left" or "extreme right" position. The image has also exactly captured the rotation of the BGA component relative to the X and Y axes, which are aligned with the rows and columns of pixels in the image.

Now, I claim the PCB can be removed from the teflon sheet, taken outside into your back yard and thrown around like a frisbee for a few minutes, then brought back inside and securely taped down to the teflon sheet again in an entirely different location and rotation compared to its original position and rotation on the teflon sheet. Now we perform the rest of the steps (step #10 to the end).

My claim is, the BGA component will be placed perfectly. My guess is, you think the enormous "drift and rotation" errors caused by grossly moving and rotating the PCB will screw up the process. Does this sum up our assumptions?

Let me try to explain why I take my [seemingly insane] position. When we move and rotate the PCB around in step #10 to make the still image of the holes in the "hole-array" lie on top of the video image of the BGA pads on the PCB, we must rotate the PCB (presumably by rotating the teflon sheet) until the X-axis of the "hole-array" is exactly aligned with the X-axis of the BGA pads on the PCB... otherwise their images would not align and correspond. The webcam has not rotated relative to the X-axis stage, so the PCB is now rotationally aligned in the exact same way (at the exact same angle) as the "hole-array" was previously... otherwise the two images would not align. Right?

Furthermore once the still and video images are aligned, we know the BGA component is exactly 50mm to the right (along the X-axis) of the webcam, because that's simply the way the device is assembled. Thus, when we move that stage 50mm back to its "extreme left" position, the BGA component is precisely located over the BGA pads on the PCB (that the video image was just displaying).

I know that's a long-winded description, but I hope it captures the point I'm trying to make. To me, the logic of this description seems solid and bullet proof (that is, lacking any logical/rational errors). Frankly, I was surprised when I first came to this conclusion, and suspicious of my conclusion until I had tried to find a fault in the logic. On a simple intuitive basis it doesn't seem possible to achieve something like this with no encoders, and in fact, no known distances at all (the X-axis stage can move any distance... 50mm is just convenient, but 48.53243mm is just as good), no alignment of the PCB with the X-axis stage, and only 1 webcam. What I think makes this work is that the steps of the described process make "certain repeatability of extreme-left and extreme-right positions" substitute for linear encoders. Seems like all we needed was a 2 position linear encoder to make this work. Well, that plus the pixels of the webcam being on the order of 0.0005" to 0.0010" on the PCB. So in a manner of speaking, the webcam plays the role of our 2-axis optical encoder - when combined with the described process.

Or so I still believe... until you set me straight? :shock:

You also seem to think this device doesn't work on QFNs and iLCCs (which are also QFNs, essentially). I'm not sure why you think this, unless you didn't understand what those square and rectangular cutout "templates" were for. These are the templates for QFNs any similar package.

The nice thing about QFNs is - your technique works (assuming the outline of the physical package is aligned with the component pads/contacts). We can assume this because the component pads/contacts run to the edge of the physical package. This eliminates the possibility of offset or rotation between the physical package and the pads/contacts.

Therefore, to place a QFN component, you simply drop it into the appropriate size square/rectangular hole/cutout "template", then perform all the steps described for the BGA. The only difference is, when you align the still and video images on your LCD monitor, you are aligning the still image of the square/rectangle template with the outer edges of the QFN component pads/contacts on the PCB. The principle is exactly the same. No need for "bumps", we only need the "outline".

If you're concerned about potentially needing a large number of square/rectangular cutout templates, note that technically and in fact we only need one large square cutout large enough to fit the largest QFN package into. We simply shove the QFN into the upper-left corner of this large square cutout "template", then align the still image of that upper-left corner of the "template" with the ends of the QFN pads/contacts its top-side and left-side in the video image. However, I'm pretty sure that we can machine (?or punch?) all existing QFN sizes in a reasonably modest size area of the SMT template plate. Again, I envision only one template plate to support all BGA and QFN components, plus a few others (like 0603, 0402, 0201, etc).

Okay, did I convince you? If not, tell me where I went wrong.
bootstrap
 
Posts: 72
Joined: Wed Jun 02, 2010 8:47 pm
Location: phobos

Re: super-cheap pick-and-place device with ~1 mil accuracy

by bootstrap on Fri Jul 30, 2010 7:52 am

I'm looking for parts to build a couple of the super-cheap pick-and-place devices that I described previously in this thread. The linear bearings and shafts are easy enough to find, but where to find a few of the parts isn't obvious to me. Anyone know where I can find:

#1: vacuum pick-up head : It would be great if the head included a switch that toggles the vacuum on and off each time the tip presses down on something. This way the vacuum would turn on when the tip pressed down against the top of the component to pick it up, then turn off when the tip pressed the component down onto the PCB.

#2: linear track end blocks : I thought I'd find these, but so far no luck. These are simply thick metal plates (or cast equivalents) that have two 15mm~25mm round holes 50mm ~ 100mm apart to plug round linear rails into. Obviously I can get this machined up pretty easily, but better if I can buy off the shelf.

#3: PCB stand : This is just a sturdy adjustable frame that can hold any size PCB ~25mm above the tabletop. I'll attach teflon feet so it can easily slide around on the tabletop.

I guess that's it. Most of the rest is easy to find (loose linear bearings, linear bearings in pillow blocks, round shaft for linear rails, etc)... or something that clearly needs to be machined custom (the templates).
bootstrap
 
Posts: 72
Joined: Wed Jun 02, 2010 8:47 pm
Location: phobos

Re: super-cheap pick-and-place device with ~1 mil accuracy

by blogger on Sat Jul 31, 2010 3:08 am

> #1: vacuum pick-up head :

you don't want the head to ever actually touch the chip in the tray and especially you don want it pushing it down on the paste in any shape or form.

you want it within < 1mm to the chip and use vacuum to suck the chip up, and then when releasing, reverse air to pressure and "puff" the chip down. definitely not pushing.
blogger
 
Posts: 43
Joined: Tue Nov 24, 2009 5:59 am

Re: super-cheap pick-and-place device with ~1 mil accuracy

by bootstrap on Sat Jul 31, 2010 7:43 pm

blogger wrote:> #1: vacuum pick-up head :

you don't want the head to ever actually touch the chip in the tray and especially you don't want it pushing it down on the paste in any shape or form.

you want it within < 1mm to the chip and use vacuum to suck the chip up, and then when releasing, reverse air to pressure and "puff" the chip down. definitely not pushing.


Interesting. Why is it bad for the pick-up cup/nozzle to [barely] "lightly touch" the component? Certainly we don't want it moving or rotating the component when it picks it up or sets it down. Is that what happens when a vacuum pick-up tool touches a component?

I would be especially surprised if this was a problem during the "pick-up" process, because in this device the components are sitting in "detents". The BGA solder-balls are sitting in holes about half the diameter of the solder-balls, for example. So I don't see how a slight downward pressure would move it in X or Y, or rotate it either. If it was sitting on a flat surface, then I could believe this is a problem.

During the "place" process the situation is somewhat less clear to me, but I still don't see why the component would tend to move or rotate, unless the downward force was significant. Nonetheless, if you're correct about this, I see two ways to address it.

Perhaps the simplest is to make the downward limit of the "Z-axis stage" be "just above the PCB". This would be easy if all components were the same thickness, but that seems unlikely (unless they only need to be within ~1mm of the same thickness before the "puff" ejects them).

The alternative is to put some kind of PCB distance probe off to the side of the vacuum nozzle that "puffs/ejects" the component, and generates an audible "beep" to indicate placement is complete, and the head should be raised again. I suppose the best scheme would be to generate an increasing pitch/volume tone the closer the component comes to the PCB to warn the operator "slow down". This feedback would be very easy to for the operator to learn, and become efficient with.

Can you explain the negative consequences of applying slight force to the component (against the paste) during placement? Also, can you explain how others deal with this problem, especially sensing the proximity of the PCB to the bottom of the component (which is very irregular, after all)? If I am correct about this device tolerating physical contact during the pick-up process (due to the templates that hold the component securely in position), perhaps the device can detect the specific thickness of the component during the pick-up, and factor that into the proximity tone generation.

Any further ideas will be appreciated.

PS: I was just playing around with my vacuum pick-up tool, and that drove home your point about the need for the "puff" of positive air pressure to place the component. Many of the components are so incredibly light-weight that molecular adhesion or slight static charge or something often makes the component not drop off the nozzle tip when the vacuum is released. Yikes! Point driven home in spades.
bootstrap
 
Posts: 72
Joined: Wed Jun 02, 2010 8:47 pm
Location: phobos

Re: super-cheap pick-and-place device with ~1 mil accuracy

by mikeselectricstuff on Sun Aug 01, 2010 5:21 am

I think the vac head does want to just touch the component, and also apply a small downward force when placing - you certainly don't want the part to have any appreciable 'free-fall', or a 'suck up' distance as this will introduce placing errors - due to stiction etc not all parts of the sucker will detatch simultaneously, so it will tend to fall at an angle instead of remaining horizontal.

I thing it should push the part into the solderpaste, which will then provide the adhesion needed to pull the part off the nozzle when the vacuum is released, avoiding the need for a blow-off.

You could put a force sensor on the head to indicate when it is palced - this could be something as simple as a spring-loaded head and a mechanical indicator showing how much the spring has moved. You probably want some vertical compliance in the head anyway to deal with things like PCB warp and mechanical height tolerance across the work area.
mikeselectricstuff
 
Posts: 164
Joined: Fri Jun 11, 2010 9:21 pm

Re: super-cheap pick-and-place device with ~1 mil accuracy

by bootstrap on Sun Aug 01, 2010 8:34 pm

mikeselectricstuff wrote:I think the vac head does want to just touch the component, and also apply a small downward force when placing - you certainly don't want the part to have any appreciable 'free-fall', or a 'suck up' distance as this will introduce placing errors - due to stiction etc not all parts of the sucker will detatch simultaneously, so it will tend to fall at an angle instead of remaining horizontal.

I think it should push the part into the solderpaste, which will then provide the adhesion needed to pull the part off the nozzle when the vacuum is released, avoiding the need for a blow-off.

You could put a force sensor on the head to indicate when it is palced - this could be something as simple as a spring-loaded head and a mechanical indicator showing how much the spring has moved. You probably want some vertical compliance in the head anyway to deal with things like PCB warp and mechanical height tolerance across the work area.


Yes, I too believe the components should always be touching something else, not flying upwards or downward through the air. But I'm ready to listen to the opposing opinions further too.

Because the components are restrained from moving in X, Y or R by the "templates", I'm not too worried about the force applied to the component during pick-up, though obviously that force should still be as tiny as possible. In fact, I would rather have the vacuum turn ON after the vacuum pick-up tip [barely] touches the component. This should assure the vacuum does not make the component fly upwards through the open air, where X,Y,R motions might occur.

I also like the idea of letting the viscosity of the solderpaste help us transfer the component from vacuum tip to PCB during placement. We don't want to smash down solderpaste with too much force, because that might squeeze the solderpaste outside the periphery of the soldermask opening, onto the soldermask itself, and possibly even up against adjacent pads/contacts/solderballs.

How we prevent pushing down against the solderpaste with too much force may be one of the more difficult problems. What would be best, if we can figure out how to implement this simply, would be to have the vacuum tip somehow "float" up and down around (or inside) another solid tube (fixed to the stage). If the vacuum tip can be pushed upwards ~10mm with only a very tiny force, then the downward Z-axis motion could overshoot by up to 10mm without smashing the solderpaste. Maybe that's what the spring was for in your comment... to apply that very light pressure.

I think we're on the same page about this part of the device. I'm just not very sure what is the best way to implement this part of the device yet. Maybe all that's necessary is a couple miniature linear ball bearings and a wimpy spring. Not sure.
bootstrap
 
Posts: 72
Joined: Wed Jun 02, 2010 8:47 pm
Location: phobos

Re: super-cheap pick-and-place device with ~1 mil accuracy

by cphoenix on Tue Sep 28, 2010 3:29 am

I don't understand why you need these sliders at all. Seems to me that the key parts of the system are:

1) The webcam, which can look down either at the template or at the PCB.

2) The pickup device, which must sometimes (but not always) be in a fixed repeatable relative to the webcam.

If you simply built a mount for the webcam, with a jig in the mount for the pickup device to fit into, wouldn't that work just fine? All you'd need to be able to do is remove and replace the pickup device without moving the webcam.

So the operation would go like this:

1) Bolt the webcam mount to the table.

2) Place the pickup device in its jig on the webcam mount. (Slide it into a square corner, or whatever. The pickup device can rest on the table and just align itself relative to the webcam, or can be clipped or set into the webcam mount somehow. It doesn't have to be fastened, just positioned; in fact it should be easy to remove.)

3) Position the template, with the device resting in it, on the table under the pickup device (so you can pick up the part without moving the pickup device sideways).

4) Pick up the part.

5) Remove the pickup device, with the part still attached. (This corresponds to the "Slide 50 mm to the right" in the original design.)

6) Take a picture of the template.

7) Put the board on the table under the webcam, and align the picture of the template with the video of the pads.

8 ) Put the pickup device back in its jig. (Corresponds to "Slide 50 mm to the left.")

9) Place the part.

No sliders, no precision machining.

I think I'll build one of these! It's a neat idea.
cphoenix
 
Posts: 2
Joined: Tue Sep 28, 2010 3:15 am