Local fiducials under package, it's ok?

[colotron]

Hello.
I was seeing a pcb design with a big module, and this board have 2 local fiducials, one of them below the module itself, and the other outside the module.
Is that a good practice?.
Could be useful to save board space and have more freedom to route the board.
What do you think about it?
Cheers.

[Plecharts]

I don't think the fiducials make that big difference in laying out a PCB, so there is a different reason why they did it probably. Was the module right on the PCB, or was it spaced off using some headers and it had some ICs under it. If there was a BGA or something with small pin pitch, it could be to get better pick and place precision.
Please post some photos so we can see it.

[colotron]

Of course, a view of the board is attached with some layers disabled to see better.
The board eagle source files are here.
In the attached image, the fiducials are circled in yellow.

Could be some difference - not too big I agree - in laying out small boards when the space is very constrained.

Thanks for your answer.

[Jon Chandler]

Since the fiducials are used to align the board and pick&place machine, locating it under a chip should make no difference.

[colotron]

Thanks for you answer Jon.
So far no dissent, it seems to be an acceptable practice.
Cheers.

[EEVblog]

You can see an example of this in my last Lecroy scope video when I sucked the chip off. Fiducial in the exact centre of the chip. Rare to see that.
The usual deal is to put it just outside the chip, 2 on opposite corners, so the vision system can get more accurate local placement for a fine pitch device like a BGA.
or at least that's the theory. I have no idea if assembly houses actually even use local fiducials any more  As the machines might be more than accurate enough with the usual panel edge fiducials.
Are local fiducials a relic of days gone by I wonder?

Dave.

[poorchava]

No. In my company we have our own assembly lines and we sure as hell use fiducials, on the other hand they are placed on the outskirts of pcb even is there's BGA or fine-pitch QFP/SOP, so modern systems can do without fiducials close to particularily hard to align component.

[colotron]

Link to Dave's video, 9m30s:


Yup, lets keep using them. But it seems the rule "place fiducials on opposite corners of the chip" is not grabbed in stone.

[free_electron]

A center-fiducial is used for precision placement of a part if the camera can not see all the pads of the footprint. in order to find the exact center of placement the camera needs to be able to see the 4 corner pads so it can run two diagonal seeking algorithms and determine the center location. if a part is really big ( like those lecroy parts ) the field fo view of the camera is not large enough to capture this in one shot. so that's when a dead-center fiducial comes into play.

[EEVblog]

A center-fiducial is used for precision placement of a part if the camera can not see all the pads of the footprint. in order to find the exact center of placement the camera needs to be able to see the 4 corner pads so it can run two diagonal seeking algorithms and determine the center location. if a part is really big ( like those lecroy parts ) the field fo view of the camera is not large enough to capture this in one shot. so that's when a dead-center fiducial comes into play.

But what if the chip center placement is already programmed in as part of the pick'n'place file?

Dave.

[JuKu]

But what if the chip center placement is already programmed in as part of the pick'n'place file?

It should be. If it is not, all three factories I've been working with lately would reject the data.

The accuracy of the machines is some percentage of travel, plus fixed error. Nothing can be done to the fixed error, but the percentage error is much smaller for 10mm travel than for 500mm. I don't have the specs here*, but they are like a fiducial every 200mm plus for yy mm parts a fiducial no further than xx mm. That's why you sometimes want local fidus. But in general, the placement machine accuracy is not the limiting factor, the soldering process is more difficult to get right.

*: Nor have I looked them for a long while; I'll just put a few fiducials on the board corners (and on middle on large boards) plus some around really tight parts.  If it feels right, you are well within specs.

[EEVblog]

The accuracy of the machines is some percentage of travel, plus fixed error. Nothing can be done to the fixed error, but the percentage error is much smaller for 10mm travel than for 500mm. I don't have the specs here*, but they are like a fiducial every 200mm plus for yy mm parts a fiducial no further than xx mm. That's why you sometimes want local fidus. But in general, the placement machine accuracy is not the limiting factor, the soldering process is more difficult to get right.

Given that chips are usually essentially "self-centering" on the pads during reflow, I'd be surprised if local fiducials (and hence increased fine positional accuracy) are really required for say an ordinary A4 size panel.

Dave.

[free_electron]

A center-fiducial is used for precision placement of a part if the camera can not see all the pads of the footprint. in order to find the exact center of placement the camera needs to be able to see the 4 corner pads so it can run two diagonal seeking algorithms and determine the center location. if a part is really big ( like those lecroy parts ) the field fo view of the camera is not large enough to capture this in one shot. so that's when a dead-center fiducial comes into play.

But what if the chip center placement is already programmed in as part of the pick'n'place file?

Dave.

Ah, looks like somebody needs a bit of an explanation 

The machine indeed goes by the coordinates given in the pick and place files. Most commonly it goes by the center coordinates. What if the board has an offset compared to the 0:0 point of the machine ? Then all programmed center coordinates will be 'off' by that offset. So that is why the pick and place head has another, downward looking camera. Instead of traveling to the pick and place coordinates given with the part , it travels there with the camera. The camera has a fixed offset from the nozzle as it 'rides' on the nozzle carriage. The camera now finds the center fiducial and compares the coordinates given by the positioning electronics of the stage ( essentially the feedback of the encoders on the stage ) +/- the coordinates inside the camera viewport to the numbers given in the file. Any residual offset is now compensated.

Boards, especially large boards do tend to 'shift' in pick and place machines under the vibration.
As for components 'settling' during reflow : that is only true if the mass of the object is less than the upward force given by the combined surface tensions of the liquid solder and only during the short point in time the solder is liquid but has not yet 'wetted' the part.
The moment 'wetting' kicks in then we deal with sideways forces , again dominated by the surface tension , but this time the pin is not 'on top' of the solder bubble, but inside the bubble. Think of it this way. The liquid solder is a little droplet of water. When the pin has not wetted yet it needs to pierce the surface of this droplet and try to get inside the droplet. Each pin has its own droplet so it is the combined force. Once wetting occurs the pins are inside the droplets. So now the surface tension keeps them nicely centered inside the droplet. Try to pull them out ( you are trying to escape from the inside, so you try to pierce the surface tension from the inside out) and the surface tension will push you back to the center.

Have a very heavy part, like a ceramic hybrid and you don't stand a chance moving it ... The combined surface tension does not have enough force to do it.
Have a very light part and you get the opposite effect : drawbridge where one pad did not flow and the part sits slanted. Or plain tombstoning where there is too much solder and the forces actually succeed i pulling the part vertically with one leg In the air.

Keep in mind that fabrication machinery has improved as well and at the time that lecroy board was made machines were less precise. Today I doubt the machines would require the fiducials.

I was at mvinix ( a subcontractor I use to assemble boards) a couple of months ago and they just installed a brand new Panasonic placement machine that handles 01005 packages ... 0805, 0603, 0402, 0201, 01005 <-this is where we are ... That stuff is literally 'dust'. If you throw a couple of these in a ziploc bag you can't find them with the naked eye ( at least I can't ).

That machine is fully self aligning and uses a very clever technique. It has one stud on the side that has a tiny crosshairs laser engraved on it. The machine first goes there with the downward camera and 'homes' itself to that point. Then it goes with the nozzle over the upward looking camera and finds the exact offset of the centre off the nozzle. It then goes and picks up the smallest part it has on its feeders , moves this to the upward camera, spins it around and centers this part I. The field of view of the upward camera. It now goes to the crosshairs and place the part there. Using the downward looking camera it now figures out the placement error of the part on the crosshairs ( the crosshairs is like a graticule with division marks so the machine can read both exact center ( which it set as 0:0 ) and the subdivisions. The machine now knows the offset between upward looking and downward looking camera as well as the nozzle position
So it can now compensate for any mechanical intolerance in the entire stage. Simply brilliant.
If maintenance was done and the cameras were removed for lens cleaning you have slight errors. No problem. During machine reset the machine fully nulls itself using a sacrificial part.

Now, this machine actually has two downward looking cameras that sit under a 45 degree angle in respect to the board. So it can actually look underneath the part... For the 01005 and 0201 parts it simply uses the pads of the component footprint as fiducials! As the nozzle with part is over the placement position the camera seek both pads on the board, apply all offsets ( stage, camera, nozzle , part ) and then thinks the part down. Before departing the placement site it uses the cameras again to verify correct placement. If ,for some reason ( vibration , part cracked , not enough solder paste on one pad ) the part has shifted , the machine will actually attempt to pick it up and reposition it. If it fails after two attempts it puts in an 'operator' notice that part xyz has a placement error and needs to be corrected by hand.

Incredibly clever bugger that machine. Costs 1.5 million$...

[free_electron]

Hmm . After re-reading the above post I realized I could have been clearer...

The part needs to be heavy enough ( mass of the part under influence of gravity which pulls it down) so it will succeed in piercing the surface tension of the solder droplets. If the ratio of part weight to 'outward' surface tension is close to 1 or smaller than one ( part is 'lighter than combined surface force ) the part will not 'drop' . It will float on the solder and you get draw bridging or tomb stoning. For very light parts like flipchip , studded or bumped packages and even certain qfn packages this is a problem. They are so light that the amount of solder on the pads need tight control or that ratio becomes close to 1 and you get soldering problems. These packages have no metal in them and very little epoxy.

Now, once the surface has been pierced and the pins/pads have wetted it is again the surface tension that kicks in, but it is now the 'inward' surface tension. The pin sits inside the bubble and is subject to a force coming from the surface all around it , pushing inwards. So the surface tension off the solder tries to center the pin inside. So this is where the part 'self-centers'. Have a mass to tension ratio that is too high here and no shift will occur. There simply is not enough force present to move the part.

And I that case you want placement as close as possible... Hence the center fiducial. Note that center fiducials cannot compensate for angular error , only for x-y error , so during placement of these beast the stage first puts the part over the upward camera to find rotation and x-y offset of part in respect to nozzle. Then the stage goes and finds the fiducials of the board to find x-y and theta( restoration) of board in machine, then it goes and finds center fiducial of the part to compensate again for x-y errors during staging , and uses theta of the board combined with theta of part on nozzle to apply rotational correction. Problem solved. The part is now where it needs to be.

These techniques are used for heavy parts where placement will not 'self-adjust' or in cases where the downward looking camera cannot see all the pads to calculate dead center.
It all depends on what kind of dinosaurs you are working with (big bulky dinosaur parts, and big bulky placement machine dinosaurs).  Have state of the art machinery and the answer is 'meh'... Machine is smart enough to figure it out.

Even older systems required two fiducials on opposite corners of a chip. One had to be larger than the other. The largest fiducial denoted pin 1 position. The smaller one was diagonally opposite allowing for center finding ( midpoint of the two ) the fiducials need to be placed at the intersection if the midlines of pin rows. You draw a line through the centers off all pads of one side. Where the lines of two adjacent sides cross : that is where the fiducial needs to be.

The upward looking camera finds pin 1 the same way. All tqfp , P&P and other square or rectangular packages with 4 sides of pins have a fiducial on the bottom of the part. There is a recessed circle there denoting pin 1. The upward camera finds that one , know how the part 'sits' o. The nozzle ( where pin 1 is ). Goes and find the larger of the two part fiducials and now knows how to rotate.  Just in case one part sat rotated in the tray...

Those markings on the bottom of parts are not there 'for fun'. But to help the pick and place machines where they need to be.