## Pick&Place machine run-time closed-loop actions

14 posts / 0 new
Author
Message

I have 0% experience in P&P machines and PCB manufacturing so sorry if that is a basic question.

Introduction:
I am curious if such a machine can be driven in a closed-loop fashion, that is if the operations it executes can process a more sophisticated data at its inputs at run-time. I also understand it is actually a computer that controls it - I just want to know how the method I presented is named in the industry..

I do not know the terminology of P&P either, so correct my vocabulary if I use funny names :)

Basic example:
I understand that whenever P&P sucks an element from the reel and it happens the element is missing (it senses "nothing sucked"), then it shifts this reel to the next element and picks that one. That is a simple example of interaction, when the action is made in the closed-loop.

Lets suppose I need to build a ~0,1% resistor voltage divider in the circuit (current flow is not that important).
One way to achieve such accuracy is to buy a highly precise resistors (rather expensive). I buy two reels of these and job is done.

As an alternative I could mount R>2 reels of less precise resistors selected so that at any possible combination of actual resistor values their respective resistance relation gives a required 0,1% accuracy of voltage divider. The problem is I need to collect the data to make it in a closed loop so lets suppose the head of the machine attaches a precision ohm-meter to a (N+1)th resistor whenever it picks Nth one from the reel. That way I have all the information about coming resistors on all R reels, the only thing I need is to make a decision on-the-fly which resistor pair to pick and place.

It does not seem to be a complicated decision but many factors can be involved. Obviously the higher R is, the less precise resistors I can buy, still getting the same effect of 0,1% divider. The whole synthesis can/should be covered using a probabilistic theory so that for example in 1 case per 1000 a resistor will have to be thrown away because it does not match the proportions. Digging deeper, a mean and variance values of resistors form Rth reel can be estimated to further improve the gain etc.

Does anybody know if such method can be used to reduce the BOM cost? Does it have a name?

No RSTDISBL, no fun!

Yes, SOT, manual 'select on test' is the only way.
P&P machines can only do what they are told....

Once the resistor is thrown out from the reel - head, its value is almost lost because will be hard to take it back.
Reading its value on the fly will decrease the process speed and I think the contract manufacturer will not be happy to slow down their machines because you want to save a little money.

Try to implement a software calibration on your product.

You may be able to select proper resistors while pick and place them but the temperature coefficient will still be a problem unsolved.

George.

chartman wrote:
Yes, SOT, manual 'select on test' is the only way.

Did search in Google and found nothing. "Select on test" does not go with "Pick and Place".

angelu wrote:
Once the resistor is thrown out from the reel - head, its value is almost lost because will be hard to take it back.

Obviously you have the option to mount for example 20 reels and loose a resistor with a probability of 1 per zillion, or to mount only 5 reels and loose a resistor with a probability of 1 per thousand, still obtaining the same 0,1% effect. You can even mount two reels only and throw away 95% of resistors - that solely depends on what you need and expect (and afford).
It is not that I wanted to point out this method forces the P&P to act like: 998-th mounted ,999-th mounted, now here comes a 1000-th so throw it to the bin, 1001-th mounted etc..
Choice is yours, but considering both the inventory and count of mounted reels is limited, I do not think throwing one of them per n is a problem as long as n is a variable - it depends on many inputs, like the accuracy of divider required, the ohm-meter itself, reel count etc.

angelu wrote:
Reading its value on the fly will decrease the process speed

Actually perhaps I used an imprecise description..
Brutte wrote:
lets suppose the head of the machine attaches a precision ohm-meter to a (N+1)th resistor whenever it picks Nth one from the reel.

It should have been while instead of whenever. I did not mean that you stop or slow down the P&P to measure the resistance! I do not think measuring (N+1)th while picking Nth one is a technically challenging task. Resistance can be measured in microseconds with arbitrary accuracy without a problem - you do not even need a fast ADC - it just needs to latch S/H circuit and the cheapest stuff able to make conversion at 10Hz will do.

angelu wrote:
Try to implement a software calibration on your product.

I can see a simple resistor divider example is not enough for you :)
Lets suppose then it is a 7-th order analog active band-path filter working in 2,7GHz range, containing 20 resistors, 7 capacitors and 3 inductors.
What I mean is the idea is generic and it does not bring any significant advantage to consider such a complicated multi-chip design so that your advice would not be rational. The problem of tolerances in the filter is of the same kind as with a given voltage divider, with the difference the algorithm of making a proper decision with RLC-s complicates significantly.

IMHO resistor divider is generic enough.

angelu wrote:
but the temperature coefficient will still be a problem unsolved.

This coefficient is also a parameter, same as drift with aging and hundreds of other unspecified variables.

No RSTDISBL, no fun!

If the 'customer' can't/won't afford 0.1% resistors, the only other option is for the resistors to be selected during test time - SOT.
Persuading a P&P manufacturer to modify a machine to do this would likely cost \$\$\$\$\$ , that's if they would even consider doing it.I think not.
Then you have to up the costs to all customers to pay for the mod, and most, if any would not bother to take advantage of it.
We make hundreds of thousands of PCBs a year down 4 SMT lines and only a handful get the SOT treatment.The rest are either designed to work or have the appropriate toleranced parts fitted.This is across military,aerospace, medical and commercial customers.

chartman wrote:
Yes, SOT, manual 'select on test' is the only way.

I got confused, but I think I get it!
By "manual select on test" you mean like it is hand made? Like the thing I described, but made manually with DVM?
I thought your suggestion is some kind of an industrial process name, or a registered trade mark of some P&P manufacturer :)

chartman wrote:
P&P machines can only do what they are told....

Again, that sentence is not clear to me. You mean the P&P machines you know cannot do that automatically?

chartman wrote:
Persuading a P&P manufacturer to modify a machine to do this would likely cost \$\$\$\$\$

This is not made for fun, but to reduce BOM. But I am not planning to persuade anybody. I was just curious if people with P&P experience know the name of such solution, so that I could read more about it, instead of reinventing the wheel.
Considering the relation of narrow/wide tolerance elements pricing, I think it could be worth in many cases. Especially in analog electronics. Not mentioning this is an additional (the very early) testing stand in the production chain.

No RSTDISBL, no fun!

Hi, I just registered to this forum..

To the questions: Some pick&Place machines are capable of measuring the component, but usually NOT while they are moving the component. Meaning that the component has to placed on some kind of measuring element inside the P&P-machine for the measurement. (The components are often so small that incorporating a measure element on the nozzle would be too complicated)

I understand the idea behind the original post, but afaik, it's not easily done with current P&P machines. What I mean is that, if your design ends up on some 10 million PCBs/year, everything can be considered. As an easier solution, I would suggest to use multiple 1% resistors in parallel.

Depending on solution you might also want to consider soldering more resistors by hand later. I've seen functional tester, which measured the frequency made by 555 and calculated necessary extra resistors depending on the freq. The tester displayed the required resistors to the operators, which soldered them by hand.

Welcome.

I do not plan to use the method described. I was just wondering if such method has it's name and wanted to google it - that is all.

Quote:
The components are often so small that incorporating a measure element on the nozzle would be too complicated

I do not think it is possible and rational to put anything on the nozzle, as usually many different components and sizes are used in one process. That would be too complicated, heavy and unnecessary IMHO. And it is not the picked element that is supposed to be measured, but a next one. So there is no point to put anything on a nozzle.

Actually I imagined it as a set of a dumb testing clips which are pushed against (N+1)th elements while N-th ones are picked. Only the pushing is made by the head(mechanically) - not the test itself, so that there can be one impedance-meter used to measure all passives in the machine. As P&P knows which clip it is pushing against what kind of element - it seems pretty easy to set measurement range and test prior to the measurement. For example if it is an RC filter which should have time constant of 1s(+-x%), then a R and C impedance need to be tailored so that their multiplicand meets the specification. Not that easy as both are RLC circuits in general.

Your story with NE555 is some intermediate P&P process with the Picking made by a computer but Placing made by a human :)

Quote:
if your design ends up on some 10 million PCBs/year

Ok, it seems my idea is completely unrealistic and does not fit the real world economics..

Let's consider a simplified example then.

I have no idea about the real life P&P tasks, so lets suppose it is some kind of a mass market instrument (domestic energy-meter) and I need a 0,1% resistor divider (0603) dividing the voltage in half (simpler calculations. Even simpler when we assume a uniform probability distribution).

With 0,1% resistors and Vcc I get:

Vout=Vcc*r1/(r1+r2)
All dividers fall into:
Vout=Vcc/2*[0,999 to 1,001]
range. Worst case.

Now, lets simplify further calculations assuming the ohm-meter measures without error. Actually it really does not matter how accurate it is (with respect to absolute value) because we need relative resistances in this case (offset cancels anyway) so it is even easier.

Neglecting higher order factors we get:
allowed_tolerance ~= (N-1)*0,1%
so with ~6 slots we can use 0,5% resistors. And for sure 7 reel slots will do.

Now, www.digikey.com, in stock, quantity = 100k. Sort by price, ascending, never mind the nominal value. Ok, 10k are quite popular.

We pick 0,1% resistor:
http://search.digikey.com/script...
It is 25ppm one.
0,035\$ at 25k is the cheapest in stock.

Comes out 3500\$.

Now we pick 0,5%.
http://search.digikey.com/script...
Same 25ppm thermal stability.
0,009\$ at 50k

Comes out 900\$.

The difference is 2600\$, the final effect is exactly the same. And same count of resistors is placed.

We can pick even cheaper 50ppm resistors and use more reel slots. Or if a probability distribution is +- known(or at least knowable at run-time), then using less than N, for example 5 reel slots and throwing away small percent of 0,5% resistors, the same effect could be achieved. I do not have the data to verify that.

I agree my example is rather unrealistic, as a 0,1% components are not frequently used. It is just the idea.

Your 10e6 PCBs a year would give 520000\$ of savings assuming one such 0,1% voltage divider is placed on each PCB.

No RSTDISBL, no fun!

Simplify here, simplify there, you simplify to the point it can not work in practice :)

Components in one single reel probably are all very close together. In your example, you could use a resistor array, I've heard the individual resistors are all very close together in value, while the absolute value might be 5% off.

PP is all about speed, what if it cannot find a resistor within the tolerance range and has to reject almost the whole reel, stalling the production process in an indeterministic way. This uncertainty is likely far more costly then just buying components with the proper tolerance.

In a mass-produced product you likely would not use these expensive resistors, software calibration is likely cheaper, or some other means to remove the dependency on precision components.

As I said before, a board gets designed to meet a market and a price.If one has to go down the route you suggest, then the design is wrong/not optimal.If a product demands a certain spec, it has to be met one way or another. If that means the only solution is 0.1% 5ppm resistors, then that is what is specified.
If the product can't stand that price, then it's a non runner.

jayjay1974 wrote:
Components in one single reel probably are all very close together.

That is the story I heard some time ago from another hardware guy. Can you post some reference? He could not.

Just to make sure, you are not talking about a pair of matched transistors which are sold in pair and do match as stated in the datasheet? That is obvious as this property in the pool (of pair) is given explicitly.

You suggest that even though the datasheet does not contain information about some property of a product pool (such information could be: "resistors are +-5% wrt nominal value but within a single reel they are +-4,9% wrt nominal value within this +-5% range") but you can still infer such property without measurements? I would like to know a little bit more about the method..
I hope it is not called "it did't not work with this reel".

Anyway, with 1/2 divider, you suggest this match can be made (at some extent) without measurements, based on the properties of the previous reels (which you still do not measure)..

I understand you obviously do not suggest: "measure the data and match resistors" instead of "measure the data and match resistors" as both methodologies are exactly the same (number of measurements can vary). The difference is first method gets only probabilistic properties of the previous pool (and though you will get a probabilistic 1/2 division if the probability distributions were dependent) and mine gets the value of the current component(and though you will get a 1/2 division within specified tolerance).

Where will you get the data from then? If you believe reel3 is similar to reel5 then where is this information from without measurements?

jayjay1974 wrote:
a resistor array, I've heard the individual resistors are all very close together in value, while the absolute value might be 5% off.

So you state that values of adjacent resistors in an array are dependent (statistically). I am not surprised, as these are made in the same factory and with the same process. The problem is manufacturers do not give the data about these dependencies. And the post is not about 0,5 divider so if you find it disturbing, pick some other value, can be 3,1415.

jayjay1974 wrote:
PP is all about speed, what if it cannot find a resistor within the tolerance range

How can you imagine in the given example to find the situation of not finding the 0,1% matching pair? It is impossible, same as 2>3. No point to discuss "what if 2>3".
Or perhaps you suggest I made a mistake in the assumptions or calculations? Please point that out.
But in all cases capacitors were mounted instead of resistors and the algorithm crashed, or the resistors were not 0,5% but 20% and the tolerances cannot be met, or (..) Well that is a different category of problems for a different topic.

jayjay1974 wrote:
In a mass-produced product you likely would not use these expensive resistors.

I agree, I suppose instead of giving 0,5% to 0,1% example, I should give 5% to 1% example. The calculations hold, the differences in price are of an order (or two) lower then.
Is it ok for you?

And I wonder who uses those 0,1% elements? They are designed for P&P..

jayjay1974 wrote:
software calibration is likely cheaper

If it was likely cheaper in mass production then Digikey would not sell those tight tolerance components, would they? We would have one type of resistor value on the market then, all the calibration would be made in software..

chartman wrote:
As I said before, a board gets designed to meet a market and a price.

I am sorry, but I do not understand your argument. As I understood, you state that:

"If the design requires a 0,5 +-0,1% voltage divider (or +-1% version),
then you must buy two 0,1% (or 1%) resistors to make that divider.
In case you cannot allow the price of these tight tolerance components - the product cannot be manufactured for such price (must be redesigned or not run)."

The problem is that it is not true.
You can do it, with the price of 0,5% (or 5%) components, without redesign. Using same amount of resistors.

chartman wrote:
If that means the only solution is 0.1% 5ppm resistors, then that is what is specified.

The requirement is not a 0,1% resistor, but a 0,1% resistor divider in here - I hope that is what you meant above.

If that is specified, then you can make divider in 73 different ways. One of them is 0,1% resistor voltage divider, the other one is 0,5% and 6 reel matched voltage divider.

So your sentence above is not true or I didn't get your idea.

It seems the cons could be:

- all designs use so many different components you cannot insert extra reels into typical P&P. - the difference in price between 5% and 1% component is so low that multiplied by 20 million gives only 20.000\$ so it is not worth mounting this ohm-meter clips for the price.
- it is not possible to measure resistance in 0,1s
- etc

No RSTDISBL, no fun!

What do you suppose our charge out rate per hour is on a P&P machine that cost Â£800,000 ??
Add to that any aditional time x hourly rate to implement your idea.We place components at the rate of 10s of thousands per hour to give the customer a low production cost.
If the final product is a cost sensitive volume consumer product, your theory may have legs.
If it's military /aerospace, forget it.Charge out rates for those are up by a factor of 5-10 and the cost of a few 0.1% resistors becomes insignificant.

chartman wrote:
What do you suppose our charge out rate per hour is

What is the difference, when we are discussing BOM cost reduction? The idea does not increase the count of elements, nor are those heavier, nor bigger. The only difference is these are cheaper. The rate of P and P does not change as well.
chartman wrote:

Again, that is not a BOM cost. This is an element of the machine. I suppose the reels can be mounted the same way as all other reels (except that two elements are exhibited instead of one). And I also do not expect the P&P with automated impedance-meter is cheaper than the one without the dongle.

chartman wrote:
We place components at the rate of 10s of thousands per hour to give the customer a low production cost.

Guess what happens when your competitors offer lower BOM alternative with the same quality? I know, you know.

chartman wrote:
If the final product is a cost sensitive volume consumer product, your theory may have legs.

It is not mine, I was just asking for the name of the method.

I was always interested in tear-downs of mass products published along the internet. BOM is about half of the retail price. The passives fall into ~1% of BOM. I have no idea about the tolerances of the components used, but I suppose sometimes you could substitute one with the other of wider(cheaper) tolerance. Not always, not in every product, not in every type of circuitry. This can be 20% instead of 5%, or 5% instead of 1% - does not change a thing in here.

Anyway, the topic is not about passives, but "run-time closed loop actions".

No RSTDISBL, no fun!