Chipsetter ONE: A desktop pick-and-place machine

[rx8pilot]

I've heard numbers like 5 microns or 10 microns.  In my business we own and operate "real" CNC machines and in many of our parts are often machined with tolerances of 0.0001 (2.5 microns).  This is a standard tolerance for a press or interference fit or for a bearing bore, etc.  I know how difficult it is to maintain such a tolerance... it requires a well built and well maintained machine with ball screws and precision encoders and precision drives, and you have to account for thermal growth of the machine.....

I come from a similar place. Mechanical engineer that owned/operated 5 axis CNC shop and designed high precision motion systems. It is very easy to get 'paper performance' of a motion system that does not account for all the real world details. I also have Quads like @Corporate666 has and they would likely struggle with these numbers even though the motion system is vastly more sophisticated and mechanically sound. Before I got my own P&P machine - I was totally naive of the process. It is remarkably unforgiving. What seems like a small problem at first, becomes a huge delay where the machine is doing nothing and 3 people are fiddling with it for an hour. That means the reliability has to be amazing. You will not have useful reliability without amazing feeders, accuracy, repeat-ability, rigidity, and more.

Just as a reference, my Quad is rated for 3600cph in which it's actual placement is significantly lower since I do not perfectly optimized the feeder layout. My best burst speed on hand placing is about 100cph or so after taking all elements into account (I know you can place the same 0805 passive much faster). After all is said and done - the P&P is only about 3-5x faster overall for small runs of boards assuming the programming is all ready to go and most of the feeders are ready. So, the placement speed is 26x faster but the process is at best about 5x faster. That will vary for each user and different design, but it does illustrate the importance of reliability. If anything on the machine is fiddly, or not accurate, or not reliable in some way - it will very quickly eat the time benefit of having a P&P machine. It took me a while to get my system reliable enough to be useful and it is a commercial machine. It only takes a tiny problem to be a big problem and that is the challenge of low-cost desktop machines.

the nature of the P&P process means it is almost exclusively a business tool - not sure why a hobbyist would need 100's or 1000's of PCB's made if they were not for sale. Being a business too, it needs to be highly reliable since you will be making promises to customers based on the expected performance of the P&P system. I am skilled in the art of mechanical design and motion systems, very high accuracy CNC, electrical engineering, and of course the practical every day reality of running an in-house P&P line for small volume. I have not yet seen anything in the desktop class that would make much sense for anything other than the most basic projects. These days, a 'basic' project can use very tiny parts - 0201, 0402, BGA, .4mm QFN, etc. Personally, I would not bother with anything that cannot guarantee success with those parts with very high reliability. It would be lower cost to spend $20k on a machine that gets it done perfect every time than a $4k machine that sort of gets it done. I did not have $20k, so I went with a used eBay machine and slowly added feeders - all in cost $10k although it is a fluke that cannot easily be repeated.

None of this is intended to discourage anyone. The goal is to put the practical reality in perspective. P&P machines do NOT assemble boards. They merely help in the process. Even an amazing machine needs a lot of work and skill to make it practically useful. A fiddly machine needs even more work and skill and the end result may in fact be worse than manual assembly. The entire process requires perfection at every stage.

[mrpackethead]

- all in cost $10k although it is a fluke that cannot easily be repeated.

I was lucky to get my 2 Yamaha machiens for a similar song..   

For business, i just dont' think that i coudl risk spending $5k on a machine that is just going to cause me a lot of fustration. 

[Xenoamor]

USD 4700...

We've got one pretty similar to this. The software is terrible but the machine is well built, accurate and fast

Probably a good niche business to write some decent software for it and promote it an/or sell as a package?

We have soon a software package working on different machines by writing a new machine controller DLL.

I'm not sure what you mean by this but if you have a better software package we'd almost certainly buy it from you

[Corporate666]

P&P machines do NOT assemble boards. They merely help in the process.

That's actually a really amazing quote that very succinctly sums up PnP machines.  I think a lot of hobbyists think they will buy a PnP machine and, suddenly, they will be cranking out prototypes like nobody's business.  But that's sort of like thinking you'll buy a $10k 4k pro level video camera and be a YouTube sensation.  It's just a tool.

Even an amazing machine needs a lot of work and skill to make it practically useful. A fiddly machine needs even more work and skill and the end result may in fact be worse than manual assembly. The entire process requires perfection at every stage.

And this hits the nail on the head.  Any tool that is designed to perform a complex task requires substantial amounts of setup and configuration.  And the less reliable any single part of the complex task is, the more time you spend dicking with the tool rather than having the tool let you go work on something else.  The more I think about it - the more I think the people promoting these hobby level PnP machines really are being outright disingenuous by glossing over any technical details, and most especially by making claims about ease of setup and programming without anything whatsoever to back it up.  And error rate really needs to be an honestly tested and properly disclosed metric.  For an average demo program with varied part sizes, rotations, picking from multiple feeders... what percentage of placements fail for any reason?

With my Quad, it's about 1 in maybe 20,000 placements that I need to do "something" on the machine, with the most common problem being the placement nozzle getting clogged with solder paste, or a feeder having the plastic cover tape get sucked into the pickup area.    But at least I know I can reliably run my machine for a full day and have maybe one issue requiring my attention.  That's a very important number.

[mrpackethead]

hat you mean by this but if you have a better software package we'd almost certainly buy it from you

Sadly lots of talk and not a lot of actual delivery, and you can't figure out who these guys are..    Still waiting for one of his "many" customers to write a review here, but as yet, there has't been one...    From the demo i saw online, the software was pretty much on par with what the open pnp guys are doing..

[mrpackethead]

With my Quad, it's about 1 in maybe 20,000 placements that I need to do "something" on the machine, with the most common problem being the placement nozzle getting clogged with solder paste, or a feeder having the plastic cover tape get sucked into the pickup area.    But at least I know I can reliably run my machine for a full day and have maybe one issue requiring my attention.  That's a very important number.

I hope that one day Desktop PNP actually comes to the party..   I've looked at it long and hard, and i can't in my head get it to work for use in business, where i just don't have the time to be farting around with it.   When the opporutnity came to buy a 2nd hand Yamaha Line, i jumped.

[rx8pilot]

The first step, in my opinion, is to make a machine that is very reliable and easy to setup, program, and change to an all new setup. After that is accomplished, make it as affordable as possible without messing all that up.

It seems all the effort is on making the system cheap as the highest priority. The 2nd priority is getting a high CPH number.

Sent from my horrible mobile....

[Corporate666]

I hope that one day Desktop PNP actually comes to the party..   I've looked at it long and hard, and i can't in my head get it to work for use in business, where i just don't have the time to be farting around with it.   When the opporutnity came to buy a 2nd hand Yamaha Line, i jumped.

A lot of guys have a dream of owning a CNC machine.  At least a CNC milling machine can do things that you literally cannot do with a manual machine - so even if the machine is light duty and slow as shit, you can let it run all afternoon and have something usable. 

People have the same dream with PnP machines, as I did.  But there is the catch-22 that there's nothing a PnP machine can do that you can't do by hand.  So you don't need a PnP machine unless you need to build boards in some quantity.  And if you need to build boards in some quantity, then you can't afford to have a PnP machine that requires constant dicking with to make it work right.

I think a lot of the would-be desktop PnP sellers grossly underestimate the customer support costs.  People buying these machines aren't going to want to fiddle with them themselves and be happy to learn and fix on their own.  It's not like a 3D printer where it's letting you do something you otherwise couldn't do.  It's sole function is to achieve a (generally) business task - manufacturing something faster than you could by hand.  So when the machine is placing every 10th part 0.5mm off or it's dropping parts too frequently or feeders are jamming or the vision system isn't recognizing SOT23-5's or some similar problem, people are going to want a phone number to call or they are going to want a speedy response to their email.  Online peer support forums and email-only support aren't going to cut it.  I think when people look at the cost of pro level PnP machines, one of the main reasons they cost so much is the level of engineering for reliability, and the other part is because of the huge amount of support the manufacturer engages in (tweaking, tuning, replacement parts, etc). 

Anyway, it doesn't look like this one is going to fund.  I don't think the creators 'get it'.  They weren't interested in open software, disclosing hardware for ongoing supply of parts and selling feeders separately until the funding didn't materialize, and then they were all about those things - except they haven't been responsive to inquiries on the place where it matters - Kickstarter.  They keep saying "we'll cover that later", except they are asking for money now.

The winning play for a desktop PnP supplier will be feeders.  Make and sell a generic feeder first.  Then sell a generic hardware system that works with off-the-shelf parts and (of course) their aforementioned feeder and OpenPNP (or their own software if they want to sell it as an option or a fork of OpenPNP).  If it's really better than what's out there, it will sell on its own.  There's no need to make it a closed/proprietary system to capture sales.  Let it sell based on merit.  A newcomer doesn't have the track record/street cred or ongoing stability to start out as a closed system.

[mikeselectricstuff]

People have the same dream with PnP machines, as I did.  But there is the catch-22 that there's nothing a PnP machine can do that you can't do by hand.  So you don't need a PnP machine unless you need to build boards in some quantity.  And if you need to build boards in some quantity, then you can't afford to have a PnP machine that requires constant dicking with to make it work right.

Absolutely. A "cheap" P&P really needs to be starting at something in the $1-2k range to bring it into a "nice toy" price range.

Entry cost could potentially be hugely reduced by having a "starter" configuration that can pick loose parts from trays using vision.
They also unnecessarily increased their entry cost with the whole cover/enclosure thing, which could easily have been made an option.

I think a lot of the would-be desktop PnP sellers grossly underestimate the customer support costs.  People buying these machines aren't going to want to fiddle with them themselves and be happy to learn and fix on their own.

I'm not sure that's the case - anyone buying at the low end is not going to expect the same level of support as Mydata etc. Also, the people buying will generally be fairly technically skilled, unlike someone buying for a routine production line to be operated by semi-skilled staff.

Anyway, it doesn't look like this one is going to fund.  I don't think the creators 'get it'.

Of all the attempts so far this has had a lot more clue than others in terms of spec, and the first to understand the importance of feeders, however I think they've missed the boat on entry price - it's just too big a risk on an unknown. ISTR reading they had some other sources of funding, so if they do produce it and people see it in the wild it might concievably turn out to be competitve with the Chinese machines. 
 

They weren't interested in open software,

Nice, but not a big deal, if it's been done right. My biggest criticism is that it's a PC peripheral, which immediately brings a ton of support issues. Considering how little processing power is needed and how many cheap SBCs are out there, making a machine be standalone is a no-brainer.

The winning play for a desktop PnP supplier will be feeders.  Make and sell a generic feeder first. 

A good feeder design could probably bootstrap the development of a full machine.

[Holmes34]

Good questions Corporate666. I'd just like to make the point he expressed 0.03-0.05mm repeatability, not accuracy. Repeatability is measured by doing the same cycle in the same conditions over a certain lapse of time and should naturally tend to 0.00mm.

http://blog.robotiq.com/bid/72766/What-are-Accuracy-and-Repeatability-in-Industrial-Robots

You are correct, of course, however on the KS page they say in the specs "Placement accuracy: 30 microns repeatability".  As a mechanical engineer by education, this somewhat makes my brain squirm, but I know what they are trying to say.

One of the problems I have with all of these companies and creators developing 3D printers or any other motion systems is the ever-more ridiculous claims about how precise their machines are.  I've heard numbers like 5 microns or 10 microns.  In my business we own and operate "real" CNC machines and in many of our parts are often machined with tolerances of 0.0001 (2.5 microns).  This is a standard tolerance for a press or interference fit or for a bearing bore, etc.  I know how difficult it is to maintain such a tolerance... it requires a well built and well maintained machine with ball screws and precision encoders and precision drives, and you have to account for thermal growth of the machine - even in a temperature controlled shop, being in sunlight will throw off such tolerances.  And you gotta check it with the right equipment - gage pins and bore mics and such.

Anyway, when some 3D printer manufacturer on KS talkes about their 5 micron or 10 micron abilities... using ACME screws and belt drives, it's bullshit.  Especially when they are then squirting out a blob of melted plastic. 

In terms of PnP... it matters in terms of really fine pitch components, especially BGA and QFN.  And most especially when you get into rotational errors.  If the placement isn't "dead nuts accurate", then it can really be a problem.    I am not saying these guys are lying - but the claim of 30-50 microns is impressive.  Very impressive.  I don't think my Quads can achieve that and they use ballscrews and a better/more rigid overall setup with a more 'advanced' nozzle system that would have less flex.  So I am skeptical of the claim.  And my feeling towards claims is the only ones that matter are those that the manufacturer is willing to stand behind.

So if they will consider a machine that isn't placing parts repeatably in the same location within 50 microns to be suffering a problem that is covered under warranty, then my hat is off to them and I am curious how they have achieved this level of performance.  If it's just a calculated number based on stepper resolution combined with gearing, then it is a worthless metric in the real world and users may be in for a rude awakening when they go to place fine pitch QFNs or BGA's or such and the machine can't do it - hence the query about whether it would be a warranty issue.

Does it really need to be "dead nuts accurate" when you are going to be reflowing the board anyway? There can be plenty of movement during reflow so if you're mucking around getting super high placement accuracy, is it not like polishing a shovel before you dig a hole?

Not to mention you have to physically transport from PNP to reflow oven!

[mikeselectricstuff]

Does it really need to be "dead nuts accurate" when you are going to be reflowing the board anyway? There can be plenty of movement during reflow so if you're mucking around getting super high placement accuracy, is it not like polishing a shovel before you dig a hole?

For chip compments no, but heavy parts with small legs like QFPs won't pull very far onto the pad. For larger QFPs, rotation accuracy is also increaingly important.

Not to mention you have to physically transport from PNP to reflow oven!

Solder paste is sticky and has surface tension. Once it's placed it won't go anywhere unless it's big & heavy.

[rx8pilot]

Not to mention you have to physically transport from PNP to reflow oven!

I pick up the PCB, and carry it to the oven. Like Mike said - the parts stay in place unless you are very rough.

Yes - dead-nuts accuracy is important. Not many boards are only 0805 passives. Anything useful these days will have bigger challenges than big chip passive parts. Once you make 100 PCB's and each one need repairs - you will appreciate the why that is true. Re-work will very quickly eat up any time the the P&P machine 'saved' you. Tiny mistakes cost large amounts of time. Time is really what you are purchasing with a P&P line. The only thing that counts is how many parts you can place per hour ALL-IN. The speed of the machine is sort of interesting - but the whole process is what really counts. If re-work of bad placements is needed, that kills all the other benefits.

[l0wside]

Couldn´t agree more. With a .5mm QFN, placement accuracy needs to be better that 150µm, rather 100µm (or 4mil). For the open-loop machines in the sub-10k range, this is a major challenge.

[mrpackethead]

Couldn´t agree more. With a .5mm QFN, placement accuracy needs to be better that 150µm, rather 100µm (or 4mil). For the open-loop machines in the sub-10k range, this is a major challenge.

Our turn of the century vintage Yamahas run in the order of 50um..

[l0wside]

But they will neither fit into your basement nor on your desk. And a couple of feeders will already cost more than an open-loop machine.

You get what you pay for, and what you are willing to sustain. Getting a 20 years old machine to run like rx8pilot did is not necessarily fun either.

[rx8pilot]

But they will neither fit into your basement nor on your desk. And a couple of feeders will already cost more than an open-loop machine.

You get what you pay for, and what you are willing to sustain. Getting a 20 years old machine to run like rx8pilot did is not necessarily fun either.

That is all true. The thing to consider though is this: Will the addition of the machine be faster and lower cost than hand assembly or outsourcing. I would LOVE to have a super small, cheap, easy to use machine where the feeders all cost $10 that could accurately place nearly any component really fast. The reality is that the current offering of low cost small machines have so many limitations that they don't really fit into a business.

[mikeselectricstuff]

The reality is that the current offering of low cost small machines have so many limitations that they don't really fit into a business.

True but for how long?
If Neoden or whoever were to really get their act together, a contract assembler buying a few Chinese machines could become a serious alternative to one $50k low-end machine from the established players, with more flexibility.



 

[rx8pilot]

True but for how long?
If Neoden or whoever were to really get their act together, a contract assembler buying a few Chinese machines could become a serious alternative to one $50k low-end machine from the established players, with more flexibility.

I don't think it will take much longer, but the winner will be the one that focuses on a business solution - not a price point. It would be nice to have 4 small machines over a single large machine for the type of work I do. A few years down the road and we will probably see that a lot of small businesses were enabled by in house assembly. Neoden has the bones of a good product - they just need to deal with the details.

At this exact moment - there is a lot of pain and suffering in assembly.