Building one of something is tremendously easy. If you’re making one of something, you can cover the insides with hot glue, keep everything held together with duct tape, and mess around with it enough that it mostly works most of the time. Building more than one of something is another matter entirely. This is the thought behind DFM, or Design For Manufacturing. [Nick Sayer] is an experienced seller on Tindie and he’s put together enough kits to learn the ins and outs, rights and wrongs of building not one, but an inventory of things. Check out this last talk of the 2015 Hackaday SuperConference, then join us below for a bit more on the subject.
[Nick]’s hobby come business is called Geppetto Electronics, where he manufactures everything from battery eliminator boards for Apple’s Magic Trackpad to replacement electronics for the ubiquitous AA-powered quartz clock movements that make time go slightly wonky. He has the street cred and has produced enough random electronic doodads to know what he’s talking about, and while [Nick]’s talk only covers the basics, there are a lot of good tips.
This is not a talk about sitting at a kitchen table with a soldering iron, reels of parts, and a stack of boards. When it comes to production, [Nick] reaches out to assemblers, and that means pick and place machines. To do this you must know how assembly works. Assemblers charge by the size of the Bill Of Materials, placement count, and board/panel count. The best thing that will help any project is minimizing the size of the BOM – for most circuits, you might be able to get away with completely ignoring the difference between a 220Ω and 330Ω resistor. If you can, that’s two items on the BOM that just became one. Remember resistors in series, too.
When you order a stack of boards from Seeed, DirtyPCBs, or OSHPark, an envelope comes in the mail with your boards neatly stacked on top of each other. This is not what you want if you’re making hundreds of things. You’ll need to panelize them, and that means following the assembler’s instructions. There will be a maximum size, and you’ll need to put v-grooves or tabs in there so you can separate them later.
These are just a few tips [Nick] has picked up over the years, but by his own admission it’s not a complete accounting of the entirety of DFM. That’s fine – it was only a 15-minute talk. [Nick] did point out one of the old EEVblog videos that covers DFM in much more depth. Still, a great talk of what to do and what not to do when building an inventory and not just a project.
Decades ago I worked for a company that had two sections. One did custom control system, think racks of minicomputers. My section created micro-based flow computers. They sold one-off systems while we manufactured thousands a year. There was a friendly rivalry between us. Our taunt was, “Anyone can get 1 or 2 units working. Getting thousands of units to work reliably is a real challenge.” I always got a chuckle out of the last line in their mechanical assembly steps: “Cut to fit. Paint to match.”
I’m in the process of getting a project of mine done ( http://smdprutser.nl/project/70w-soldering-iron-controller/ ) However the manufacturer (somewhere in the shenzehn area) didn’t want a panelized but just the gerbers. I guess they did make them for me. I also went to a small factory during Hackercamp shenzehn 2015, where you could see how it is done. The PCB’s were indeed panelized or at least they got a guidance strip to transpaort the project through various machines.
I guess they are specalized in small runs and they run multiple projects during the day. It was facinating to see the modularitry of the fcatory. The price didn’t change much depending on how many boards you get assembled, bsicly it was a setup fee and a few dollars per board, mostly lineair up to 1000 :)
Getting something into production is quite a learning curve :)
So far, I’ve found success with dealing with individual contractors for assembly and PCB fab. I’ve spent quite a bit of time comparing different vendor combinations, including various combined PCB fab and assembly houses. There are a bunch of ways that the process can be split up. Some assemblers will manage the parts procurement process for you (but they’ll charge you for that), some want you to just ship them all the reels. Some combined fab and assembly houses won’t take PCBs for assembly other than theirs.
Once you find a formula that works, it’s tempting to stick with it, but it’s important also to not get complacent about trying to find the best deal, but realize that there are a lot of components to “best” – not just price.
For me it is a hobby so I try to get a good relationship with a English speaking Chinese :) I think that is more important (at least on the scale I’m operating) then saving a few bucks. That is prolly the market this manufactrurer focus on.
And I do learn a lot on this project which I hope to apply to my next projects. On my blog I try to document the lessons learned :)
Can I ask what manufacturer / turn-key you are using? I was hoping to try out the game too in ** volume, hopefully not too steep a learning curve xD
I use Small Batch Assembly. They’re great for the small end of the scale.
DFM is certainly a consideration, but it is also often used as a way to decrease production costs at the expense of product maintainability and reliability. It can also be used as a way to support really costly ideas from consultants making claims about the benefits of their services. I saw one of these consultant types arguing to replace a few dollars worth of off-the-shelf parts with a custom machined item of significant complexity to save on the time spent by the $8.hr assemblers.
My hero for these was Muntz, who took the idea and eliminated parts, leaving sets that were less tolerant of component variation and less sensitive to signals. Considering he became both an icon and bankrupt in the TV business suggests that DFM taken too far can result in ultimate failure.
The real trick isn’t approaches or buzzwords, but simply to understand as much about the product, its application, its fabrication, its distribution, and all the money involved.
Because I’m an engineer, I was gifted a book, “The Tale of the Scale,” in which a non-engineer clearly demonstrates the challenges those not willing to spend time understanding a problem will face in solving it. At the end he breaks even and got to write a book about how to take years to design a product that could have been done in a few weeks by a qualified engineer.
I absolutely agree. You cannot properly apply DFM techniques without being intimately familiar with the product you’re making and its engineering. The fundamental rule I discuss in the video is that after all of your tinkering the thing doesn’t work, then it’s all for naught.
One thing I didn’t mention that I should have is that it’s not enough for one to work, really. Your design has to be robust enough to accommodate all of the parts variations you will encounter in your manufacturing run, and robust enough to function properly for its entire service lifetime. And you can go even further – devices that use radio frequencies must pass FCC Part 15 muster. Devices that use line voltage power may need to get UL (or regional equivalent) approval.
And – again – the risk you run is that you go and build 1000 of your thing and it comes back and is worthless. All of the money that production run goes right into the landfill along with your production run.
It’s a big, big subject, and all I was able to hope to do was highlight some of the mistakes I’ve made along the way and some of the resulting lessons.
Thanks for the reply – I’m reminded of the quote attributed to Edison – ‘I didn’t fail 1000 times, I found 1000 ways not to build a light bulb.’ Thanks also for the talk.
Favorite anecdote from a prof of mine with a Civil PhD, He got a job analyzing shoulder-fired missiles. They’d gone through development without hick-up, but the first production units disintegrated early in the flight. It was at an epoxy joint between two body tubes. Checked the protolab and they are going over the joint carefully with acetone before applying the precisely metered and mixed epoxy. Go to production and they have oil contaminated rags to clean the joint, epoxy prep is some of this – some of that, and smear at it a while before slathering it on. Sigh.
I think a lot of projects get knee-capped by having carefully made prototypes that disguise the slim margins. Even moving a production line can cause a design ‘we’ve made these for years’ to suddenly become 100% scrap.
YES!! A million times this…
As an Industrialization Engineer this is all i do. The company im at does it to the extreme. DFM is only manufacturing. My company has us run design for manufacturing, test and assembly. To simplify we call it DFx. We also having gating levels in the project, Called Stage Gates. each gate has things that need to be done to move on with the project and it is broken up between all the departments. My group has 18 areas (specified in 18 documents) and each document has like 5-15 things to accomplish. (sorry to ramble on, i could actually do an 8 hour video on what we do and still not cover it all)
Anyways, i agree with the above responses this DFM process could take a lot of time. The funny thing is that the guy that took 2 years to build something and break even is exactly how long my company takes to design and deploy something. Which seems normal to me now after 12 years with the company. So im not sure where three_d_dave says could be done in weeks comes from. unless the project was super simple to design and build.
Just google Sony PS4 development. You will see they started in 2008 and not released until 2013 (and only the us) So they took 5 years to develop and deploy. So complexity of the project really determines length of project.
My company designs/builds complex devices with embedded linux. Always trying to build a faster better product using new Processors and FPGAs, going from nor flash to NAND flash (which has all sorts of strange problems running an OS on NAND flash) i miss NOR flash just not the cost of it. Also new mechanical designs (possibly moving parts)
It’s not as simple as having the right engineer. You need a good schematic engineer, board engineer, mechanical engineer, software engineer, systems engineer. Try all that by yourself on something complex and its going to take a couple years.
How long has the simple design of Oculus Rift taken? started in 2011. kickstarter around 2012 Only development units went out. 2013 hd prototype, 2014 dev kit 2, version 1.0 of SDK out in 2015. Facebook buys it. End of 2015 and where is our oculus rift for consumers? Maybe a real engineer would have had it already out and running. How hard could it be right? put a fast imu, HD display, couple lenses and a ski goggles and bam your done right? two weeks, ha.
To save reading the book, it was a ten on&off years to put a two strain gages onto a small metal piece so the output was roughly linear. His first attempts resulted in nearly 0 output over the range of the applied load. 2 weeks would have been enough to accomplish this task. He apparently didn’t ask anyone who ever did stress work to help him. He didn’t create a sellable product – just a design for a load cell. He was trying to make a thinner electronic bathroom scale, apparently assuming that the last 100 years of scale-makers had missed a breakthrough.
He did get some patents, but it looks like no one ever made any scales based on them.
He’s not a stupid guy, but he is a really bad engineer and he documents how not to design things without expressing any lessons he should have learned. In his own field he seems quite intelligent and he would probably be as critical if I wrote a book about how to handle housing policy without doing any research on the topic ahead of time.
http://sollyangel.com/
http://sollyangel.com/category/design-invention/
http://www.amazon.com/The-Tale-Scale-Odyssey-Invention/dp/019984982X (Check the used prices for the hardcover)
This was a design for a load cell that doesn’t seem to have ended up in a product, over 10 on&off years of development. It took around a decade to design one that worked, not to produce a sellable product. It should have taken a stress analyst about 2 weeks to develop this part.
In another (now hidden, lost, missing, deleted, moderated post) I had links to his page and to Amazon. Prices for his hardcover are as low as $.01. There was no realization even at the end of the book that he could have looked at how strain gages work and how stress/strain works before sketching up solutions.
He’s a smart guy, but way out of his field. From one Amazon reviewer – he wrote the wrong book and should have discussed how he reversed an ugly situation in housing policy in Africa rather than this self-indulgent story.
gearvr first dev kit 2014, second dev 2015, consumer nov 2015
…unless n=0
Nice.
Should have been “building 10^n is not the same to build 10^(n-1), n>=0”
you beat me to it
+1, as soon as I saw “25 comments” I knew someone already had to have posted this
Dude.. I’ve got one of your Crazy Clocks. Really really cool.
Great job on everything. Looks and works perfectly. And you got it shipped out super quick. Thanks!
Only thing I can add to the wish list is storing multiple modes in a rom or something so you don’t have to reflash it, but that could have been a marketing thing to get people to buy multiple units. I would have paid an extra 5 bucks to have it’s modes be switchable.
You’re not the first to mention mode switching. But the number one goal of the design was to make it as small as possible so that it could fit inside the movement.
That said, the ATTiny85 does have one spare pin you could use. But if you really wanted to get sophisticated, you’d need to switch up to the Tiny84 to get more pins.
one pin is enough: hold button for 6 ticks to get mode 6 .
Re flipping gerbers to avoid DirtyPCB’s tracking numbers on front panels: I tried that and it didn’t help. I guess it’s either random or they always put in on a side with silk screen. For front panels I now use Elecrow who doesn’t add tracking numbers.
Good to know. The last batch of front panels I made I went with MG circuits, and they came out perfectly. OSHPark is also an option, as long as you’re good with white-on-purple.
All my designs with dirtypcbs.com and their new service had the numbers on the front. Did you do something special on your designs?
Apple’s Tragic Macpad…
I see what you did there…
Alas, my Magic Trackpad Plugger is sort of an orphan product now. Apple revved it and now it’s got non-removable rechargeable batteries. I originally designed it because I saw a bunch of instructibles where people had used dowels to connect USB power straight to the trackpad. I didn’t think that was a good idea, since two AA batteries are nominally 3 volts, not 5. So my design included an SMD 3.3v LDO.