My rule of designing stuff

[throbscottle]

Rule number one. Don't lay out the pcb till you've finished the schematic and shown it works. Dammit.

Well it does say "rule" in the title, not "rules"...

 

[Scotophor]

Rules are for wusses!

 

[throbscottle]

That's why there's only one - "rule is for wusses" - you just can't say that without sounding like it's the wusses's job to rule!

 

[tunedwolf]

I built 10 small micro boards for a pal a good few years ago, they used a texas otp microcontroller that was in a 44pin flat pack. Anyway, got to board no. 8 and decided to take a break and test one of the built boards. Nada, nothing, just sat there doing diddly squeek. Much probing, measuring and thinking later, I realise that I had been picking the processors out of the "yet to be programmed" box. I had bought 25 of them just in case of wee accidents and when I opened them, I just flipped the box lid over and began programming them. After which I went and had lunch, came back and picked from the wrong side. It was really obvious too, just couldn't help myself

Rule no. 1 went fine, it was rule no.2 that got me

 

[MrAl]

Hi,

Here's a fairly good strategy:
1. Design
2. Simulate
3. Order Parts
4. Breadboard
5. Layout
6. Etch
7. Assemble
8. Box
9. Ship

Note between Design and Layout there is "Breadboard", where it gets actually tested.
In some cases the layout and etch may come before breadboard, so you can test the actual PCB too.

 

[canadaelk]

7.1: Test
7.2: Curse
7.3: Troubleshoot
7.4: Fix
7.5: Test
7.6: Document Fix.......

 

[Tony Stewart]

To avoid endless design fixes and reliability failures and field returns on high volume items or environmental failures, I have always been successful with this philosophy when quality, cost and time of design all are critical . That being said, I have designed products for Lucent from contract to part-functional prototype ( 1U custom rack sheet metal, artwork , silk screen , boards, assembly, power supply sourced ) appearance, and user interface. in less than 8 weeks all from a paper napkin spec.

1. Write a Design Spec... Define all requirements for yourself in writing; I/O Functional, Cost budget, Environmental requirements for EMC, Acoustic, Climate, Mechanical ( e.g. drop test and UL Coke spill test on top.)
Confirm it with customer
- Break-down the design spec into smaller chunks , Functional blocks, User Interface, Appearance , DFM, DFT (Testability requirements), cost budget for BOM and split budget for each section, design for low labour and max profit based on quoted price
- Delegate the tasks and set goals for completion of each section with major milestones and schedule daily quick reviews and major design review as required with customer.
2. Create a Virtual Design Team & Start the Design - Brainstorm all options, and weighting factors for each to determine best balance of cost, performance, time. Define all the technical issues early e.g. is air flow required? Does fan make too much noise? Can a temperature controller be made for $3 , will the layout cause interference , will shielding be needed, line filters? WHere are the hot spots? Is it realistic. What happens if a part fails ? Will it be detected easily by customer? What is the repair plan? All the while options for hardware vs software and Make vs Buy for any portion of the design
3. Have someone else check your work. unless your expertise is exceptional
4. Review all the issues daily. Requirements, Budget, Time, challenges. Don't stop until you are sure nothing is overlooked.
5. Get it built by reliable sources. 3 day PCB turn. Get Sheet metal builder to design the box for volume orders.
5.1 Be creative like for an 50 wire ribbon cable in 1.5" height space design an origami cable folding method from board to front panel non-std customer connector interface and farm out the design to a shoe repair buddy with creative thinking for doing 7 precision folds for $3 in volume 300 pcs.
6. Test it for extremes and find all the weak links then fix them. e.g wider component tolerances, hair dryer test and cold spray or dry ice & picnic box
7. Review hands on Prototype with customer in 8 wks
8. Help Purchasing with lowest cost sourcing of special parts ( Like low profile hex screws, thermistors, and 24V 1.5" fans

Will finish later.... work to do.

Bottom Line is the best designs must start from a the best Spec. with nothing left out and it should not be "Implementation Specific" unless based on a preexisting requirement. Meaning a good design never starts with what MOSFET or uC should I use . NEVER ! A good spec is the basis for DVT and documentation after to be verified by QA and eliminates false assumptions.

A good design spec will cover all the things found in the spec for a resistor. Electro-mechanical requirements, Climatic Environment, limitations, not to exceed, input out functions , Failure modes ( open or short) Fault Detection requirements, Fault Isolation requirements etc etc. before any implementation design even begins... even if the requirement is None, it should be stated. This oversight is the biggest fault with rookie designers. I dont mean a 100 page Military spec but at least as long as detailed as any commercial spec for a component, now applied to a system.

 

[gary350]

1. Design
2. Build proto type
3. Test
4. Make changes
5. Order parts
6. Layout board
7. Build
8. Test
9. Play, have FUN.

 

[throbscottle]

This is turning into quite an impressive thread, considering I was just moaning at my own lack of vigilance in er, everything really, for my little hobby project I was Sooooo confident was right first time!

Keep it up folks!

 

[Externet]

Actually, the electronics industry makes the prototypes on real PCB. If changes / fixes / improvements are needed, another tweaked real PCB is sent to be manufactured until all goes well to release to production and then ordered in large quantities.

 

[Blueteeth]

Always add redundancy. Any spare PCB space not required for groundplane, or connector clearance, add a few footprints with breakout pins/pads so you can add jumper wires/0 ohm SMD resistors/caps, extra decoupling caps, and of course - test pads, both SMD and through-hole for hooks. If you don't need em? don't solder parts to them. I also often add solder jumper pads, which I leave open until I have fully tested everything on board, this can be power to a micro, or an enable line for a chip, then just a blob of solder, and its ready.

On double sided boards that aren't high frequency (read: microwave), plenty of via's helps too, so if needs be, you can use these as small holes for soldering kynar/enameled wire for jumpers/tests.

Probably only done runs of 50+ maybe 20 times, but I've made hundreds of PCB's, and I still forget things. The 'prototype' phase usually lasts 2-4 home etched boards before I send it off for mass production, and even then I get nervous. Nothing worse than having to do a small 'two jumper wire and a resistor' mod on 40 boards

 

[throbscottle]

Nothing worse than having to do a small 'two jumper wire and a resistor' mod on 40 boards

Really? I thought there was nothing worse than paper cuts

 

[KeepItSimpleStupid]

I always thought he best design breaks within a week after the warranty is up. That's perfection.

The stuff I built or fixed for a research lab, I used the "reusable", "recyclable", "repairable", "modular/reconfigurable" and "I don't want to see it again". Unfortunately, during my tenure not much became "recyclable" or the ability to re-use it's [arts to make something else. "Re-configurable" seemed to be the norm. Stuff I built was still being used 25 years later.

I absolutely loved building controls with DIN rail stuff.

Some stuff that didn't use my philosophy, I "refused to fix" or pulled the "It has to conform" to this before I will deal with it.
The idiots in one case designed a shutdown controller similar to mine. It's features:
1) Sequential alarms. You have no idea what tripped the alarm or if there were two.
2) If the alarm momentarily tripped, you had no idea what alarm did it. The alarms didn't latch.
3) Instead of building it into a wall cabinet, they built it into a rack. Nearly 98% of the sensors were building related.
(shutdown buttons, air velocity, fire alarm, etc.)
4) Connected via screw terminals so it it was nearly impossible to remove and troubleshoot.
(I would have at least went with slides)
5) No readout on air velocity and no latching meant you never knew when to clean the sensor.

Eventually, I managed to run all of the building stuff into a big box and connect the rack via a large CPC connector.
Now at least it could be removed and reconnected with a few inter-connects to the second rack case.
And I got power run to that interconnect box, to run the air velocity alarm. The box needed an indicator during an upgrade for some reason, but i forget what it was for.

On the systems I designed, I would not allow a light bulb to be powered from a supply in another room. It had first go to the termination bock and then to a relay that used the box's power to power the indicator. That way, I was not likely to shut down a remote system accidentally when troubleshooting.

 

[Mosaic]

Sometimes, when building something for yourself....you leave out a lot of spec.
Then you can take advantage of creative thinking in midstream. Sorta like leaving room for feature creep.
My latest little project.....an Iron Man Arc Reactor that's downsized for kids as a wearable used such an approach.
You see, as a low volume DIY for folks at home...you cannot spec. a custom injection moulded housing. You have to make do with household item availability.
So I find a common item for a housing & faceplate and work backwards, changing and eliminating features based on space constraints and the # of I/O pins left, based on the chip size that would fit.
So basically the housing had a big impact on everything else. So I had 4 specs, dimensions, internal battery, thru hole discrete parts and blinky lights. The rest just sorta happened over 3 or 4 days.

The squeeze to fit all thru hole parts and still maintain decent brightness while compensating drooping battery voltage with only about 0.3V headroom over the LED Vf had me switching from a constant current supply for the 7 LEDS to a current mirror approach. I managed to do away with the 7 ballast resistors at the cost of about 12% higher current draw for the mirror ref. So that caused me to tweak the PWM behaviour to achieve lower average currents while maintaining interesting light effects.

The end result was different from what I had envisaged, but still a fun gizmo. It 'created' itself during the build.