How many PCB prototypes does your company allow in hardware design?

[Just_another_Dave]

Designing a PCB that works perfectly in a single iteration is pretty hard. When I designed PCBs in my previous employment, the project managers used to consider that it wasn’t going to happen and, therefore, they focused on mitigating the impact of the problems that could happen. Prototypes were normally designed to be as modular as possible, so even if something failed, other subcircuits could be tested. Additionally, breaking the system into several PCBs meant that you didn’t need to replace the entire prototype if something failed.

Additionally, each module was reviewed by the entire team as well as by a second team that was not involved in the project before manufacturing it (sometimes someone that doesn’t know anything about the pcb can find routing mistakes that people involved have overseen)

[Psi]

Designing a PCB that works perfectly in a single iteration is pretty hard. When I designed PCBs in my previous employment, the project managers used to consider that it wasn’t going to happen and, therefore, they focused on mitigating the impact of the problems that could happen. Prototypes were normally designed to be as modular as possible, so even if something failed, other subcircuits could be tested.

Exactly.   

If a manager/company is relying on the prototype to work first time then they're not doing their job.
If a manager/company is expecting the prototype to work first time then they don't understand prototyping and are probably wasting resources that could be better spent iterating on the design.
If a manager/company is planning for the prototype to have some issues and trying to mitigating those risks as much as practical then everyone is working correctly as a team and you're lucky to work there as opposed to the other 2 options.

[dietert1]

I also think peer review is crucial in order to remove expensive errors and share responsibility. And avoid overly complex prototypes but rather make some test board for design confirmation. I often used test setups with evaluation boards and other modules before even thinking about PCB design.
One needs to learn enough to be able and bodge the first prototype PCB into a working one, so its revision will be 99.9 % correct. Once a prototype is working, there shouldn't be a problem to order the revision.

Regards, Dieter

[coppercone2]

I mean I think its like a sport team, you wanna win, its not guaranteed, the better your team the more likely it is, but its chaos and you should not expect miracles. Their supposed to be like coaches. Only difference is that in most cases coaches played the game and know the game very well and are retired because of injuries or age... at least the best ones are like that.. and most of them are very interested in the game outside of work like a hobby/past time. With how corporate America works, you are likely to get one that never played the game and is not interested in it outside of work... so you need to set your bar a little lower.

And the analogy falls apart more because its so much more chaotic and complicated then a sport game in terms of variables, plus the performance lasts for possibly years.

I think it has to do with setting reasonable goals. I.e. that you expect the power rails should work for the first time and that the traces are sized not to fuse... there is a reasonable standard you can go by in terms of expected outcome.

[free_electron]

Are we talking the pcb layout or the entire design ?

Entire design : as many as needed. Build a number of prototypes. Possibly sections of the circuit. Simple rectangular board to get bench testing going and develop firmware. inject lots of testpoints and dummy parts or cuttable pathways.

As the design nears completion start working on form and fit. Work within mechanical constraints: size, shape.
After thermal , emi , vibration and other aspects have been tested respin the board. During this cycle component selection may change. Different connectors, transformers, capacitors. Different locations. Final enclosure shape may change.
So the board needs to adapt.

In telecoms i respun board because of components impeding proper airflow for cooling in the 24 inch telco racks. Or respin them because they had issues during assembly ( all heavy components on one side giving problems in the pick and place machine.

In automotive I made modules that hade many late changes. The factory floor people complained it was difficult to plug in the wiring harness because of an odd bend they had to do on the conveyer. it was difficult to reach-in to plug in three connectors. Scrap the entire design, go to a different connector, position it differently. Later on we ditched it again because there was an additional cooling duct required above it for the power hungry computer module above it. ( between glove box and footwell ). Things change during design, especially when the board is part of a larger system. All these aspects were looked at. How difficult is it to assemble, how does the module fit, thermal, vibration, assembly time, installation time. Mosfets did not behave as intended. Global part shortages. Footprints were made so they could hold many different parts allowing for late-game changes. Optimize for mass production so it was faster to assembly with higher yield. It's a continuous process.

Space was yet another can of worms. A launcher undergoes extreme vibration during take-off .J-bend stacked ceramic capacitors literally flew off the board. Ferrite core transformers would crack their core and bits and piece would go flying.
Connectors had issues with foreign material ingress. Do we put the male or female on the board ? How many times will a connector be removed during assembly or launch operations. Where is the risk for damage. A bent pin on the wiring harness is a disaster as you can't easily swap the harness. But you can easily swap a module. A socket can get plugged by debris. Sometimes connectors need ability to be inspected visually. We're talking 38999 family onnectors that cost 500$ a piece... Someone has to reach into the interstage to plug in the GPS simulator while standing on a boom lift. Come launch day : remove simulator and plug in the stage computer.

So these boards underwent extensive testing on vibration tables. The enclosures woudl flex and warp and needed reinforcement ribs. This would impede part placement and mounting hole placement. Boards would vibrate in spots and crack vias so additional mounting holes were required, or needed shifting of the holes to stop resonance points. Then all of a sudden the module was to be rotated or flipped in the nosecone or around the fuel tank and now the entire dynamics changed ... -> respin board. We need more sensors inputs -> respin. we need more actuators. we also want to monitor xyz -> respin respin respin.

Same for aircraft. Rotate, consolidate, add stuff. The power controller got mechanically so complex that it became unreliable due to all the interconnects. Migrate to rigid flex. The board folds post-assembly and 90% of the interconnects are gone as they are now part of a single piece board instead of 4 boards plus associated wiring harnesses. Only two 38999 connectors and a power busbar remain. Electrically the system has not changed. But the pcb layout has shifted technology , form , assembly and installation to make the entire assembly more reliable and easier to manufacture. No more harness mistakes or bad connections between boards. It is one board and can be fully tested pin to pin before going into the mechanical assembly.

it is unreasonable for a design to be frist time right, let alone the board to be first time right.

The only place where i have known first time right is in silicon design, but that is a totally different world. There you have the tools and the timeline to do this. Simply becasue building the first run is so massively expensive. You can't prototype. PCB is cheap ( a few k$) and fast (3 to 4 days if needed) . Silicon not so much ( millions of $ for a maskset) and 6 to 8 week turnover in the factory.
In silicon world you take your time and simulate the snot out of the design. You have the tools and data to do it ( which also cost a fortune)

For PCB level this is not required. The respin is fast and cheap. The PCb level simulation tools are nowhere near the capabilities in the silicon world. They don't need to be different problem to tackle, different cost, different domain.

[coppercone2]

Mosfets, space

in this case my sports team analogy becomes "expectations on the performance of flash gordon" and/or vulcan baseball team management

[djacobow]

Not only is "getting it right" on the first iteration unrealistic in most circumstances, it's not even a smart goal. It wastes a lot of time with extra reviews and stress that are unlikely to turn up the issues.

The cost of pcba had moved so much that it's better to optimize for quick cycles. I do like to make sure that early spins are highly testable and as amenable to hand rework as possible. I generally like to try to get a board working via rework before spinning it and I get very cross with colleagues who find one issue and immediately start a new spin. I think it's important to take each board as far as your can get it, to uncover as many issues as possible before spinning.

We usually get to an adequate board in three goes, but often management has moved to goalposts by then. Our product people are ... not good.

[RoGeorge]

Listen to Rick Hartley, then design good PCBs from the first try:

[wasedadoc]

Why should the hardware designers be expected to get it right first time when the software guys never do?   

[AndyC_772]

I can't email you a hardware update.

[wasedadoc]

I can't email you a hardware update.

The software guys didn't have that copout before the hardware engineers developed flash memory.

[max_torque]

There is a modern effect called "integration hell" which unfortunately tends to often dominate fast paced "high tech" projects, and generally those that are poorely managed (and IME engineering managemen is most often either shockingly bad or effectively non-existant.  Integration hell basically is that state where all the different teams of people working on the final solution have to put all their stuff together, and yup, i pretty much never works.


So the answer to "how many goes should it take to get to a production spec solution really, really, really depends upon two critical facts:

1) has the team desiging and developing this new pcb done this before?

and

2) how much "new" is there being implented?


So how many iterations you should plan for should clealy simply be an inverse measure of the answer to those two questions multiplied together.

So, you have a highly experienced team, doing something they have all done for years, well, honestly proddy spec should be board two imo (always a prototype for mostly physical reasons and typically because often the mech design is simply not progressed enough to allow a production representation, and why wait when you can get started!

But if you have a green team, incorporating lots of new stuff, yup here, a Unit Testing approach (ie lots of small individual test boards to prove out and develop each new bit) is clearly and demonstrably (ime) going to provide, on average the best return for their effort!



(PS ive done automotive electronics to the highest level (F1 and OE etc) for 30 odd years now)

[djacobow]

There are a lot of reasons, like EE being a much older and more established discipline and being built on other engineering disciplines with standards of how to do things. Software has a ... different ... pedigree and continues to do so.

But also, let's not kid ourselves; software is way more complex than board design, like many orders of magnitude so.

[twospoons]

Our standard project work flow allows for 3 planned revisions at minimum. Sometimes it takes more than this to iron out every last issue (we have an entire team dedicated to product validation). And thats for everything from simple 2 layer designs up to 12 layer HDI designs.
Expecting a prototype to be perfect is stupidly unrealistic.

[thermistor-guy]

How many PCB prototypes does your company allow in hardware design?
...

What about your experience?

In my professional telecom days, three.

First one was to prove the concept, and get some key features going.

Second one was to correct the mistakes in the first, including layout issues. Production would look at this
and suggest changes to reduce factory costs (assembly, testing), Production would look over the first one too,
but would focus on the second.

Third one was very close to the production version. Production would look hard at this, and confirm their
jigs/tooling against it. This would undergo full compliance testing. Often it would go into production.

Only once did I put a 2nd prototype into production. The layout lacked some load capacitors for a quartz crystal (doh!),
but it was a low-volume board, and had no other issues. So, through-hole caps were added to the board during
assembly. The layout revision was deferred until it was justified - either by some other problem, or by a circuit
change that would give a notable cost reduction.

More than three prototypes was taken as a sign that everyone involved weren't good at their jobs.

[tszaboo]

There is a modern effect called "integration hell" which unfortunately tends to often dominate fast paced "high tech" projects, and generally those that are poorely managed (and IME engineering managemen is most often either shockingly bad or effectively non-existant.  Integration hell basically is that state where all the different teams of people working on the final solution have to put all their stuff together, and yup, i pretty much never works.

This happens 100% of the time, if you put software guys in charge. Or hardware guys who have to report to software guys. because they have this idea that they can manage it, because it's similar. It isn't. I would much rather have a mechanical engineer as manager than a software guy. At least they understand that time sheet is a waste of time, and why I need to wait 2 weeks with the project and do nothing when waiting for the boards to show up.

[Psi]

Yes, managers who don't understand what they are managing is a big problem.  Especially when they "think" they understand it.

To manage effectively you need to hold a degree or better in the same industry and have at least 8 years hands-on experience (in the trenches).  And if you're managing both hardware and software you need that in both those areas.

[free_electron]

Oh, forgot one thing : any manager that tries to implement "agile methodology" should be fired into the sun.

[themadhippy]

To manage effectively you need to hold a degree or better

WALOB

[coppercone2]

a guile

[electronx]

The last PCB I made was tried for the first time in front of the customer and it worked successfully, I was literally sweating my ass off.There's a huge stress problem here. I think it's time for me to quit my job, I've made him thousands of dollars,

[RoGeorge]

The last PCB I made was tried for the first time in front of the customer

Why doing that?

[electronx]

I designed it  in 3 weeks, I didn't even have the chance to try it. I explained this to him, he thought it was simple to design the circuit.Profit per pcb is more than 1000 dollars

[Psi]

Oh, forgot one thing : any manager that tries to implement "agile methodology" should be fired into the sun.

I assume you mean, "implement agile methodology in a situation where it has no place"
Agile does work for some things but knowing when you should vs should not use it is key

[tszaboo]

Oh, forgot one thing : any manager that tries to implement "agile methodology" should be fired into the sun.

They are not implementing agile. Only, standup meetings, weekly reviews, timesheets, an agile software management tool that's incapable to follow waterfall methodology, kanban boards, and they keep on changing the requirements even after we started shipping the product. But it's not agile, you are mistaken.