Tips for improving reliability?

[patrick_2]

I'm designing aftermarket engine management for cars, a standalone ECU I want to sell for profit. I want to do as much as I can reasonably afford to improve reliability. So far a lot of the parts are derated to 20%, spec'd for operating at temps higher than required. Critical circuits have been tested on the bench for months, and some have been run in an oven to do accelerated testing though not all have gotten this yet. We have not done any vibration or cold temp testing.

I'm looking for any ideas to improve reliability, mainly regarding the pcb side of things. Anything to consider or ideas regarding design or PCB layout to improve mechanical, electrical, or thermal reliability would be appreciated!

[free_electron]

Are all your parts AEC-Q compliant ?

then next step is robustness of the design ( input conditioning and output protection ), ESD and load dump handling. lastly EMC ( but susceptibility and generation )

[CatalinaWOW]

Depending on mounting an ECU will see a fair amount of vibration and some fairly extreme thermal cycling.  Testing is the only way to really sort out vulnerabilities here, but making sure that large and/or heavy components are adequately mechanically supported and checking that modal frequencies for PWBs don't align with expected vibration input are the normal precautions.  Be cautious about tight clearances (things that might touch which translates to bang together) with slight bends.  It is usually faster to test than analyze thermal cycling problems, but looking at the CTEs of objects that are rigidly attached to each other at one or more points helps.  Things like the case and PWB, or heat sink and PWB.

[T3sl4co1l]

EMC, and environmental.  Relatively large electric fields (10 V/m, various transients including ESD).  Wide temp and humidity swings.

I'd be surprised if a bare proto board can last a week in say, Seattle, or Miami?  The kind of thing you can basically only solve by: 1. sealing the board in a hermetic enclosure (or at least IP67), 2. conformal coating or 3. potting (or any combination of the above).  Actually very simple, and obviously standard procedure, for automotive, but if you don't know, you don't know.

Tim

[Gyro]

Some boards I've seen recently in high vibration humid environments have been 'shallow potted', ie. enclosed in a shallow plastic subframe within the enclosure and then potted to a depth of about 3-5mm above the PCB surface. This is sufficient to securely anchor larger items but avoid problems with allowing for things like capacitor venting and connector mating.

[Brutte]

Visit some scrapyard and buy several modern ECUs. Could be for engine or for power steering. Just focus on safety-critical. And then just follow these designs.
Mind that an ECU that is bolted in a passenger compartment has much different construction than the one that rides on an engine or on a belt driven powered steering.

[nctnico]

I'd be surprised if a bare proto board can last a week in say, Seattle, or Miami?  The kind of thing you can basically only solve by: 1. sealing the board in a hermetic enclosure (or at least IP67), 2. conformal coating or 3. potting (or any combination of the above).  Actually very simple, and obviously standard procedure, for automotive, but if you don't know, you don't know.

It entirely depends on where the circuit is mounted. My previous car from Mazda had an injection control module and an ECU. The ECU was under the feet of the front passenger seat. This was a simple FR3 board with through hole and SMT components in a metal enclosure. Just like you'd find in a TV. The injection control module was a different story. This was mounted underneath the air filter box in a place where it had some airflow from the front but it was far from the radiator. All connections where spot welded and it had a ceramic substrate with semiconductors bonded onto it. The whole thing was submerged in silicon snot too.

[patrick_2]

Are all your parts AEC-Q compliant ?

then next step is robustness of the design ( input conditioning and output protection ), ESD and load dump handling. lastly EMC ( but susceptibility and generation )

Thanks. 99.8% are AEC-Q compliant, the couple things that are not, one is redundent, other is derated and in both cases I could not source a part that hit the standard and had the specs I needed.

I have I/O conditioning taken care of. I have not done ESD testing, need to look into that.

[patrick_2]

Depending on mounting an ECU will see a fair amount of vibration and some fairly extreme thermal cycling.  Testing is the only way to really sort out vulnerabilities here, but making sure that large and/or heavy components are adequately mechanically supported and checking that modal frequencies for PWBs don't align with expected vibration input are the normal precautions.  Be cautious about tight clearances (things that might touch which translates to bang together) with slight bends.  It is usually faster to test than analyze thermal cycling problems, but looking at the CTEs of objects that are rigidly attached to each other at one or more points helps.  Things like the case and PWB, or heat sink and PWB.

It will be in the cabin so luckily won't see temps from engine bay. Thanks for the info! I do plan to rely on testing and not simulation for thermal and vibration problems. Any good ideas to do accelerated vibration testing? I have a small oven I can program to do cyclical thermal cycling, don't have anything right now for vibration testing other than real world use.

[patrick_2]

Visit some scrapyard and buy several modern ECUs. Could be for engine or for power steering. Just focus on safety-critical. And then just follow these designs.
Mind that an ECU that is bolted in a passenger compartment has much different construction than the one that rides on an engine or on a belt driven powered steering.

Thanks, I now have six modern ECUs on order. I have went though some older ones but not anything in the last few years, so it will be good to do so. Great idea!

[CatalinaWOW]

Depending on mounting an ECU will see a fair amount of vibration and some fairly extreme thermal cycling.  Testing is the only way to really sort out vulnerabilities here, but making sure that large and/or heavy components are adequately mechanically supported and checking that modal frequencies for PWBs don't align with expected vibration input are the normal precautions.  Be cautious about tight clearances (things that might touch which translates to bang together) with slight bends.  It is usually faster to test than analyze thermal cycling problems, but looking at the CTEs of objects that are rigidly attached to each other at one or more points helps.  Things like the case and PWB, or heat sink and PWB.

It will be in the cabin so luckily won't see temps from engine bay. Thanks for the info! I do plan to rely on testing and not simulation for thermal and vibration problems. Any good ideas to do accelerated vibration testing? I have a small oven I can program to do cyclical thermal cycling, don't have anything right now for vibration testing other than real world use.

I'm having a mental block right now and can't remember the name for impulse input testing.  Basically you hit the DUT randomly with a bunch of precisely calibrated hammers.  If done right (which is very difficult, the hammers beat themselves up) you can get rapid and repeatable results.  I mention it because if you don't care as much about quantifiable results you can quickly evaluate the general robustness of a design and associated processes by setting up some solenoids and whacking your device around a bit. 

I mention processes because very frequently the cause of failures is only indirectly the result of the design.  Less than perfect solder joints.  Pieces of debris banging around in the enclosure or wedged under a mounting point were they totally change the loading.  Nicks in wiring from improper stripping.  Improperly crimped joints.   Good design can make these things less likely, but not eliminate them.  Bad design can make them almost impossible to eliminate.  And if you are chasing reliability you also chase process related things that many here don't believe in.  Latent ESD damage.  Latent damage from pop corning.  Electromigration.  Tin whiskers.  The list goes on and on.

[patrick_2]

Depending on mounting an ECU will see a fair amount of vibration and some fairly extreme thermal cycling.  Testing is the only way to really sort out vulnerabilities here, but making sure that large and/or heavy components are adequately mechanically supported and checking that modal frequencies for PWBs don't align with expected vibration input are the normal precautions.  Be cautious about tight clearances (things that might touch which translates to bang together) with slight bends.  It is usually faster to test than analyze thermal cycling problems, but looking at the CTEs of objects that are rigidly attached to each other at one or more points helps.  Things like the case and PWB, or heat sink and PWB.

It will be in the cabin so luckily won't see temps from engine bay. Thanks for the info! I do plan to rely on testing and not simulation for thermal and vibration problems. Any good ideas to do accelerated vibration testing? I have a small oven I can program to do cyclical thermal cycling, don't have anything right now for vibration testing other than real world use.

I'm having a mental block right now and can't remember the name for impulse input testing.  Basically you hit the DUT randomly with a bunch of precisely calibrated hammers.  If done right (which is very difficult, the hammers beat themselves up) you can get rapid and repeatable results.  I mention it because if you don't care as much about quantifiable results you can quickly evaluate the general robustness of a design and associated processes by setting up some solenoids and whacking your device around a bit. 

I mention processes because very frequently the cause of failures is only indirectly the result of the design.  Less than perfect solder joints.  Pieces of debris banging around in the enclosure or wedged under a mounting point were they totally change the loading.  Nicks in wiring from improper stripping.  Improperly crimped joints.   Good design can make these things less likely, but not eliminate them.  Bad design can make them almost impossible to eliminate.  And if you are chasing reliability you also chase process related things that many here don't believe in.  Latent ESD damage.  Latent damage from pop corning.  Electromigration.  Tin whiskers.  The list goes on and on.

Hmm. I was planning to do some drop testing, basically just drop it from 3-4 feet on concrete a bunch and then see if anything failed. It will be inside a metal enclosure. I definitely care more about quickly knowing if a design sucks so I can fix it and if it's tough, I don't really care about the numbers as long as the results are that it's solid.

The solenoid things is interesting, I have a couple 100lbF solenoids and a capacitor bank to fire them, so I could build something like you describe. In all seriousness, would it be just as good to just repeatedly drop the thing on concrete multiple times to test for mechanical robustness?

[CatalinaWOW]

Dropping on the floor is fine as VERY rough screen.  Results can be all over the place because of the variable dynamics of first point of contact.  On one drop you will stress things relatively lightly, on another you will break things that are far stronger than necessary. 

Whatever approach you take try to stress all three principle axes.

The advantage of the hammer approach is that you get hammer strokes a roughly second interval, so faster than dropping.

You should get some idea of your real environment to provide some context for your tests.  An assembly perfectly suited to the passenger compartment of a sedan could fail uniformly and quickly in the hell of a main motor mount for a satellite booster.  I suspect a little googling will find some relevant specs or measurements.

[CatalinaWOW]

The mental block broke.  HALT (highly accelerated life testing) and HASS  (highly accelerated stress screening).  Both processes can used to screen designs and once a design is in production to screen product if that makes sense.  Google will find reams of material on the subject.

[patrick_2]

Quick update and a question.

I got two of the six ECUs in today, a Ford ECU and a Toyota ECU. The ford is a 2015, the Toyota a 2008.

On the ford, the via size is tiny, I haven't measured it yet but at most 8mil drill, possibly smaller, and the anular ring is just tiny. The Toyota is 12 mil vias and with MUCH larger anular rings, the fords are so tiny in comparison.

Everything I've read says bigger vias are more reliable for several reasons. By my eye, the Toyota ECU is less complex (no surprise) but probably more reliable as the trace width, trace spacing, and via size are all larger on the Toyota vs the Ford.

Thoughts? If the primary concern is reliability, then bigger is better, right?

[asmi]

1. sealing the board in a hermetic enclosure (or at least IP67)

You've got to be careful while sealing stuff that goes through wide thermal cycles, as the moisture from the air that is sealed in can condensate/freeze and ruin the board. I had this specific failure mode with my device that was mounted outdoors, I solved it by adding a small pouch of desiccant inside before sealing the package.

[PlainName]

When I did some kit that would go on a motorcycle I built a vibration tester which was basically a hinged platform with the moving end attached to the center of a 6" speaker cone. I was primarily concerned about engine vibration, so with this I could select any frequency and amplitude within the audio range. Then I attached an accelerometer to a real bike and ran it over some tricky roads, and played back the recorded accelerations through the test rig.

The inertia of the test rig means you can't simulate every vibration you might encounter, not playback a recording exactly, but it is very repeatable.

[Jeroen3]

The mental block broke.  HALT (highly accelerated life testing) and HASS  (highly accelerated stress screening).  Both processes can used to screen designs and once a design is in production to screen product if that makes sense.  Google will find reams of material on the subject.

HALT testing is fun. You basically thermal cycle it (-40 to +125) while vibrating at the maximum of the test equipment (eg: 60g).
If your equipment dissipates enough you'll have the test house people running around and swapping LN2 bottles continuously  .

[CatalinaWOW]

Quick update and a question.

I got two of the six ECUs in today, a Ford ECU and a Toyota ECU. The ford is a 2015, the Toyota a 2008.

On the ford, the via size is tiny, I haven't measured it yet but at most 8mil drill, possibly smaller, and the anular ring is just tiny. The Toyota is 12 mil vias and with MUCH larger anular rings, the fords are so tiny in comparison.

Everything I've read says bigger vias are more reliable for several reasons. By my eye, the Toyota ECU is less complex (no surprise) but probably more reliable as the trace width, trace spacing, and via size are all larger on the Toyota vs the Ford.

Thoughts? If the primary concern is reliability, then bigger is better, right?

To my knowledge it is not absolute size, but ratio of size to process capability that is most important.  If the Ford vendor is capable of making 2 mil vias an 8 mil via will potentially have excellent reliability.  There are several interacting factors going on here so simple rules like bigger is better may not apply.  Of course the less complex ECU has an inherent reliability advantage.  Which might be overcome by implementation.  To understand what I mean by this, think about all ECUs.  These days they have 16 or 32 bit microprocessor containing literally millions of transistors.  Very complex by some metrics.  But that isn't dominant because the large market for semis has driven the reliability of the chip itself very high.  Complexity has to be evaluated very carefully.

[patrick_2]

When I did some kit that would go on a motorcycle I built a vibration tester which was basically a hinged platform with the moving end attached to the center of a 6" speaker cone. I was primarily concerned about engine vibration, so with this I could select any frequency and amplitude within the audio range. Then I attached an accelerometer to a real bike and ran it over some tricky roads, and played back the recorded accelerations through the test rig.

The inertia of the test rig means you can't simulate every vibration you might encounter, not playback a recording exactly, but it is very repeatable.

Thanks, I may build a little rig like that and do some testing. I probably have enough laying around to build something on the cheap. Got a 10" subwoofer I don't use I could use for the driver. Would you be willing to answer some questions about how you built your test rig?

Ok, so update. Thank you again for the recommendation to buy some junkyard ECUs. Got all 6 apart now and put them under a microscope and measured some things. Gained a lot of insight in several areas. Pretty interesting, very different design approaches depending on the brand. I did not expect to see that much difference between them. No consensus on design even from the big automotive companies.

[PlainName]

Would you be willing to answer some questions about how you built your test rig?

Sure. It was a while ago but my memory hasn't quite escaped yet

[T3sl4co1l]

1. sealing the board in a hermetic enclosure (or at least IP67)

You've got to be careful while sealing stuff that goes through wide thermal cycles, as the moisture from the air that is sealed in can condensate/freeze and ruin the board. I had this specific failure mode with my device that was mounted outdoors, I solved it by adding a small pouch of desiccant inside before sealing the package.

Yup, and even after manufacture, there's moisture diffusion through plastics, say.  Hermetic sealing is usually done with metal, glass or ceramic enclosures and seals of the same sort (welded, fritted, soldered..) or epoxy.  PITA for most electrical purposes, let alone servicing or debugging.

Tim

[nctnico]

Actually using a hermetically sealed enclosure is not a good idea at all (unless you fill it with a gas like Nitrogen). There will always be moisture going in & out. In my experience it is better to have a hole at the bottom of the enclosure so water can get out. A conformal coating over the board does the rest.

[T3sl4co1l]

Actually using a hermetically sealed enclosure is not a good idea at all (unless you fill it with a gas like Nitrogen). There will always be moisture going in & out.

No, and this is contradictory!

If there's always moisture in and out, then how would fill gas change anything?  Dry air, N2, Ar, SF6 -- they'll all equilibriate with ambient gasses, including H2O, O2, even what small amounts of He and H2 are around.

The point is to have a seal that is impermeable even to those.  Vacuum apparatus does this every day.  Metal-metal and metal-glass seals are typical, but even gasketed joints can be used to a certain extent.

This is the difference between a truly hermetic seal that does not allow gaseous diffusion, and something that is merely sealed against relatively rapid pressure changes like a plastic IP67 enclosure.

A sealed, but non-hermetic, enclosure might equalize in days, years or centuries, but if it's permeable to gas, gas will diffuse in and out, and condensation and oxidation will be a problem at some point.  (Obviously, if it's in the centuries, that's good enough for commercial purposes, but still not truly hermetic.)

Tim

[nctnico]

Actually using a hermetically sealed enclosure is not a good idea at all (unless you fill it with a gas like Nitrogen). There will always be moisture going in & out.

No, and this is contradictory!

If there's always moisture in and out, then how would fill gas change anything? 

I wrote 'unless' and the gas is under pressure. But in general a sealed casing without taking any special precautions is worst than having a hole in it.