ICT or not to ICT, opinions and experience about test methodology

Both require DFT and each has their own +/- points. WE use both. I guess the biggest + for ICT is where proven designs have no 'marginals' and the production process is already well proven. Functional - is that if you want a THOROUGH JOB, YOU MIGHT HAVE TO DESING IN EXTRA DFT features to get close to 100% coverage.

Functional. But we do test all the power supplies. That can be done with DVM test points, a ribbon cable, or, preferably, BIST. Most of our products have BIST, and that includes reporting all the power supply voltages, and occasionally some currents.

We try to do enough functional testing to catch any possible assembly or part error, and to verify the specs. If it's not measurable (say, one bypass cap out of 20 the wrong value) it must not matter.

We have an optical inspection machine that catches obvious assembly errors and bad soldering.

So, if there is some component that would cause the whole board to go up in a puff of smoke (maybe a voltage regulator) it might be smart to find some way to test for soldering defects before powering it up.

Otherwise, if a failed component or unsoldered joint allows a board to pass the functional test, why is that component there at all? (I guess ESD protection parts would be exempt from this question.)

Jon

Turn the question around: how helpful has your *past* use of this practice been in catching errors in YOUR process?

Is your process likely to change (degrade?) that would suggest less (more?) need for this "safety net"?

Or, the "functional test" doesn't provide complete coverage for faults!

On Thursday, January 26, 2017 at 9:10:40 AM UTC+11, snipped-for-privacy@gmail.com wr ote:

sistors, transistors, diodes, inductors, caps etc to find production faults

only functional testing. Functional is good if we can trust the SMD process and internal control to mount the components properly

Cambridge Instruments initially skipped in-circuit testing on the boards fo r their computer-controlled electron microscope, the S.360.

It was a disaster. The argument was that every board was attached to the VM E bus, and you could test the boards through the bus interface.

Sadly, there were a bunch of faults that stopped VME access, and it took a long time - and expert testing - to find and fix those faults. Crummy admin istration meant that some boards failed automatic testing several times bef ore they got human attention.

Lots of the faults were in connectivity - whisker connections, gaps in copp er, solder bridges or solder connections that hadn't actually got connected . Pre-testing boards for whiskers and gaps before they were loaded subseque ntly proved to be worth the expense. Unsurprisingly, the printed circuit sh ops that could do it in-house made fewer duff boards.

The production of the S.360 microscope ran at about half the rate it should have until management paid for the in-circuit test fixtures and programs, about a year after the S.360 had hit the market, which gave the competition enough time to bodge comparable features into their electron microscopes.

It represented a lot of lost potential sales.

You need to know a lot about the faults you actually going to find before you can make an informed choice.

Another thing that has helped reduce early failures is a "bake" cycle, running the units (possibly at slightly elevated temperature) to try to weed out weak units.

Of course with quality parts and a good design this should no longer be necessary - though an occasional sample run might be worthwhile, and less costly.

An example of this:

I took a look at some power boards a customer was having problems with. One of them, a simple visual inspection showed a D2PAK wasn't properly soldered.

From the looks, I'd supposed the paste had just beaded up on the pad, and not wetted the tab. Should've been caught by AOI, but sometimes people are hasty.

During burn-in testing, it roasted itself to death, from the lack of heatsinking through the tab and thermal pad.

It went through ET just fine, because the solder made continuity, but it didn't survive burn-in because there wasn't enough thermal dissipation.

(Curiously, despite the oxidation on the tab showing the transistor had well and truly been cooked, and having blown a fuse in the process -- the transistor ohmed out fine, so I desoldered it, cleaned up the tab and pad, soldered it back in, and it worked like a champ. Go figure!)

These boards were built by a well known(?) American CM, so a reminder that quality is an ongoing process, not something you can simply assert. Tested boards cost more. But lost user experience due to infant mortality is much, much worse!

Tim

I'm not sure I understand the difference? Functional means you just turn it on and see that it works?

We don't test any individual circuit boards. (besides a visual inspection for backwards tant's. and sometimes a power up to see there are no shorts or stuff.)

Once an instrument is assembled, there will be a test to see all the gains and such are right. For "involved" apparatus, I'll also take test data to test the whole thing. That is a bit of an art... "Oh there is a little extra wiggle on this 'scope trace... " Click.. turn the room light off, it's gone. (Photodiode pick up) Or, "oh that's not right." and it has to be investigated more. Lots of built up experience.

But we are very low volume.

George H.

We have AOI visual inspection. BIST we have also, and maybe would like to extend for more test coverage

An alternative way is to do sample testing in flying probe, which does not need test points, but that test will only find problems that are consistant, won't find single boards

Cheers

Klaus

Functional testing should be a test to cover most of the functionality of the device/circuit. So if a component is "off", you should see a result in the Functional Test, otherwise it is not needed in the first place

But, an "off" component could just be a slightly changed gain or crossover of a filter, and will be difficult to discover with high probability in the test

We try to loop back signals from importance nodes in the circuit to spare pins in the microcontroller to do a better BIST, but that may in some cases result in a design that uses more components that just sticking with the ICT

Cheers

Klaus

If you have good assembly and AOI, unless you are building something life-critical, good functional test should be enough to get your shipping-level quality way up. I doubt that internal probing would help much.

We ship fairly complex gedgets that have multi-million hour field MTBFs, in some cases zero failures out of thousands shipped. RMAs are usually from customer abuse, like the corner of the box caved in.

We have one German customer that was using VGA cables to connect boxes, rather that the proper custom cables. That shorted all sorts of things and fried some power supplies.

We used to do a 24 hour burnin on every production unit, but it never found problems so we gave up on that.

Temperature testing first articles is pretty much necessary.

We functional test to make sure a box meets all guaranteed specs, and we try to design tests that will catch any assembly error or component failure. This is almost always automated. For example, if a D25 connector has six pins that are documented as ground, we maks sure all six are actually ground.

We use oscilloscope masks to snoop things like pulses, to check for overshoot or oscillation or other uglies. We prefer that no operator presence is required, so test+cal can run for hours if required, for full test coverage.