4

We periodically get some boards produced and assembled in small runs of 50 - 200. The PCBA company just makes and loads the boards (SMD reflow), we do the programming and testing after delivery

Sometimes there will be one or two boards with a problem, for example a missing resistor or a solder bridge.

We just received a batch of 200 boards, and in the 26 already tested, 8 have problems such as programming failed or sensors not working properly.

Questions:

  • What is a typical failure rate for small production runs?
  • Can we assume all the problems are soldering related? (Assuming components sourced in working condition)
  • What steps can we take to minimise the failure rate?
  • What response can we expect from the PCBA manufacturer? Is it reasonable to ask for a refund of assembly costs on the affected boards?

Update:

A particular QFN chip seems to be failing, see an example below. It seems some rework was attempted, which suggests an solder bridge was noticed during a visual inspection. (resistors are 0603 for size reference)

enter image description here

geometrikal
  • 4,921
  • 2
  • 23
  • 41
  • I would be looking for a new assembly house at that yield. – Matt Young Aug 18 '15 at 03:37
  • Have you actually diagnosed those 8 as specific assembly problems or are you guessing? If assembly problems, what are they? I would expect 99% yield even in those quantities for a well designed board and decent assembly house using good (not eBay) components. Getting back a few dollars for assy would not be my first concern. – Spehro Pefhany Aug 18 '15 at 03:42
  • @MattYoung this is the first time we have had a major problem with them. – geometrikal Aug 18 '15 at 03:54
  • @SpehroPefhany Previous runs have been pretty good, athough slightly different layout. It is just this latest one. Unfortunately I am not physically present to do the testing myself but all of the major components are sourced from digikey /mouser / element14 and the PCB is the same as the last run of 50 which had no problem. One of the boards where the sensor doesn't work shows some rework. Can we do something like get the PCBA company to load a test program that runs some diagnostics? Is that a typical thing to do? – geometrikal Aug 18 '15 at 04:01
  • Are they doing advanced optical inspection (AOI)? – pjc50 Aug 18 '15 at 08:54
  • 1
    Do you ask the fabricator and assembly house for their input to your layout for DFM and DFT? It is possible they will accept whatever you send, but if you involve them in the layout and testability process, you may find your yields go up significantly. – Peter Smith Aug 18 '15 at 09:33
  • The best person to do a root cause analysis on the failures is you. Figure out what the leading causes of failure are, and find out who's responsible for them. – Nick Johnson Aug 18 '15 at 10:52
  • @pjc50 I don't know what AOI exactly entails but they do visually inspect as evidenced by some rework on one of the boards. – geometrikal Aug 18 '15 at 12:14
  • @PeterSmith Good idea, there is one component in particular with very fine pitch that seems to have problems. Of the previous few runs of 50 and 100 they have only been 1 or 2 failed boards. I do supply them with complete eagle files to help with component placement. – geometrikal Aug 18 '15 at 12:16
  • @NickJohnson Yes of course, I'm trying to get to the bottom of it (I'm not at the lab so troubleshooting remotely) but I looks like the fine pitch 16-QFN might be the cause in some. – geometrikal Aug 18 '15 at 12:19
  • I ask because sometimes a design or layout error will be interpreted as a PCBA problem, especially in early days for that board. I have seen marginal designs get 20% fallout with ~perfect assembly. – Spehro Pefhany Aug 18 '15 at 14:09
  • @SpehroPefhany it could well be. I might post another question soon regarding layout for a particular part. Are there any common layout problems I should be aware of? – geometrikal Aug 18 '15 at 14:31

3 Answers3

4

1) Typical failure rates before test can be anywhere between 1% and 100% (I have never seen 0% on a batch of any size!)

2) Many of the problems will be solder related - short and open circuits - some may be incorrect components, incorrect orientation, actual component failures, and so on. Failure to program may be a consequence of another failure, etc.

3) This is the big one...

4) If there was nothing specific about quality in the contract, there may be nothing you can expect from the manufacturer. Some level of failure pre-test is to be expected, correcting that is the purpose of test. If you can identify systemic defects - e.g. most boards have solder bridges between adjacent pins on U1 or J3 - you can negotiate to have the lot returned and reworked so they are (visually) good. And specify visual inspection and correction of these areas next time.

But certainly report, and ask if there is any way to improve. They may give your next job to Pamela (she's good!) or they may quote for visual inspection, or refer you to their board test department for a quote...

And back to (3). As Michael Karas says, you can fully specify a test procedure and make that part of the assembly contract. That may cost : per unit, in test time, test equipment hire, etc : but also up-front tooling costs for test fixtures and test programming. That can be expensive : a bed of nails test fixture can be several thousand dollars, and a test program may take a few months (for a complex board).

Alternatively, if the assembler is already performing device programming, they may be willing to program a testsuite in, run that to do a basic sanity check on the hardware, then reprogram with the application.

It may also restrict your choice in assembly houses to those willing and skilled enough to do this.

Or you can design for test, in particular, make sure every signal on the PCB is accessible either via a connector, or a test point (such as an easily accessible pad on the underside of the PCB). Then you can build and program the test setup yourself. This may make more sense - especially if you are prepared to rework the defects rather than outsourcing that.

A few more points :

  • if this is the first batch from a new layout, you may find something different : perhaps all the soldering errors are clustered around one IC or connector. That's quite likely to be a design problem : suck up the rework for this batch, but look around for improvements.

The assembler themselves are the best party to talk to - maybe they recommend a different footprint for that IC than the one you're using. They can live with yours but it's not ideal and some change like longer pads to wick off excess solder, or solder resist overlapping the pads, or some variation gives better results with their process.

  • For ground/power shorts tests on a 100 pin connector on one board, I wired up a bank of switches, 4 pins to a switch, to the mating connector. (The other end of every switch was commoned). If any pin was shorted to power or ground, 2 buzz tests told me (common to ground, common to V+). Then I'd open switches until the buzz stopped, and I only had 4 pins to buzz out by hand. Well worth the time spent making it for 100 boards...

  • Sometimes, blowing under the chip with a hot air pencil can clear these shorts.

Community
  • 1
  • Thanks Brian, test points are a good idea. The board is actually quite simple so a simple test for shorts on one particular component could be done quickly. – geometrikal Aug 18 '15 at 12:24
2

You really cannot go back to the assembly house to complain if they met the build agreement which was to simply assemble blank boards with no testing.

If you have greater expectations than that then you need to spell out all the specific details in the assembly agreement/contract. And then along with that you need to support your expectations so that the manufacturer can be properly enabled to execute according to the contract. This support may include:

  1. Programming data files
  2. Programming procedures
  3. Programming pods and supporting software
  4. Test procedure including pass/fail criteria
  5. Known good reference boards
  6. Training on to perform programming and testing
  7. Specification of expected yield rates and acceptable rework levels
Michael Karas
  • 56,889
  • 3
  • 70
  • 138
  • Is board testing on their end common for small runs? I will let them know so they can improve for next time. – geometrikal Aug 18 '15 at 05:45
  • It all depends on what the different contract manufacturers offer. Some may offer testing as an chargeable extra, some always include it (and may be more expensive in the first place) and others may not be able to offer it at all. – og1L Aug 18 '15 at 10:09
2

To try and answer your questions:

1) The yield will depend on the contract manufacturer and the technology used on your design. I think to expect 99% yields for any small number production runs (especially if it is a new design) is a bit optimistic but I would expect it to be 90%+.

2) There are a couple of possible problems: Pick'n'place-related (wrong components or wrong orientation), soldering-related (shorts, opens...), handling-related (ESD-issues, damage during transport...).

3) As mentioned by someone else: Work with the CMs during the design phase. Certain device packages will be more problematic during production and the CM should be able to advise. Often a certain extra-small package isn't needed and replacing it with an "easier" package may improve yields.

The other thing you can do is to specify some testing as part of the PCB assembling. I would at least expect some AOI-testing - maybe more depending on the value of those PCBs.

4) That really depends on what you agreed and what the problems turn out to be. If you have agreed a minimum yield then you are obviously in a stronger position for any discussions. I would first talk to the CM before demanding any money back. They should be interested in resolving those issues and keeping you as a customer happy.

og1L
  • 461
  • 3
  • 5
  • Thanks og1L, I will pass 2) for my colleagues to check. We didn't agree to anything just manufacture, I didn't even realise that could be done. :) – geometrikal Aug 18 '15 at 12:29