When development boards fail, is it the fixture or the finish?

Herewith is yet another striking reminder that in marriage as in life, communication holds the key to better living. And going green is not always as easy, or as virtuous, as it seems.

The Players:

OEM: Full of really very incredibly smart people developing state-of-the-art technology within a highly competitive and accelerating innovation environment. Did I mention they were chock full of really smart people?

CM: Builds boards for OEM.

Test Service (TS): Hired by OEM to provide ICT fixture and program design. At OEM’s direction, TS will deliver said fixtures and programs to be run on CM’s equipment. Perceived as a non-value-added service by some players in the development process. This perception that they have little of relevance to contribute leads to minimal involvement in the feedback loops addressing various critical process problems.

The OEM was happily engaged with having the CM build its boards. The bare boards in this process were manufactured using the ENIG (electroless nickel immersion gold) process. All was right in the testing world, as far as this process and surface finish were concerned. Contact was great; repeatability was great; throughput was great, and, most important, assembly reliability was great. With this past as prelude, the OEM tasked the TS with building test fixtures and test programs for the next generation of products. TS saluted, and off they went to fulfill this task, designing leading-edge ICT fixtures capable of reliably and repeatably hitting 20 mil targets.

Smack dab in the middle of the test development process, the OEM was told by one of their Big Name Customers to change their assembly process to one using no-clean flux and OSP (organic solderability protectant) board finish. The reason: ENIG was insufficiently green. Big Name Customer was all about being green. Green, of course, being indisputably good.

Except that the OEM neglected to inform its TS of this change. TS beavered diligently along under the assumption that the surface finish remained ENIG. The fixture was manufactured with the expectation of contacting flat ENIG pads. Test development proceeded oblivious to the change; in fact, initial testing went smoothly with ENIG-coated prototype boards, and all remained well with the world. Soothing music continued playing.

Chaos! Trouble came fast with the first production build of 20 boards. CM delivered to the TS to test and validate the fixture and program. The test development engineer at TS placed the first board in the fixture, and the program blew its mind. Outputted was a long list of opens, suggesting contact problems with the fixture. An inauspicious start to a verification process to be sure. A second board was placed in the fixture, with similar (bad) results. The same results with a third board, followed by a fourth board. Utter and complete failure. Contact problems were now being strongly suggested. Something was seriously wrong with this fixture.

Time for forensics. A painstaking, step-by-step debug of each and every failure on the first board is taken. Each contact issue is examined; the amount of flux and residue on the board is noted. Could this be a no-clean issue? Each problematic location on one board is manually cleaned. Following cleaning, the board is placed in the test fixture and tested again. Improved results are achieved! On the trail. A second manual cleaning cycle is undertaken for the sake of thoroughness. The test program runs flawlessly after this additional cleaning step. As a control, board number two is cleaned in the same manner, with the same results. And so on with board numbers three, four, and the balance of the 20 piece lot, cleaned in collaboration with CM. All 20 boards are run through the identical cleaning cycle and returned for test. All 20 boards run successfully through the test, with virtually no contact problems. Success. Solution found.

Or so TS thought. Unfortunately those same green rules driving the change in board surface finish to OSP also drove the cleaning process or, more accurately, forbade it. The cleaning process employed was not allowed in volume production. The SMT process at CM needed to permit the boards with the no-clean flux residue to be tested.

Meanwhile, the test development team at TS subjected the 20 initial production boards, as well as the development boards, to additional scrutiny. They took note of the fact that the development boards with ENIG finish exhibited test probe marks in the dead center of their gold test pads. By contrast, the production boards with OSP finish exhibited solder balls in those same test pad locations with probe marks sliding down the side of the solder balls. The probe marks terminated at the base of each solder ball, where flux/no-clean residue had accumulated.

More forensics. Suspicion now pointed to the probability that cleaning the boards was a necessary but not sufficient condition to solve the problem, more a case of addressing a symptom but not the virus at the root of the disease. Repeated observations of test pad impact suggested that the test probes, notably the 0.050" and 0.039" probes, were striking the solder ball and glancing away at full test actuation and extension. There was little or no spring force being applied at full extension. Consequently, solder balls on test pads with high amounts of solder presented an insufficiently flat or planar central surface to retain the probe in intimate contact. As noted, the probe would deflect from the solder ball into the no-clean flux at its base, thereby registering as an open. No-clean flux presented a soft target for a sharp, blade-style probe head, resulting in the probe becoming stuck in the flux. Cleaning this flux, as TS and CM did for the first 20 production boards, permitted the probe to slide back into partial contact with the solder ball, thus masking the root cause of the failure.

From forensics to engineering. Several modifications needed to be made to the test fixture. First, probe guiding needed improvement to enable test probes to strike the center of each test pad and reduce – if not eliminate – their natural tendency to deflect from the dome of each solder ball. The fixture as originally designed contained many bottom-side vectorless test sensors. This meant that the bottom plate of the fixture contained numerous large open spaces, thus allowing no guiding of probes in those regions. This problem was remedied by rebuilding the bottom plate, the major improvements being the removal of many vectorless test sensors from the bottom plate and closing or sealing the previously relieved areas. The “zero-flex” relief was removed and board supports added so the probe guide hole length was maximized. Further, if the solder pads with OSP finish were not presenting a flat surface to the probes, then the probes needed to present a flat surface to the solder pad. The head style of the probe had to be more of a blunt object, rather than the sharp knife originally installed. A small, serrated style probe head was selected as the replacement. Last, guide holes were enlarged on the bottom plate to handle the larger serrated head style.

Once the modifications previously described were made to the bottom plate, boards that had not been through any cleaning process were tested. The results were very much improved. Most boards would run successfully without any need for retest or manipulation. The operative term is “most.” As with any OSP/no-clean process, some lingering contact issues and false failures remained, but far fewer than before. (They cannot be 100% eliminated when using fine-pitch spring probes.)
Moral of the story: The adverse side effects of the OSP/no-clean process can be significantly ameliorated with some preparation before the test fixtures are built. The key to those preparations is to keep your test service, whether internal or external, fully informed of any process change. They, like the government, are here to help, but at a lower cost.

Robert Boguski is president of Datest Corp., (datest.com); rboguski@datest.com. His column runs bimonthly.