Estimated reading time: 6 minutes

Testing Todd: What Do You Mean 'Passed' Isn't Enough?

It’s Friday afternoon, and the shipping deadline is approaching rapidly. Your high-visibility Class 3, Level C product is just about done with electrical test and should fly through FA and barely make it. You are relieved and look forward to Saturday on the lake. Just to make sure, you call the ET department and check on the yield. The lead in ET replies, “96% final.” You are quite happy with the news as you manufactured enough overage to compensate for that loss of 4%. Relieved, you head out the door pleased that the new customer delivery has been met.

Not so fast! Your cellphone rings just as you press the disarm button on your getaway car, and your quality manager is on the other end. They say, “We have a problem.” You happily retort, “We had a 96% yield! They passed ET!” But your quality manager replies, “You better come to the lab,” and you have the same feeling as when the dealer flips a blackjack just when you’ve doubled-down. Their final comment before hanging up is, “We better not ship this order.” Feeling your boat sink before you even get it off the trailer, you reluctantly head to the lab.

So, what happened? Although you had a 96% yield, it has been discovered that, within that 4% of failures, is a defect that puts the entire order at risk. In this case, I’m talking about barrel voids. ET had detected barrel voids in the 4% of failures. The decision now must be made as to whether the “passed” product can ship as reliable or should the order hold for evaluation. Looking from the outside, the answer is fairly clear that it should hold to evaluate the type of void detected and whether it poses a potential field failure once it leaves the manufacturing facility. This decision, however, is critical, and honestly, there are many times the order doesn’t go on hold and ships instead.

How reliable is this order in the long run without verification? It’s a risk that, unfortunately, is taken due to production pressure and high revenue deliverables. Sure, in most cases, there may be no negative results from this decision, but it only takes one failure at assembly to cause a line stoppage and a costly return to the manufacturer. This also results in the OEM and/or CM possibly questioning the reliability of the manufacturer, and in today’s market, reputation is key!

From a reliability standpoint, we need to quickly assess what risk we may have uncovered when faults are detected during electrical test. What are the showstoppers, and what statistically shows a low latent risk at the assembler and beyond? In most cases, isolated inner layer defects, random shorts, or surface solder tails are not statistically significant on the overall long-term reliability of the product. These are usually either reworked (if allowed) or scrapped. Many OEMs no longer even allow repairs on their product, so the latent risk is fully removed as the board is scrapped.

The significant defect that requires much more scrutiny is the void. This can be even more important when microvias and blind and/or buried vias are involved. These are the defects that may hide in a “passed” board only to manifest during assembly due to thermal stress and high temperature during solder flow. Once the void is identified, it is crucial to identify its type and what substructures may be involved (in the case of sub-part stack lamination). If the void is determined to be a bubble or air entrapment type circumferential void, it could be an isolated case. This may be isolated to a specific flight bar on a plating line, and some sampling of other board serial numbers processed on that flight bar may be indicated.

However, if a taper plate or thin copper void is determined, it may indicate a wide range of risk on the entire load that was processed. Failed bonding on a microvia can also indicate an undetermined plating issue or even an anomaly in sub-part lamination. Now, the risk on the overall long-term reliability of the product has become very high. Shipping product when defects like these are found, even in a small percentage, blows the statistical curve of ET yield. A 96% yield does not bode well when that 96% may be hiding 100% of potential field failures.

We must determine what risk is present, which requires high-resolution resistance testing of the barrels or, alternately, 4-wire Kelvin testing. In most cases, this is done on the smallest holes as the higher the aspect ratio, the greater the risk. Studies have been done to calculate the theoretical resistance that can be expected based on the copper weight and aspect ratio. With this test, you can determine if you have a risk within the remainder of the product that “passed” electrical test. We must remember that standard electrical test is measuring continuity resistance at 10 ohms and above even for Class 3, Level C product. Taper plate and thin copper will not be detected under these test conditions. The fluctuations in resistance of a good barrel versus a suspect barrel will be in the milli-ohm range. This is undetectable with standard ET, which is why we can have a 96% yield in ET with a hidden train wreck just waiting to happen.

The problem arises here that now we have an entire order that must be screened but the direct access to the barrels is not available due to solder mask and/or via plug. The best results in 4-wire Kelvin testing is the direct probing of the opposing sides of the barrel. Probing at the first opportunity from the barrel introduces more copper and thus increases the mean resistance of that given barrel. What happens is that the resistance master developed from that longer circuit now becomes too large to accurately detect the small changes in resistivity that thin copper or taper plate may cause.

Remember, the test is looking for milli-ohm changes in resistivity from a good barrel to trigger a fault. The detection guard percentage is adjustable and typically set to around 25%. If you have excess copper in the circuit, the total resistance end-to-end is 1 ohm, and the fault trigger is 25%, you would have to see a 250+ milli-ohm change in that circuit to trigger a fault. That is far too high to detect the type of fault in question.

The main solution is pre-planning with these types of product. Small hole size, high aspect ratio product requires in-process screening. Trying to perform 4-wire Kelvin test on fully masked and finished product will not identify the potential latent defect unless the barrels are accessible from both sides. This test should be performed before solder mask and after all plating processes are complete. This allows the direct probing of the high aspect ratio barrels, which will deliver the most accurate results. This can also be a sampling from each flight bar from plating to identify if there was a potential systemic issue across the entire load or just perhaps an issue with just one flight bar alone. Performing the test at this stage increases your confidence in reliability as resistance fluctuations will be detected before costly final processes are performed. If the test is fatal and caught early enough, a restart can be performed with as minimal an impact as possible on delivery.

What we have seen today is that “passed” is not always passed. We must be diligent to scrutinize the failures found during routine electrical test as a high yield in ET may not indicate high reliability. Improperly reviewing the failures, and especially overlooking the potential impact of a detected void, can turn a 96% ET yield into a 0% yield in the field. This may result in a devastating monetary hit to the manufacturer not to mention the reputation hit in this extremely competitive market.

Todd Kolmodin is VP of quality for Gardien Services USA and an expert in electrical test and reliability issues.