Bunch of Shorted MLCC Caps in R+S Analyser

[knobtwiddler]

Hi,

I have an R+S audio analyser that has at least 3 MLCC caps on the analogue side that have gone low-R (typically 5-10R - managing to blow a pair of VRegs on the secondary of the transformer). 1 of them is in the PSU (1u) and 2 of the others (22uF) are Rail-0V on the analogue input board.

The R+S has 2 power supplies: An SMPS for digital duties, and a linear for the analogue board. I haven't investigated the digital side yet, so have no idea if any caps there have suffered the same fate...

I am wondering what might have caused 3 caps (counted so far...) to fail?

We know that heat and mechanical stress causes SMT MLCC caps to fail. But R+S do not skimp on mechanical bracing, and the machine is 17 years old.

Over-voltage? But how would that happen, if each rail is fed from a linear Vreg?

This machine was bought broken, and I'm having a go at seeing what I can fix before throwing in the towel and having open wallet surgery at R+S. But I am wondering what could cause a whole bunch of ceramic caps to fail? Bad batch? (2 different values have failed...).

If I got an idea of what might have subjected the machine to do this, it'll give an idea of what else needs careful checking. Answers on a postcard, please!

TIA

[Radio Wrangler]

Time

Time and defects.

A small proportion of MLCCs are shorted out of the box.

After that, the failure rate is significant.

David

[Keith956]

Could an input have been overloaded? Sometimes input protection clamps to supplies which can then damage all sorts of things.

Always consider what you'd least expect - could the caps have gone because the voltage regs / other circuitry overloaded them?

[duncanlowe]

Strain, caused by flexing of the board is a common failure point. But actually too much support or in the wrong place can be a cause rather than a solution. I'm aware of at least one module that had a number of MLCCs filtering the various main supplies. One of these was mounted on the other side of the board to a support leg and it turned out this concentrated the problem and a number of the units got cracked caps. These don't manifest straight away. It takes time for the metal to grow through the cracks, with voltage on. It won't grow unless the voltage is present. In this application it was quite serious as the available current was quite capable of blowing the capacitor apart and it was permanently powered.

[cmjones01]

Yes, MLCCs do that sometimes. Not as often as Tantalums by any means, but they do go short-circuit. Mechanical stress (including that incurred during manufacturing or rework) makes it worse.

Chris

[Radio Wrangler]

Thermal shock, wrong cooling profile, all sorts of things can exacerbate the problem. There are also 'soft termination' types for automotive use which have some resilience against board flexing. Worth considering for all ceramics in 1206 or larger sizes.

Even properly handled, expect some duds right off the film strip packaging.

David

[duncanlowe]

Process in the wider term is critical. One customer of a company I worked for insisted on passing PCBs through the process that had been strain gauged close to the most vulnerable caps (they decided which ones) before approval. The production site thought we were nuts. Anyhow, we ran the process and the strain was OK, just. So we went to volume production. Just a few weeks in, as a result of late firmware changes, the product was undergoing rework by reflashing the firmware. Many of the reworked products were failing EOL test, essentially dead. I got them to send me a few samples for analysis. On every sample I tested, a particular MLCC on the output of the VReg had gone short, so no disaster on the input with exploding MLCCs, but the VReg current limiting so there was no voltage to the uC.

Guess what. They had changed the process since the strain test! Less trimming on the PCB edge meant they were having to force the PCB into the jig for assembly onto the motor that it controlled, bending it in the process. They thought I was some kind of magician for working it out so fast, but in fact it was the customers previous experience of the problem and therefore requirements that made it relatively easy to figure out!

[knobtwiddler]

At the firm I am involved with, we had lots of R-R MLCCs go short a few years back. The product is still in manufacture, revised for through-hole ceramics. Initially, I thought the problem was due to mechanical stress. However, we also found caps had gone short on small (FR4 / 4-layer / 2oz) daughter cards, which are mounted via chunky brackets at 90-degrees, so I can't see how these cards could've flexed. We put it down to thermal damage...

How on earth do OEMs that make volume electronics such as mobile phones account for the variability (i.e. failing off the reel) in reliability? Do they have jigs that test them as they spool onto the P+P line?

Thank you all. This has been an enlightening experience. I am now scratching my head, wondering how many of the R+S MLCCs should be replaced... I have 2 other identical R+S machines (similar age), and - touch wood - they seem ok... I hope they stay that way, as they are a workhorses that are powered on all day.

[cmjones01]

Originally Posted by knobtwiddler

How on earth do OEMs that make volume electronics such as mobile phones account for the variability (i.e. failing off the reel) in reliability? Do they have jigs that test them as they spool onto the P+P line?

In my experience relatively few MLCCs are faulty off the reel. Later failures can be managed simply by very careful process control during board assembly and test. The best PCB subcontractors I've used are obsessive about how parts and boards are handled and stored and the temperature profiles used during soldering, and they make sure that everything's as repeatable as possible. It pays off in yield and reliability.

Curiously, I've found that having everything absolutely right brings the cost down, too, because there's far less wasted time, material and uncertainty. In one memorable case I found that this super-careful approach brought the cost of a board packed with BGAs with hundreds of pins and, yes, gazillions of little ceramic capacitors, down to half the cost it had been at another subcontractor who, shall we say, had a slightly more relaxed approach. Both firms were in the UK.

Chris

[knobtwiddler]

From reading the thread it would appear that MLCC caps need careful handling and design implementation if you want them to be reliable. In my case, TH MLCCs have been reliable, whereas SMT ones have definitely not (in designs used at my firm, as well as on the R+S analyser per subject title).

Would it be fair to say that TH MLCCs are vastly easier to implement and handle than their SMT brethren? My experience would suggest this to be very much the case. Does yours?

[duncanlowe]

There are solutions for MLCCs if there's a possibility they could be strained, is one of the resistant types, such as the flexible terminal / soft terminal types. Of course, they cost more.

[Aub]

What's MLCC stand for?

Thanks
Aub

[Cobaltblue]

Multi layer ceramic capacitor

[Aub]

Thanks for that. Single layer ones are enough for me 🙂

Aub

[knobtwiddler]

Originally Posted by duncanlowe

There are solutions for MLCCs if there's a possibility they could be strained, is one of the resistant types, such as the flexible terminal / soft terminal types. Of course, they cost more.

Hi Duncan,

In my experience using leaded MLCCs over several decades, they've been highly reliable - most likely because the leads take the strain. However, in the last few years, I have seen many SMT MLCCs fail, in a variety of places, not just at the firm I work at.

I am quite spooked at the prospect of using SMT MLCCs, which is a problem, as leaded caps are becoming scarce and can only become rarer in years to come.

NB - we ought to be booting up the R+S soon... Fingers crossed! Either it will be a happy day, or a frightening bill from R+S as an alternative (it may be that other issues exist in the machine, not just the caps that failed short).

Thanks,

[duncanlowe]

SMT MLCCs are extremely widely used, and in general aren't a problem. But they can be. It's tempting to use them on DC power supplies because they are small for a given capacitance. But DC and cracking (caused by strain) is a disaster waiting to happen.

Leaded components aren't really an option for mass production these days. As someone else said earlier, process control is key.

[G0HZU_JMR]

I've seen this issue a lot at work over the years. Usually the worst offenders are MLCCs in the long and thin 1206 package.

It's best to avoid fitting an MLCC dc blocking cap near an RF connector like an SMA PCB end launch connector. Even careful handling of a PCB when fitting a cable to the SMA connector can result in a cracked MLCC in this scenario.

There is also a great deal of difference in immunity to this problem. About 12 years ago I designed a high power broadband RF switch using PIN diodes and I used 1nF 1206 200V caps in the prototype on the high voltage control lines. These caps were especially thin (as in not very tall) and it was basically impossible to handle the PCB without cracking several of the caps at a time and these weren't near any connectors. By contrast I have some old MCU boards here that have Murata 100nF 1206 caps on them and these 1206 caps appear to be indestructible in terms of cracking even if abused with a soldering iron numerous times. I use these old MCU boards as a source of reliable 100nF 1206 caps for prototype boards here at home.

[G0HZU_JMR]

Some MLCC failures also take a long time to show symptoms. If there is a small crack then over time moisture gets in and eventually something will break down due to the long term presence of the moisture as it seeps into the crack.

Things get much worse if you try and prototype using a thin PCB material. The first thing I do if I mill an RF board from 0.02" (0.5mm) thick material is to add strengthening on the underside to minimise PCB flexing. Without this the board would soon develop symptoms of broken MLCCs. This can manifest itself as a subtle intermittent change in signal strength when the board is touched or it can lead to complete loss of signal.

[knobtwiddler]

Some MLCC failures also take a long time to show symptoms.

Thank you, Jeremy. This would be my experience also. I had a bunch on some cards that were mounted vertically, by a solid alloy bushing. The cards were small (about 60mm x 40mm) and 1.6mm thick. I can't see how they sustained flexing in use (the assembly would need a very heavy whack), so can only conclude that they either had defects from manufacture that manifested themselves retrospectively; or defects due to handling. They looked like little peppercorns... The client didn't report a bang, thankfully. It was interesting to see how most of them went short and triggered the PSU protection, whereas others went pop, and the unit continued to work with charred R-R caps (50v - across 34VDC R-R).