Faulty name-brand ceramic capacitors - common?

[Corporate666]

I have experienced a 500% increase in failures in the field over the last few months for one specific product.  I've gotten some of the units back and it was just this week I was able to go through all the failed units, do a thorough post-mortem and tabulate the results.

It turns out a capacitor on the power input was failing as a dead short, causing the traces on the board leading to that cap to ultimately massively overheat and go up in smoke. 

The capacitors are CL21A226MOCLRNC from Samsung.  I went through our stack of boards that failed post-assembly testing and were awaiting rework, and found that most of the failures there are also due to this same capacitor.  I then went through the drop-bin on the pick and place machine (where it drops parts it could not place for whatever reason), and pulled out all of the aforementioned caps and tested them... about 5% were bad - just a dead short.  I looked under a microscope and there's nothing unusual to see, no cracks or problems with the solder points or anything else.

But the fact that I saw so many failed parts in the PnP reject bin tells me that it wasn't the reflow process, nor the testing process, nor the use in the field that was causing the issues... it was bad parts right off the reel.


I was wondering if anyone has seen anything like this before?  Digikey has ~200,000 of these caps in stock, so it seems to be a pretty common usage part, and I am very surprised that a major manufacturer would be shipping parts with this problem.  But maybe I have just been lucky in the past?

[georges80]

It's an MLCC - so likely internally shorted and likely due to an assembly issue or PCB layout issue (i.e. micro cracks leading to internal short).

I'd find it hard to believe they are faulty right out of the chute... I'd pull a bunch of parts straight off the reel and measure/test before any pnp machine etc has gotten to them.

IF you find them faulty on the reel then I would contact Digikey to figure out what is going on.

cheers,
george.

[GreggD]

Could someone be bending the pcb and causing micro cracks ?
Is there a heavy component on the pcb so that if the pcb is picked up by one end the heavy part flexes the board.
This is quite common and causes a dead or near dead short.

[Leo Bodnar]

I have just ordered a few hundred from DK.  Maybe I get lucky too - will let you know.

I get ceramic caps failing shortcircuit now and then but they are usually already assembled and nowhere close to even 1%.

Leo

[wraper]

In my experience MLCCs fail because of mechanical stress. PCB may have 10s of the same ceramic capacitors on the single power rail, and it's always same single cap failing somewhere nearby to the mounting hole or board edge. Larger sizes are more susceptible.
You won't see something like this unless cut capacitor in half.

[wraper]

But the fact that I saw so many failed parts in the PnP reject bin tells me that it wasn't the reflow process, nor the testing process, nor the use in the field that was causing the issues... it was bad parts right off the reel.

I rather suspect PnP machine was smashing them while picking up or something like this. Are there multiple such capacitors on the PCB and only particular destinator failing?

[Leo Bodnar]

I have seen capacitors with visible cracks and chips right off the [undamaged] tape when I have been assembling prototypes manually. 
They were  1206 with high C / high V value for that package so not a cheapo 0.1uF runarounds.  Can't remember exact brand but definitely well branded - TDK, Murata or Kemet.

Anyone who worked in manufacturing will know that weird things do happen and when they do happen there are usually a lot of them.

Leo

[Corporate666]

Well, a plot twist... not all of the failed boards were PnP placed.  The board doesn't have too many parts on it, so if the PnP is already set up for another job and we just need a run of 20 boards or such, we will hand-place them.

Due to our manufacturing process, I know which boards were hand placed and which were PnP placed, and a proportional quantity of the boards with failed caps were hand placed.  I also went through the tub of components that they use when hand placing boards and tested a bunch and found several that were failed with a dead short also. 

Nothing in the PCB or assembly process has changed - and based on the batch numbers and estimating how many parts are taken off the reel for hand assembly, not only were all the bad parts from a single reel, but it looks like it was just part of the reel that was bad.  In other words, it appears that maybe just a couple of meters of that reel had a ton of bad parts and the rest were fine.

I wonder if something could have happened during reeling or during storage/transit to cause this?  Like the reel got squished by something or someone dropped something on it in the shop? 

[Leo Bodnar]

I wonder if something could have happened during reeling or during storage/transit to cause this?  Like the reel got squished by something or someone dropped something on it in the shop?

Do you use DK re-reeling service or buy full reels?

Leo

[TimNJ]

Ceramic caps, especially those with class 2 dielectrics (X7R), are prone to endcap separation due to mechanical stress. Caps placed near the edge of a board or near heavy components are more likely to fail. Also, an incorrect soldering temperature profile, or even worse, hand soldering, can cause the endcaps and ceramic chip to expand at different rates and crack. NP0/C0G caps don't experience these problems as much.

You might want to take a look at MLCCs with "flexible terminations" like FT-CAP from Kemet: http://www.kemet.com/Lists/ApplicationBriefs/Attachments/58/Flexible%20Termination%20Technology%20(FT-CAP).pdf

These caps have a layer of a epoxy (of some sort) that make them more resilient to mechanical stress.

[Leo Bodnar]

I know this is internet but could we please concentrate on the fact that a lot of failed caps have never been on the PCB in their entire life?

[rx8pilot]

Following this thread.....

Sent from my SM-G892A using Tapatalk

[thm_w]

5% is incredibly high, did you contact Samsung with the production date to see if they can provide any insight?

I have just ordered a few hundred from DK.  Maybe I get lucky too - will let you know.

I get ceramic caps failing shortcircuit now and then but they are usually already assembled and nowhere close to even 1%.

Yes, same experience here, a few ceramics failed after production (<0.1%), and a handful failed returned from the field (<0.2%). Its always the same ones in the power section, so maybe they are stressed or due to their larger size (1210 10uF 25V X7R Murata).
In the field case they could have been abused (overvoltage, roughly handled in shipping, etc.), so cannot say for sure if they were actually defective.

[wraper]

Yes, same experience here, a few ceramics failed after production (<0.1%), and a handful failed returned from the field (<0.2%). Its always the same ones in the power section, so maybe they are stressed or due to their larger size (1210 10uF 25V X7R Murata).

It's not "OR" situation. Larger sizes are more susceptible to mechanical stress. I would look into how caps are mechanically located and how these boards are depanelized. You could, say, rotate them by 90^o and this could fix the issue.

In the field case they could have been abused (overvoltage, roughly handled in shipping, etc.), so cannot say for sure if they were actually defective.

If there is mechanical stress at production, they may fail short later. Overvoltage unlikely to be the issue, MLCC easily survive 10x of the rated voltage.

[Psi]

I've seen some 10uF Yageo 0603 caps fail semi-shorted recently. But being cheap Yaego caps it didn't surprise me that much.

[SeanB]

Position dependant on the reel only, so my guess is that the reel was dropped at some point on edge on a hard surface. Check the reel itself for edge damage, and you will find a flat spot where it deformed. Likely the reel tape inside was not tightly wrapped, and the loose turns had enough shock from hitting each other to crack the capacitors as they impacted inside the reel tubs and crack them.

Take the reel, and sacrifice a few meters of capacitors, peeling the tape off after attaching the reel to a hard flat surface, and look inside the plastic tubs to see the indents the capacitors would have left there on impact. Use a microscope to see, and take a photo of dented versus undented, and have a word with the supplier and the shipper, for both dropping shock sensitive items and for not packing properly in a box with adequate shock dampening material.

[Corporate666]

I wonder if something could have happened during reeling or during storage/transit to cause this?  Like the reel got squished by something or someone dropped something on it in the shop?

Do you use DK re-reeling service or buy full reels?

Leo

Bought as a full reel, new in package.

[Corporate666]

5% is incredibly high, did you contact Samsung with the production date to see if they can provide any insight?

I haven't, but I will probably do that this week.  I don't imagine much of a result... especially since the reel has since been used up and the only parts I have are the failed ones, those in the hand-placement tub, and the miplaced parts catcher from the PnP.  At best they will ask for some of the parts back for analysis and maybe will send me some replacements.  We'll see. 

[T3sl4co1l]

FWIW, 1210 is large, but even the conservative IPC doesn't worry about them: in fact, they recommend 1210 as the largest reliable size.

In any case, forget about soldering defects, parts were bad off the reel.  Can't people read?

Second the suggestion of calling Samsung and notifying their quality dept somehow.  Or if not them directly, then the supplier -- for that matter, I suppose you might ask if you can get a refund/credit or replacement for the bum parts.  Then if they decide to notify the mfg, it's up to them (and they probably have better contacts than you, anyway?).

Tim

[Leo Bodnar]

Bought as a full reel, new in package.

Corporate666, I have just received DK order.

There is good news and bad news.

Good news is there are no shorts.

Bad news is capacitance is about half of what you'd expect.  They are all in the 13-14µF range.

Leo

https://www.digikey.co.uk/products/en?keywords=1276-6780-1-ND
1276-6780-1-ND   .07430   7.43 T CAP CER 22UF 16V X5R 0805
HTSUS: 8532.24.0020   ECCN: EAR99
LEAD: LEAD FREE   ROHS: ROHS
Mercury: Cert on File.
For more information contact RoHS@DigiKey.com
COUNTRY/ORIGIN:CHINA

[jbb]

Half the expected value is odd, unless it's due to DC bias. Do you know what voltage is applied by your capacitor meter?

X5R and similar dielectrics actually loose a lot of capacitance when DC voltage is applied. Some manufacturers are more forthcoming than others about this. There should be a capacitance vs DC bias curve in the data sheet or part technical information to help with your design.

[Leo Bodnar]

Half the expected value is odd, unless it's due to DC bias. Do you know what voltage is applied by your capacitor meter?

I have three meters from various brands and they all show the same value. Don't really have time to look it up but i'd be amazed if there is any DC bias beyond few mV.  I trust Agilent not to be complete idiots.

Leo

[BFX]

@Leo
Capacitors >10uF should be measured at 120Hz +-20%.
I see on your meter 1kHz. That could be problem.

[Ice-Tea]

I noticed that you never outright say whether you bought them at DK or not... Did you?

Also, I agree: get in touch with Samsung. Failing components off the reel should be in the low ppm region. Or, you know, nothing at all.

[wraper]

@Leo
Capacitors >10uF should be measured at 120Hz +-20%.
I see on your meter 1kHz. That could be problem.

This, also a significant part of capacitance drops because of ageing. Leo, reheat the capacitor with the hot air or soldering iron, let it cool down for some time and measure again.

[Leo Bodnar]

@Leo
Capacitors >10uF should be measured at 120Hz +-20%.
I see on your meter 1kHz. That could be problem.

Not that much different - still 25% down from nominal.

Leo

[wraper]

Not that much different - still 25% down from nominal.

Leo

This is completely normal. Do as I said above and capacitance will be within rated tolerance.

[Leo Bodnar]

After Pb-free reflow cycle it went up to 19.8µF.  I'll let it pass.
Leo

[thm_w]

This, also a significant part of capacitance drops because of ageing. Leo, reheat the capacitor with the hot air or soldering iron, let it cool down for some time and measure again.

Had no idea it was this significant (10% after 1000hr on X7R): http://www.johansondielectrics.com/ceramic-capacitor-aging-made-simple

As you said:

After the soldering process the capacitors have essentially been De-Aged. Capacitance measurements may be erratic in the initial 10 hours after testing. This is due to the initial capacitance value, dielectric type and the time between reflow and the capacitance measurement. For this reason it may be necessary to wait for the capacitance to stabilize after reflow before testing. In “High K” dielectrics the capacitance may also appear slightly high after the soldering process. This is normal as the capacitance is intended to be stable after 1000 hours so that there is adequate capacitance throughout the life of the circuit

Another reason to not touch Y5V..

[Leo Bodnar]

Another reason to not touch Y5V..

Also partial reason for why you can seemingly fix some hardware by magically "reflowing" it in kitchen oven.
Leo

[Siwastaja]

Another reason to not touch Y5V..

Even X7R can be quite horrible, some "poor" X7Rs are as bad as some other "good" Y5Vs. You always need to check the actual curves of the exact component (if available). Rule of thumb: a smaller ("too-good-to-be-true") device usually has worse DC bias, ageing and temperature characteristics, all at once.

So simply avoiding Y5V might not be enough. A lot of homework to do when designing .

(Edit: sorry for OT. The issue here is the case of broken components on the reel / taken from the reel, not soldered.)

[wraper]

Another reason to not touch Y5V..

Also partial reason for why you can seemingly fix some hardware by magically "reflowing" it in kitchen oven.
Leo

It's not like this actually unless device is poorly designed. When heating shorted/leaky MLCCs, they often start working normally for some time. Therefore trying to find shorted component on the rail with a lot of MLCCs on it may be a big PITA. Say you desolder some MLCC or MOSFET and it becomes fine. And then you don't know which part was actually faulty. Lois Rossmann was talking about this issue on his youtube channel as well, when explaining GPU reflow nonsense performed by con artists or clueless. Particular models of Macbook may have shorted MLCC under GPU and when "reflowed" it starts working again for some time.
If I suspect shorted MLCC, I apply voltage to the rail from lab PSU and look with thermal imager as basically this is the only easy way to know for sure that short is in particular MLCC.
This could also explain why OP has many returns. Shorted caps might become working when reflowed and pass all tests. And during MLCC production this could happen as well as they are heated several times and then pass the tests.

[jmelson]

Do you have any of these still on the original tape?  I'm wondering if the placement/handling forces on the P&P machine is possibly cracking the parts.  If you can find defective parts still in the tape, that would eliminate any handling issues in your facility.

Hmmm, 22 uF in an 0805 package?  Wow, that's a lot of capacitor in a small body.  And, they claim low moisture sensitivity.

Jon

[Corporate666]

I noticed that you never outright say whether you bought them at DK or not... Did you?

Also, I agree: get in touch with Samsung. Failing components off the reel should be in the low ppm region. Or, you know, nothing at all.

Yep, I bought them direct from DK, new on the reel.

I got in touch with Samsung and after much transferring and people answering the phone who had no idea who I should talk to, I finally got someone who knew what they were talking about.

Their answer is that they have never heard of such an issue before, that it must be something I did, but I am welcome to send them any parts in my possession that failed testing and they will replace them like-for-like.

So basically totally unhelpful.

[IDEngineer]

You always need to check the actual curves of the exact component (if available). Rule of thumb: a smaller ("too-good-to-be-true") device usually has worse DC bias, ageing and temperature characteristics, all at once. So simply avoiding Y5V might not be enough. A lot of homework to do when designing.

This. Exactly this. I just handled populating the BOM for a new product and I spent more time on cap selection than any other type of component. There is NO consistency across manufacturers... sometimes Samsung has the better curves in its spec sheet, other times Taido Yuden, sometimes Murata, etc. Yaego doesn't even publish some important curves for many of their caps, so their products don't get approved for our projects.

There are no ideal components, but caps are especially weird beasts. We try hard to never run them beyond 33% of their rated voltage (5V bypass caps are always at least 15V, and often 25V) and then we consult the curves to see how they are derated with DC bias. We recently dropped 4.7uF/25V caps in one 12V application because we found that a 10uF/25V we were already using had MORE than 4.7uF effective capacitance at 12V. (Imagine what the 4.7uF looked like at 12V.) Yes, it violates our "33%" rule but if we're making an informed decision, operating within specs, and still below 50% rated voltage, why stock both parts? With the increased volume of the single part the price was basically a wash. The package sizes were the same too.

[georges80]

I noticed that you never outright say whether you bought them at DK or not... Did you?

Also, I agree: get in touch with Samsung. Failing components off the reel should be in the low ppm region. Or, you know, nothing at all.

Yep, I bought them direct from DK, new on the reel.

I got in touch with Samsung and after much transferring and people answering the phone who had no idea who I should talk to, I finally got someone who knew what they were talking about.

Their answer is that they have never heard of such an issue before, that it must be something I did, but I am welcome to send them any parts in my possession that failed testing and they will replace them like-for-like.

So basically totally unhelpful.

I wouldn't say unhelpful, but realistic. Pull parts off your reel and send them in for testing if you are positive that the problem was not caused by mishandling in the supply chain or within your own assembly/storage facility.

I've sent components to manufacturers and that is the correct way to determine the actual problem versus expecting the internet to resolve the issue... Not being harsh, but consider that they have the equipment to determine if the faults are in handling or in the original Samsung assembly. I've used and use Samsung caps (as well as other mainstream sources) and never seen issues that weren't directly attributed to handling/assembly issues on the assembly floor - even then it is in the percent of a percent (or less)...

cheers,
george.

[Corporate666]

I wouldn't say unhelpful, but realistic. Pull parts off your reel and send them in for testing if you are positive that the problem was not caused by mishandling in the supply chain or within your own assembly/storage facility.

I've sent components to manufacturers and that is the correct way to determine the actual problem versus expecting the internet to resolve the issue... Not being harsh, but consider that they have the equipment to determine if the faults are in handling or in the original Samsung assembly. I've used and use Samsung caps (as well as other mainstream sources) and never seen issues that weren't directly attributed to handling/assembly issues on the assembly floor - even then it is in the percent of a percent (or less)...

cheers,
george.

They are not offering to test and analyze the reason for failure... they stated that this has never happened before, and therefore must have been something I did, but they would replace any failed caps with equivalent parts if I send them in for exchange.

I've had similar "service" from Atmel back in the day.  I have also had excellent service from Alpha Omega when I was having issues with a regulator failing abruptly in a specific application.

YMMV, I guess.

[exmadscientist]

Can't say I'm surprised to hear that it's Samsung who made these parts. I've seen one too many odd things with Samsung components and have since blacklisted their passives. Murata is only a tiny bit more expensive and supplies much better specifications and design support tools. TDK, Taiyo Yuden, and Kemet are also good, but rarely better value than Murata. Yageo C0G parts are decent (C0G being harder to screw up than X5R/X7R) and cheap.

I agree that selecting capacitors takes a huge amount of design time. About the only thing that takes longer is choosing a diode, but there are many fewer diodes in a design and so the total time is much less. The advice to stick to 1210s and smaller is well taken; the big ceramics are just too fragile.

[Siwastaja]

they stated that this has never happened before

And, following this logical principle, it will, by definition, never happen .

[sean0118]

Send them into Dave to dissect under his microscope. 

[wraper]

Send them into Dave to dissect under his microscope. 

Dave does not have tools to cut MLCC in half and I don't recall him having a microscope with enough magnification.

[IDEngineer]

The advice to stick to 1210s and smaller is well taken; the big ceramics are just too fragile.

Plus, if you need higher capacitance, you're better off using multiple caps in parallel anyway. The unwanted ESR and ESL is thus paralleled, reducing the effect of these unavoidable imperfections.

[T3sl4co1l]

The advice to stick to 1210s and smaller is well taken; the big ceramics are just too fragile.

Plus, if you need higher capacitance, you're better off using multiple caps in parallel anyway. The unwanted ESR and ESL is thus paralleled, reducing the effect of these unavoidable imperfections.

Not that you have to worry about ESR with ceramics.  ESL is a part length thing, so that's true, but you're probably in need of a better solution if you're that critical on it (i.e., adding ferrite beads and additional filtering stages to meet EMC).

Personally, I'm not opposed to 1812s, and of course any larger caps on a leadframe are fine (but they're usually prohibitively expensive).

Tim

[Ice-Tea]

I noticed that you never outright say whether you bought them at DK or not... Did you?

Also, I agree: get in touch with Samsung. Failing components off the reel should be in the low ppm region. Or, you know, nothing at all.

Yep, I bought them direct from DK, new on the reel.

I got in touch with Samsung and after much transferring and people answering the phone who had no idea who I should talk to, I finally got someone who knew what they were talking about.

Their answer is that they have never heard of such an issue before, that it must be something I did, but I am welcome to send them any parts in my possession that failed testing and they will replace them like-for-like.

So basically totally unhelpful.

You pretty much have to find a faulty component *on the reel*, without processing further. If you find one of those, they have no place to hide. But I agree with others: chances that it's a Samsung issue are slim to none..