EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: Psi on November 11, 2022, 04:26:53 am
-
So unfortunately I got some bad ceramic caps from a PCBway PCBA order.
Kind of my fault, I should have specified a list of known good sources where the were allowed to buy from in the BOM.
They were supposed to be TDK C3216X5R1V156M160AC 15uF 35V
But when measured I found something like
5% were ~31uF
49% were ~27uF
30% were ~24uF
10% were ~15uF
10% were ~10uF.
When powered from 34V on PSU about 1 in 30 will immediately crack and goes dead short.
Obviously they are non-genuine and probably not 35V caps, but I was just interested in the massive wide range in capacitance.
Are ceramic caps binned for capacitance and or voltage, and maybe these are unsorted?
-
For shits an giggles, here's the aftermath of one going short on a 6 layer 3oz pcb
(https://www.eevblog.com/forum/projects/bad-ceramic-caps-(exploding)/?action=dlattach;attach=1636610;image)
-
That pic is truly painfull. |O
-
(...)
Are ceramic caps binned for capacitance and or voltage, and maybe these are unsorted?
I do remember an extremely interesting episode of the Amp Hour regarding this topic. The guest was employed by an international manufacturer of passive components, AVX if I remember correctly.
Dave asked about the same topic: are capacitors sorted at EOL to fit the specified design target? The answer surprised me: the characteristics of MLCCs are defined by design; as long as the process is stable, the capacitors will fit the specified voltage, capacitance, ESR, etc., without any need for sorting.
Who knows where your caps come from. The process is seemingly out of control, although the capacitance is generous. They may be perfectly fine at half the rated voltage, as in most applications.
To be honest, I would be nervous at 34V DC on 35V OEM caps as well. Regarding the aftermath, is there no current limitation or fusing in your circuit? Or maybe the capacitors do not crack short, but slighty resistive instead, leading to large heat dissipation?
-
There is a NASA study, that found that you need 0% de-rating of ceramic capacitors.
I don't think they were considering how low the quality of chineseum components can go.
From the wide spread, I would guess these are Y5V or some lower grade parts.
-
To be honest, I would be nervous at 34V DC on 35V OEM caps as well. Regarding the aftermath, is there no current limitation or fusing in your circuit? Or maybe the capacitors do not crack short, but slighty resistive instead, leading to large heat dissipation?
Same, the 34V was just me testing them to see how many blew up at supposed rated voltage. Normally they have around 14V DC on them with maybe a little ringing.
They definitely 'crack', I can see the fracture line under the microscope where an inner layer becomes visible after separating.
-
internal layer fracture line / crack.
You can see the crack on the surface at the right of the image and then proceeds downwards at a shallow angle.
(https://www.eevblog.com/forum/projects/bad-ceramic-caps-(exploding)/?action=dlattach;attach=1636865;image)
-
is there no current limitation or fusing in your circuit?
Normally I'd put two in series, but in this application the datasheet has specific requirements about how close the cap has to be to the chip. Putting two in series would increase this distance a bit.
I dont think a fuse would have helped in this case. The current had to pass through a 6A PFET and that pfet survived.
I think it may have been a slow burning short after the cap failed. But hard to say. It burned down at least 3 layers into the PCB so it seems to have been burning for a while.
I had others exploded when I was testing them after the fact.
-
Could it be a bad batch? Also, the capacitance will decrease under bias.
-
Could it be a bad batch? Also, the capacitance will decrease under bias.
I think it was just counterfeit batch. PCBway purchased the part number I spec'ed but because I did not specify where they were allowed to buy from it came from a non-authorized seller.
If they all measured similar then I might have assume just a mislabeled part with lower voltage than required.
but lots of failures combined with an absurdly wide capacitance range makes me think it was unsorted junk/rejects.
-
How did you measure them? Desoldered and measured? Or did you get excess parts on tape and measure them?
If desoldered, they may have been damaged in assembly, or more likely, during desoldering. MLCCs require careful soldering, a well maintained process. Desoldering safely is almost impossible, which is why I never try to reuse desoldered MLCCs.
-
To be honest, I would be nervous at 34V DC on 35V OEM caps as well. Regarding the aftermath, is there no current limitation or fusing in your circuit? Or maybe the capacitors do not crack short, but slighty resistive instead, leading to large heat dissipation?
Same, the 34V was just me testing them to see how many blew up at supposed rated voltage. Normally they have around 14V DC on them with maybe a little ringing.
They definitely 'crack', I can see the fracture line under the microscope where an inner layer becomes visible after separating.
For testing the 34 volts should be perfectly fine. Good 35 volt ceramic capacitor probably handles in short term hundreds if not couple of thousand volts before sudden failure.
Voltage (and capacitance) rating on ceramics is line drawn on water: Actual capacitance at rated voltage could be as little as 5% of rated capacitance and on the other hand the dielectric would handle even lot higher voltage than the rated volts but you don't have even that 5% capacitance left.
-
When powered from 34V on PSU about 1 in 30 will immediately crack and goes dead short.
Are the ones that measure significantly high on capacitance the same ones that blow up at 34V?
-
https://nepp.nasa.gov/files/25613/Teverovsky_TranDielectrics2012.pdf
Even if you had 6.3 volt ceramics sold to you as 35 volt rated they should not fail instantly.
6.3 volt rated ones should handle 300 to 600 volts before breakdown.
-
The ones I tested capacity on were from the pile of caps I pulled off the PCBs.
-
The graphs in this thread suggest fairly strongly that the voltage rating is baloney, at least for the good manufacturers.
https://www.eevblog.com/forum/projects/is-ceramic-capacitor-package-relevant/ (https://www.eevblog.com/forum/projects/is-ceramic-capacitor-package-relevant/)
Obviously I'd never do a design with an insufficient rating on it, that's just bad engineering. But I suspect the process is so good and so cheap that they just make the maximum voltage rating and then sell it at all levels below that too, so the robots who look for exact spec matches are happy.
If your supply chain is iffy then you are probably buying garbage. It's not worth playing the "dodgy supplier" derating game - it's a race to the gutter. Today you have to derate by 50%, tomorrow it's 75%, and soon it will break if you look at it funny. I don't have time for those kinds of shenanigans.
-
But I suspect the process is so good and so cheap that they just make the maximum voltage rating and then sell it at all levels below that too, so the robots who look for exact spec matches are happy.
Nope, running caps of the same series/design but different voltage does produce a different failure curve. I'd be guessing the voltage binning is from a very fast leakage + HALT test that weeds out impurities/inclusions/point defects.
-
The ones I tested capacity on were from the pile of caps I pulled off the PCBs.
Very possibly damaged by "pulling off" the PCBs, and not fakes. Especially large size ceramics need super-careful desoldering.
I have had good quality Murata 1210 MLCCs get damaged and fail as short circuits by just soldering them in, in uncontrolled process and poor pad layout.
-
If desoldered, they may have been damaged in assembly, or more likely, during desoldering. MLCCs require careful soldering, a well maintained process. Desoldering safely is almost impossible, which is why I never try to reuse desoldered MLCCs.
I've had no issues desoldering and reusing large MLCCs, with the proper tools: hot tweezers, or hot air and leaded/low temp solder. If you went wild trying to desolder them with a typical iron and force I could see that.
Of course something caused them to fail in the first place, so who knows what that was.
Very possibly damaged by "pulling off" the PCBs, and not fakes. Especially large size ceramics need super-careful desoldering.
I have had good quality Murata 1210 MLCCs get damaged and fail as short circuits by just soldering them in, in uncontrolled process and poor pad layout.
Soldering in by hand or?
-
If desoldered, they may have been damaged in assembly, or more likely, during desoldering. MLCCs require careful soldering, a well maintained process. Desoldering safely is almost impossible, which is why I never try to reuse desoldered MLCCs.
I've had no issues desoldering and reusing large MLCCs, with the proper tools: hot tweezers, or hot air and leaded/low temp solder. If you went wild trying to desolder them with a typical iron and force I could see that.
Of course something caused them to fail in the first place, so who knows what that was.
High capacity MLCCs (at bleeding edge of volumetric density) have always been touchy/risky/low reliability. Hand rework on production lines to pull off the hot/shorted ones and install a replacement isn't that unusual. But reuse is a step further than I would go!
-
But I suspect the process is so good and so cheap that they just make the maximum voltage rating and then sell it at all levels below that too, so the robots who look for exact spec matches are happy.
Nope, running caps of the same series/design but different voltage does produce a different failure curve. I'd be guessing the voltage binning is from a very fast leakage + HALT test that weeds out impurities/inclusions/point defects.
I'd be interested to read a test report if you know of any.
This earlier comment suggests that they don't bin them after manufacturing:
I do remember an extremely interesting episode of the Amp Hour regarding this topic. The guest was employed by an international manufacturer of passive components, AVX if I remember correctly.
Dave asked about the same topic: are capacitors sorted at EOL to fit the specified design target? The answer surprised me: the characteristics of MLCCs are defined by design; as long as the process is stable, the capacitors will fit the specified voltage, capacitance, ESR, etc., without any need for sorting.
Maybe it depends on the manufacturer and how mature and conservative the process is, as time goes on they may have the confidence to test less and throw higher bins into the lower bin SKU much as the CPU manufacturers do with their processors (slightly different motivations there though, they want to milk as much revenue from selling low speeds as possible). And in a very similar way, you can't count on "overclocking" your voltage ratings because it might blow up in your face ;D
-
Would you consider 15uF 35V 1206 as 'bleeding edge' and at higher risk of failure.
There does seem to be a change going above 10uF at 1206. Price jumps a lot. and voltage falls from 50V to 35V
-
Maybe it depends on the manufacturer and how mature and conservative the process is, as time goes on they may have the confidence to test less and throw higher bins into the lower bin SKU much as the CPU manufacturers do with their processors (slightly different motivations there though, they want to milk as much revenue from selling low speeds as possible).
Off topic, but it's a weird argument. They want to milk as much as possible revenue from high speed models as the same die can be sold for much more. However they need to satisfy the demand for lower cost models as those are sold in larger numbers. When process is immature they may have only a handful of dies which satisfy requirements of high tier. However as process becomes more mature, they will have more of higher bin dies which need to be downgraded to satisfy demand in lower price market segment. They would sell those dies as high tier if they could do so without destroying pricing structure.
-
It can be very hard to test lower binned parts to verify they're ok for your application. When you buy/test them you don't know if they truly are at the lower spec or not.
It's common to have manufacturers making more 'good' binned parts than they can sell. So many get sold as the lower bin parts and you may end up testing those ones without knowing.
Usually this is more of an issue with ICs, not really an issue with passives as testing them is easy and does not require specialized gear
-
Very possibly damaged by "pulling off" the PCBs, and not fakes. Especially large size ceramics need super-careful desoldering.
Soldering in by hand or?
Layout was two parallel polygon fills (Vcc and GND) without solder mask. No thermal reliefs for capacitors, because capacitors were retrofitted. Hand-soldered close to each other, side by side. Out of ~200 capacitors, ~2 failed, and only after >1000 hours of continuous operation followed by power-off then power-on.
Non-current-limited DC battery bus 10V, part was Murata 50V X7R 10uF 1210.
You can find appnotes that are quite demanding if you want manufacturers to guarantee MLCC reliability. It's not limited to zero board flex. Thermal reliefs, balanced pad thermal masses, PCB artwork so that warping is avoided, correct reflow temperatures, etc. I have seen an appnote that basically forbids hand soldering completely.
I desoldered all the capacitors and soldered in new ones, more sparsely, and VERY carefully by using minimal amount of solder and minimum amount of time soldering each terminal. Failures stopped.
EDIT: I found my old post about the failure, you can see a photo and the issue being too much solder: https://www.eevblog.com/forum/projects/mlcc-short-failure/msg699778/#msg699778 (https://www.eevblog.com/forum/projects/mlcc-short-failure/msg699778/#msg699778) . But note, there was no board flex! Many people think it's enough to avoid board flex.
-
Would you consider 15uF 35V 1206 as 'bleeding edge' and at higher risk of failure.
There does seem to be a change going above 10uF at 1206. Price jumps a lot. and voltage falls from 50V to 35V
That's somewhere close to the edge, only a few years ago it would have been bleeding ;) If you have lots of failures from the batch consider handling error/failure of the reel before placement.
-
Capacitance variations could also be due to aging depending on the dielectric.
Class 2 have ferroelectric material and can degrade. Fortunately it is reversible.
https://www.edn.com/class-2-ceramic-capacitors-can-you-trust-them/ (https://www.edn.com/class-2-ceramic-capacitors-can-you-trust-them/)
-
Capacitance aging is a thing. However you won't get 30uF instead of 15uF after heating. In a range of +20% above nominal at most, and quickly dropping close to nominal in a first few hours. So as there are 30uF caps, most likely a mix of right and wrong capacitors were assembled. Or maybe just all wrong capacitors and some of them lost part of capacitance due to cracking.