Reasons for NOT using SMD ceramic caps?

[daqq]

Hi guys,

I've started using SMD ceramic caps of higher values for power decoupling as well as the main filter cap in a switched DC/DC converter in a lot of places recently. They seem much more cost effective than tantals while keeping really great parameter - even when using double the number of them to keep the capacitance at least at that level. They are small, easy to use, easy to spread around the board. Also, they die more gracefully than tantals/electrolytes. Their ESR is great, other stuff as well, they seem absolutely perfect for power decoupling as well as main filtration caps... so, where's the catch?

I'm aware of:
Piezoelectric effects - buzzing from the caps when something is done at audible frequencies, and minor voltage fluctuations from external vibrations
Capacitance dependent upon voltage - can be reduced by overdesign (just slap 2-3 times the capacitance there)
Prone to cracks with mechanical stress
Great temperature dependency of certain dielectrics (not all)

So, if I'm OK with, or have taken steps to prevent/mitigate this, why should I use a tantalum or other for that kind of application (assuming a digital design)?

Thanks,

David

[mikeselectricstuff]

As long as you stick to X7R/X5R and avoid Y5V and similar rubbish they're fine.
I'd avoid using the smallest available size for a given value, but these are often more expensive anyway.
I typically use 0805 up to 1u, 1206 up to 10u/16v & 2u2/50v, and 1210 over that.
You do need to make sure that whatever voltage regs you use are happy with the low ESR - most modern ones are designed to cope and mention it in the datasheet.


 

[Simon]

I'll use ceramics up to a few hundred nF low voltage but after that your looking at tradeoffs. As Mike said stay away from the tiny ones. Capacitance is a physical thing, that's why IC's have next to no capacitors in them, expecting too much capacity out of too small a space is silly. I work with 1206 and up as I can't be arsed to bother with tiny fiddly things

[tszaboo]

Some linear regulators define minimum ESR on the output, which the ceramic can be lower. You might not see oscillation right away, just in certain conditions, which is hard to track down. Just make sure your voltage regulators like ceramics.

[daqq]

I did not know that voltage regulators might actually have a problem with that. I'll keep an eye out for that, thanks!

[hans]

The same can be said for in-rush current and voltage overshoot with ceramics. Their low ESR means there is little to no damping and with a long cable you've created a very nice LC circuit. I've been caught by this when a regulator with Vin abs. max 36V powered from 24V was blown up during hotplugging with a reasonable cable length (like a meter at most - not twisted). I measured voltages up to 48V for a brief moment (as a LC circuit connected to a step response will charge up to 2x Vin)

I since then don't do hotplugging on proto boards, but also make sure the power network contains either TVS protection and/or some sort of aluminum/tantalum capacitors that provide some dampening.
Some inputs of DC/DC will still require ceramic input capacitors for their high-frequency switching currents. In that case I usually will stick a relatively small (compared to the tantal/alu caps) capacitor close to the DC/DC.

[coppice]

So, if I'm OK with, or have taken steps to prevent/mitigate this, why should I use a tantalum or other for that kind of application (assuming a digital design)?

Looking at the converse issues, there are strong reasons for avoiding tantalum caps. In many environments you are banned from using them. They have two key problems - they present a fire hazard when they fail, and whiskers forming inside the capacitor can short the plates together. Most tantalums are used in situations where the current capacity of the system is adequate to blow out the tiny shorting whiskers, so you just see a hiccup in the system. Until this was figured out early electronic telecoms systems would give unexplained resets every now and then. Now they use solid aluminium electrolytics and ceramic capacitors, and you can see extremely long uptimes these days.

[tszaboo]

I did not know that voltage regulators might actually have a problem with that. I'll keep an eye out for that, thanks!

http://www.ti.com/lit/an/snva167a/snva167a.pdf Application note for further reading.

[timb]

So, if I'm OK with, or have taken steps to prevent/mitigate this, why should I use a tantalum or other for that kind of application (assuming a digital design)?

Looking at the converse issues, there are strong reasons for avoiding tantalum caps. In many environments you are banned from using them. They have two key problems - they present a fire hazard when they fail, and whiskers forming inside the capacitor can short the plates together. Most tantalums are used in situations where the current capacity of the system is adequate to blow out the tiny shorting whiskers, so you just see a hiccup in the system. Until this was figured out early electronic telecoms systems would give unexplained resets every now and then. Now they use solid aluminium electrolytics and ceramic capacitors, and you can see extremely long uptimes these days.

I'm surprised early telecom systems didn't use tantalytics.


Sent from my Smartphone

[mikeselectricstuff]

The same can be said for in-rush current and voltage overshoot with ceramics. Their low ESR means there is little to no damping and with a long cable you've created a very nice LC circuit. I've been caught by this when a regulator with Vin abs. max 36V powered from 24V was blown up during hotplugging with a reasonable cable length (like a meter at most - not twisted). I measured voltages up to 48V for a brief moment (as a LC circuit connected to a step response will charge up to 2x Vin)

I've been caught by that one as well- fortunately there was a polyfuse I could replace with a 1R resistor

[T3sl4co1l]

Looking at the converse issues, there are strong reasons for avoiding tantalum caps. In many environments you are banned from using them. They have two key problems - they present a fire hazard when they fail, and whiskers forming inside the capacitor can short the plates together. Most tantalums are used in situations where the current capacity of the system is adequate to blow out the tiny shorting whiskers, so you just see a hiccup in the system. Until this was figured out early electronic telecoms systems would give unexplained resets every now and then. Now they use solid aluminium electrolytics and ceramic capacitors, and you can see extremely long uptimes these days.

Hmm, never heard that as an operational mechanism.  They do usually contain some silver, but it wouldn't make sense that it would be able to grow whiskers in that direction (i.e., through the solid electrolyte and into the dielectric).

My understanding is, the dielectric is normally cracked all over the place, it's just to a matter of degree; and where it's cracked, it can break down, which causes the electrolyte (MnO2) to heat up, release oxygen, and turn into Mn2O3, an insulator.  Effectively self-healing the short.  At manufacture, parts are tested to ratings, but the rating is easily halved from the stress of soldering; it's further reduced by strenuous environments (e.g., high supply ripple, wide temperature swings).  The self-healing process is further aggravated by high peak current availability, such as unfused, low impedance supplies.  (Which would seem to suggest that the voltage has to drop suddenly during the self-healing process, which could explain those random upset failures you mentioned.  I haven't heard anything about it, probably because that would be a rather embarrassing failure mode -- talk about popcorn noise!)

The two theories (or arguably, hypotheses, depending on corroborating evidence) should exhibit much the same symptoms and effects, though.  Or even, if it's just a matter of "whisker" being the presence of material, versus "crack" being a presumed absence, it's pretty much the same thing on a small enough scale (or if the crack happens to be filled with a 'whisker' of the solid electrolyte).

Tim

[David Hess]

T3sl4co1l, your understanding is the same as mine.  AVX and other including NASA have detailed documentation about solid tantalum capacitor reliability.

I will add that I have never run across a voltage derated solid tantalum which failed.  Voltage derating makes for fewer defects at a lower voltage and raises the threshold for thermal runaway.  I remember old marketing literature and application notes which said voltage derating was unnecessary for solid tantalum capacitors and some designers followed this advice.

They now make special surge rated solid tantalum capacitors which have a controlled ESR or other characteristics to prevent this issue.

Ceramic capacitors have their own failure modes including destructive shorts.  As noted earlier, their low ESR needs to be accounted for when they are used for bulk capacitance.

[coppice]

Hmm, never heard that as an operational mechanism.  They do usually contain some silver, but it wouldn't make sense that it would be able to grow whiskers in that direction (i.e., through the solid electrolyte and into the dielectric).

My understanding is, the dielectric is normally cracked all over the place, it's just to a matter of degree; and where it's cracked, it can break down, which causes the electrolyte (MnO2) to heat up, release oxygen, and turn into Mn2O3, an insulator.  Effectively self-healing the short.  At manufacture, parts are tested to ratings, but the rating is easily halved from the stress of soldering; it's further reduced by strenuous environments (e.g., high supply ripple, wide temperature swings).  The self-healing process is further aggravated by high peak current availability, such as unfused, low impedance supplies.  (Which would seem to suggest that the voltage has to drop suddenly during the self-healing process, which could explain those random upset failures you mentioned.  I haven't heard anything about it, probably because that would be a rather embarrassing failure mode -- talk about popcorn noise!)

The two theories (or arguably, hypotheses, depending on corroborating evidence) should exhibit much the same symptoms and effects, though.  Or even, if it's just a matter of "whisker" being the presence of material, versus "crack" being a presumed absence, it's pretty much the same thing on a small enough scale (or if the crack happens to be filled with a 'whisker' of the solid electrolyte).

British Telecom, AT&T and some defence bodies all gave the explanation I wrote in the late 80s and early 90s. This was based on their analyses of sporadic crashes in systems running continuously. When quirky telephone exchanges had all their solid tantalums changed to solid aluminiums, which were pretty new at that time, the quirks went away. This suggests the conclusion of sporadic shorts was correct. Its quite possible their analysis of the actual chemical mechanism was wrong, and perhaps yours is a later explanation.

[coppice]

I will add that I have never run across a voltage derated solid tantalum which failed.

How much do you expect to derate them? 16V and 25V solid tantalums used at 3.3V and 5V certainly aren't immune to catastrophic failures.

[T3sl4co1l]

I've got a 1993 vintage 16 bit LAN card here (IBM 60G0611) that has four 22uF 25V tantalums and one 10uF 16V.  Think I traced the power lines before and it only uses +5V and GND from the edge connector.

Also good illustration of the correct way to do Ethernet clearance, isolation, routing and ESD, and of layout / placement / routing practices typical of the 90s.

5V on 25V parts is a rather excessive derating I'd say, but for most purposes, 1/3rd is considered "high reliability" and 1/2 "good enough" I think.  So, 10-16V parts should be fine otherwise.

Tim

[David Hess]

I will add that I have never run across a voltage derated solid tantalum which failed.

How much do you expect to derate them? 16V and 25V solid tantalums used at 3.3V and 5V certainly aren't immune to catastrophic failures.

All of the ones that I have run across which failed were 6.3 volts on a 5 volt supply and 16 or 20 volts on a 15 volt supply.  Derating helps but by itself is not enough if the surge current is high which may explain other failures not related to dielectric breakdown.

[Bud]

As Mike said stay away from the tiny ones.

That depends. I found smaller size cap work better for bypass function in switching power supplies I built. They filter out more high frequency noise, perhaps because they have less lead inductance.

[David Hess]

As Mike said stay away from the tiny ones.

That depends. I found smaller size cap work better for bypass function in switching power supplies I built. They filter out more high frequency noise, perhaps because they have less lead inductance.

I am surprised you could measure a difference unless they were a lot smaller.  Physically smaller and lower value capacitors have a higher self resonant frequency.  I have occasionally run across designs where the decoupling capacitors values were selected to create a null at a specific interfering frequency.

They make surface mount ceramic capacitors with the contact terminations on the long sides for extra low ESR and ESL.

[coppice]

As Mike said stay away from the tiny ones.

That depends. I found smaller size cap work better for bypass function in switching power supplies I built. They filter out more high frequency noise, perhaps because they have less lead inductance.

I am surprised you could measure a difference unless they were a lot smaller.  Physically smaller and lower value capacitors have a higher self resonant frequency.  I have occasionally run across designs where the decoupling capacitors values were selected to create a null at a specific interfering frequency.

They make surface mount ceramic capacitors with the contact terminations on the long sides for extra low ESR and ESL.

Physically smaller capacitors are easier to get really close to the thing you are trying to decouple. However, he might have meant smaller in both size and value.

[a210210200]

Another problem is sometimes in certain decoupling applications you simply need a high value cap which is electrolytic type. Attempting to use ceramic caps in this case would look absurd. Also if board flex is an issue you can get reliably problems in things like automotive/industrial settings where they have mechanically buffered ceramics at increased cost. (Not really sure if those are needed or how widely they are used)

A strange case is that SMD ceramic caps for students all look very similar so it is really easy for them to mix stuff up and start wasting the entire reel getting new parts. I've color coded them by using different dielectrics which all meet/exceed spec to at least help that (SMD ceramic caps have no markings typically). Some don't seem to know how to use a multi-meter with capacitance functionality which we provide for free with instructions but that bit is being worked on... (To be fair our leads are really cheap/old and probing a free ceramic is a bit difficult especially down at the 0603 level for an untrained student who never worked with the stuff before)

We also provide a hard template boards for people to stick the bits of cut tape onto so they can keep track but usually it ends up all in one box with a pile of lose 0603 bits everywhere and the student lab a total mess. (With all the copper wire gone every year...)

[Simon]

As Mike said stay away from the tiny ones.

That depends. I found smaller size cap work better for bypass function in switching power supplies I built. They filter out more high frequency noise, perhaps because they have less lead inductance.

I think we are confusing small size and small value maybe ? all SMD caps will have little to no lead inductance, but physically smaller parts will have more problems like lowering their value as the voltage rises. Also I find 1206 parts easier to work with for me on size and for layout as it makes routing easier when traces can go under the part.