Choosing the Right Decoupling Capacitor for Lower Frequency Transients

I’m in need of a decoupling capacitor to capture lower frequency transients in my main power rail, running at 5V - 2 A. Currently, I’m using a Tantalum 100uF-16V capacitor, mainly for its SMD and compact features. However, I’ve come across information suggesting that Tantalum capacitors might pose a risk of flaming out in the presence of voltage spikes exceeding their rating. Any recommendations? Would it be safer for me to switch to a different one?

There are several issues here. Actually bursting into flames has been greatly improved over the years. Tantalums have excellent storage capacity and low ESR, but I would still probably go with ceramics. My reasoning is that the ESR is usually much better, and the cost is usually much better. Throw in that whole “bursting into flames” thing and I think that’s the best choice. If you have to have really high capacitance try a surface mount aluminum electrolytic.

2 Likes

Some good points have been raised in answering the initial question, but there are a few more things that may be less obvious and so benefit from being pointed out.

Tantalum capacitors.
Bursting into flames is a real failure mode, (and the only one that I’ve seen happen in the lab), but this is generally the result of either reverse polarization (i.e. assembly error), or violating the surge current rating (a design error). Component derating is necessary, depending on your industry, a 50% voltage derating and 30% surge current derating is not uncommon. When this is taken into account, the lifetime of the part can be very long (many quality test instruments from 30-40 years ago use these in large numbers and are still working fine. Note that only some Tantalum capacitor ranges are 100% surge tested, and in some industries this is a mandatory requirement (Space, Automotive etc.). If you are worried, select a family of parts that is 100% tested. Also be aware of thermal limitations and how they change the performance when getting close to the termperature limits.

Aluminium Electrolytics
First thing to look for is how much you are going to have to derate these to get a good lifetime. They often have lifetimes of 1000 - 10000 hours, and this applies when you have the highest voltage, temperature and ripple current. 1000 Hours is just 42 days, so serious derating of all three parameters is the only way to get a good lifetime from these parts. Although the failure modes do not usually involve fire, they can explode. From a circuit design point of view, their high frequency performance is driven by their size and construction, so like for like, expect Tantalums to out perform Aluminium electrolytics practically every time.

Ceramics
As has already been said, certain aspects of these parts make them a great choice; especially their super-low ESR. Naturally, there are some gotchas to be aware of. The ceramic itself is not the same in each device, they are therefore classified according to their properties. The “best” (by this I mean the closest to being an ideal capacitor) is the C0G (a.k.a. NP0) type. They are not really affected by anything, but they do not come in large capacitance value. When you want a large capacitance, say >1uF, there are compromises to make. Wide temperature range means you lose capacitance at the extremes. Having a constant DC bias also reduces capacitance. And the one that usually gets missed is ageing (the orientation of the barium titanate molecules becomes aligned as time passes so they can no-longer move slightly in response to charge), which also reduces capacitace further. These other ceramics can be placed in order of best to worst, this results in this sequence X7R (best), X5R, Y5V (worst). There are more types than these three, but the show the same pattern. The initial message in this thread referred to 100uF 5V Tantalum parts. The nearest you can get in ceramic is 100uF, 6.3V (or better, 10V to allow for voltage derating). The most cost effective ones are in 1206 footprint packages. You can’t get these in X7R, but you can in X5R. When you take into account the bias voltage induced reduction in capacitance, you might have something like 70uF initially, but given time, this decreases. After 2 years, it is likely to be in the 20-30uF range. If your circuit needed 50uF minimum for stability (say it is an output capacitor in a regulator circuit), and two years after selling the product you begin to get field returns, this could well be the issue. There is nothing you can do to stop this process, so you have to compensate for it in advance by putting in say 3 parts in parallel, or go with the 10V version instead and have 2 in parallel. The worst thing about this is the temptation to desolder and measure a capacitor off a faulty board, then measure its capacitance. This does not work because the heat from the desoldering process anneals the molecular structure and it will measure as good. If you want to measure it, you need to cut it off the PCB without damaging the terminations and measure it; then you will see it had dropped to 20uF!
Another classic gotcha is to use ceramic capacitors for the class X and Y capacitor filters in mains power input circuits. They may not have a DC bias, but still experience ageing. As a result, your board that had happily passed the EMC tests, when put back through the lab now fails on conducted emissions (and sometimes radiated emissions) on the power cable. The ceramic capacitors are probably to blame. For things like that, the film types remain a safe bet if you want to be sure that the product will stay the course when deployed.
A final thought on ceramic capacitors concerns where you mount them. If you use “V-groove” breakout for depanelising boards, this technique has a nasty habit of breaking capacitors that are very close to the edge. If you can’t move them more inboard, you need to talk to your capacitor supplier who will be able to advise you on minimum distance from board edge (which is dependent on both orientation and package size) to give you confidence in the breakout process. The other alternative is to specify a routed profile, and now you just need to be concerned about the precise locations of the “mouse-bites” that hold it into the panel.

From this, the obvious remark is that there is no panacea. But armed with the understanding of what you’re up against, you can better make an informed decision of which parts to use.

3 Likes

Thank you for the detailed insights! Given my requirements for a decoupling capacitor to capture lower frequency transients on a 5V - 2A power rail, I am leaning towards using Ceramic capacitors due to their low ESR keeping in mind the potential need for multiple units in parallel to ensure long-term stability and reliability.

For the ceramic capacitors, you mentioned the aging effect and its impact on capacitance over time. How can I best mitigate this aging issue to ensure long-term stability in my circuit?

To continue the ceramic capacitor discussion, it is best to understand the ageing a little better. The effect is described and shown as a graph against time on the Johanson Dielectrics website.

The critical bit is to see that the capacitance decay is exponential and has flattened out after about 2-3 years of use.

There’s not much that you can do to stop this happening, but since these dielectrics see a reduction in their dielectric coefficient as you apply a progressively higher bias to the part, this loss can be dealt with to some extent. A 5V circuit could use 6.3V parts, but the 10V parts will experience less bias induced loss of capacitance. If someone produces a 100uF 16V X5R part in a sensible footprint rather than an enormous one, that would be better still.

It follows that the best advice is to work out the minimum capacitance your circuit needs, then use the ageing related applications notes from the manufacturer whose parts you plan to use and do the maths. From this, you now know the minimum number of parts to get the capacitance you want.

Although this is not related to your initial question, the ESR of something like a 100uF 10V X5R ceramic capacitor will typically come in at least 10x better than the tantalum device you were originally considering. This can have more consequences than you might think. If this capacitor is located at the power input of your product, and you apply power from an AC-DC power converter via a long (say 1-2m) cable, think about the circuit this represents.

The ceramic capacitor, with its really low ESR and also ESL looks like a perfect capacitor. The cable from the AC-DC converter to your product makes a good approximation to a perfect inductor. The combination gives you an LC circuit with a high Q factor (the ESR is what lowers the Q factor). If you powered up the AC-DC converter first, then plug in the cable into your product the high-Q LC circuit says you will get a surge that could upset or damage the components on your board. This effect does not happen if you connect the AC-DC converter to your product first and then turn on the power to the AC-DC converter because its voltage rise time is much slower than the voltage jump at the moment you plug in the live cable into your product. Analog Devices have a really good applications note on this effect, look on their website for applications note AN88 “ceramic input capacitors can cause overvoltage transients”, it will also tell you how to prevent this problem.

3 Likes

Thank you for the comprehensive explanation and the additional insights on managing the aging effects of ceramic capacitors.

Your advice on the potential issues with low ESR and ESL in ceramic capacitors forming high-Q LC circuits with long cables from AC-DC converters is eye-opening. I wasn’t aware of this surge risk. I’ll check out Analog Devices’ application note AN88 as you suggested to learn more about preventing these overvoltage transients.