Home [en] [de] /

My Little Capacitor Blog

Intro

A capacitor is a capacitor is a capacitor? Oh, you wish. This page is dedicated to a few of the numerous complexities associated with the little buggers.

Want to access quick navigation directly (N)?

Derating Practices

While any part should have no problem sustaining its rated limits (e.g. voltage or power dissipation) for a while, it is often unwise to subject it to these conditions in the long term. The practice of derating refers to picking safe operating conditions so that the part can be expected to last at least as long is it needs to.

By default, I would pick as a rule of thumb:

A peculiarity found in electrolytic capacitors is that their voltage ratings generally are working voltage ratings (you may have seen "WV" printed on some), with short-term surge voltage rating being 20-25% higher. Even then, there is substantial variation between different manufacturers and capacitor series in practice, as you'll find when doing some leakage current testing (which I have). Some from high endurance series I would trust running at full rated voltage all day every day (at least when new – it's probably not a good idea if you still want them to work in 40-50 years), while a lot of the time I'd rather not be exceeding 85% or even 80% of that right out of the gate if I can help it.

Here is a list of various degrees of derating for electrolytics that I have identified:

Now you're not entirely exempt from luck of the draw – I have still seen at least one example of failure in a nominally good-quality (if 85°C) general purpose series cap that had been properly derated to 50% (mind you, that case was accompanied by scores of SMD electrolytics being stone-dead, so I assume the unit had been running hot for years, possibly due to inadequate ventilation), while in other cases some had failed short but by no means those seeing the highest voltages among their peers. So ultimately we're talking rules of thumb here. Plus, it should be obvious that a part fundamentally less prone to drying out will generally be lasting longer even under identical conditions (temperature, voltage, ripple current and capacitor ESR relative to volume).

Another complication arises at low rated voltages, where the dielectric gets thinner and thinner. While at least some makers seem to be engineering in some extra voltage headroom (even in nominal 16 V parts) to prevent excessive leakage and potential failure after prolonged no-voltage conditions have been gnawing away at the oxide layer, I wouldn't bet my life on it and conduct leakage testing if in doubt. Lower-voltage types certainly do exhibit an increased failure rate from experience. When faced with the choice of 3300µ/6.3V vs. 2200µ/10V of same size and similar ESR for a 5 V rail, I would generally side with the latter unless I had enough confidence in capacitor quality.

↑ Back to table of contents

Entry last modified: 2026-10-19 – Entry created: 2026-10-09

More Tips for Treatment

As a hobbyist (or "hobbeast" if you prefer) you have one option that a large-scale producer may not: You can decide to reform your electrolytics on a lab power supply with some series resistance prior to installing them, in order to give them the best fighting chance of making it the next couple of decades. 24-48 hours may be a bit much to ask for, but 10-30 minutes at 120% of rated voltage should help them calm down a whole lot. (If your lab PSU can only do 30 V, you may need a lot of patience for 50V types and up. It can still help cushion the blow of initial power-up.) This is particularly recommended for any stock that's been gathering dust for more than 2-3 years. Given that a lot of what I've been buying has been 3-10 years old and at times downright historic (plus I tend to be interested in leakage current), I'm now doing that on a fairly regular basis.

In this context, it is also worth mentioning that bending capacitor legs right at the seal may damage said seal and lead to drying out long-term, so I would very much recommend holding the leg with small needlenose pliers at the seal so that a small unbent section remains. This way you can still create sharp bends to accommodate a different lead pitch and whatnot. (Side note, I would not resort to overly elaborate gymnastics when replacing axials with radials. Just two 90° bends outwards a few millimeters outside the seal, and back down where needed.)

Speaking of best practices, large temperature differentials in surface-mount MLCCs can lead to cracks forming, which as you may know is a precursor to silver migration and eventual shorting, a very common failure mode for these (and the increased frailty associated with lower rated voltage doesn't make it any better). Therefore, traditional hand-soldering is not recommended for them or not without a great deal of caution anyway (e.g. having a tip large enough to cover both ends at once, pre-heating, whatever). Hot air is a better choice for them.

Double-layer capacitors (supercapacitors, goldcaps) for memory backup should be charged gently (100+330 ohms in series will not go amiss) and most of all reside in the coolest place you can possibly find, as heat is a primary contributor to aging. There is a special place in Hell for equipment designers who decided that an unreachable spot behind a VFD (which obviously gets warm in operation) would be just about perfect. I am seeing a near 100% failure rate in constructions like that, and getting the old cap out is a major challenge at the very least. Right next to a toasty spot in the power supply is similarly problematic, though at least it makes it easier to fit a replacement. By contrast, every device with it placed somewhere on the board far away from heat sources is still working fine for me (well, except the one where the cap is fine but the problem is a degraded SRAM chip known to be troublesome).

If you want the electrolytics in your equipment to last, use it sparingly but regularly. Anything that has been "rode hard and put away wet" (used extensively followed by years and years of storage) is prone to having one or more go short.

Long-term standby operation is an underappreciated "silent killer".
I once had an original '90s Bose Wave Radio on the bench in the local repair cafe, which it turns out basically leaves most of everything running in standby (over 7 W from the mains). Not only was it chock-full of dust, the main filter cap squeezed in right next to the amplifier heatsink proved stone dead (it was a 1000µ/25V Nichicon) and misc. 100µs weren't exactly exciting in the ESR department either.
And then there's my Pioneer F-757MkII which had a bunch of general purpose series caps strategically grouped around three permanent heat sources in a rather cramped power supply layout. I had to recap almost the entire power supply section (more than a dozen caps), and it was a pain too because the legs were bent over and as a result I had to fix a bunch of PCB damage as well. Interestingly, all the higher-grade caps still measured completely fine, including the big 3300µ/35V filter cap (the sensibly-placed goldcap also proved to be working just fine). At least I was able to reduce thermal load by changing the mains transformer primary tap to 240V and largely eliminate one major heat source, and many capacitors were upgraded to lower ESR / higher endurance types, but it was a major ordeal either way.

↑ Back to table of contents

Entry last modified: 2026-10-26 – Entry created: 2026-10-09

Supercapacitor Alternatives for Memory Backup

Double-layer capacitors (supercapacitors, goldcaps) fundamentally are neat little devices: They can store a lot more charge per volume than a traditional electrolytic though generally with an ESR in the double-digit ohms, which makes them attractive to applications like providing power backup for things of relatively modest current draw. (Nowadays they also make low-ESR ones that really pack a punch for power buffering applications.) This made them a popular choice for memory backup from the mid-1980s into the 1990s and beyond.

However, they are not necessarily the most robust devices. For starters, the one thing they appreciate even less than overvoltage is heat. Placed somewhere in the middle of the board in a device that doesn't get very warm, they can last for many decades, but placed near heat sources in the power supply or sandwiched between the board and a VFD they tend to call it quits prematurely, losing all of their actual capacitance and just retaining a bit of dielectric absorption (something they are leaning on very heavily in general, so they are closer to a cross between a capacitor and a battery). Charging them with no current limit may also generate a fair bit of heat internally, so some series resistance is generally advised. To make matters worse, the common 5.5V rated type actually consists of two 2.7V cells in series, and if you've ever had to deal with cell balance in battery packs, you'll probably know what happens when one cell goes bad. Not exactly a recipe for graceful aging.

I had purchased some new-to-me supercaps and noticed a somewhat disappointing self-discharge rate in them. First lesson learned: Oversizing a supercap is absolutely pointless if internal leakage dwarfs your actual load. They also tend to require a fair amount of time before their DA has been recharged properly, which is not ideal when a device may only be used relatively briefly every once in a while.

At the same time, I knew that a traditional 2200µ/16V electrolytic can keep Kenwood KT-880L memory alive for way longer than its capacitance and the microprocessor (TC9147BP) datasheet would suggest. On paper, that's around a 15 minute time constant. In practice, I've left it unpowered for a week and the memories still were all there. I can only assume that actual current draw is way smaller than specified, and effective capacitance is being bumped up by dielectric absorption.

Now that was a device from 1986. As we move into the 1990s, the specifications for memory retention current only went one way: down. That made me wonder: Could you replace a supercap with a traditional electrolytic and still retain sensible memory backup times? Their leakage current tends to drop precipitously when voltage is derated substantially (hence why Kenwood opted for a 16V type in a 5–V application), and while not every series seems to be equally suited, I've seen very good voltage retention in some. In such a chill application, they can basically last forever, and their larger size is not a drawback in any device containing lots of air to begin with.

The opportunity for a proof of concept came when my Onkyo T-422's memory retention dropped to less than a week after it no longer saw regular use. Inside, I found a Murata 0.1 F goldcap of the vertically standing variety accompanied by a 22 ohm series resistor in the power supply section, with only a fraction of a volt on it. This area happens to be fairly warm, and I had already replaced a rather dead electrolytic close by earlier. After removal, the part only measured a few hundred nanofarads, so whatever it was still doing was almost entirely down to DA.

I had acquired and thoroughly pre-formed a Nippon Chemi-Con LBG 6700µ/25V recently – arguably a bit oversized (and not the smallest part at 18x25mm DxL), but if anything would work, it would have to be that, right? There was plenty of free space for it over the micro, I ran some twisted red and black wire to the original spot, and a bit of hotglue was used to keep the cap in place. I let the unit run for a good half an hour to charge and started taking voltage measurements every once in a while. Well, 90 days later, voltage had reached 2.0171 V from an initial 4.81 V, with the latest time constant computing to 127 days, over four months. And since memory seems to be retained down to the 0.7 V vicinity, odds are it would work for that much longer, too. Given that my original target had been a two week retention time, we can call that a resounding success.
(At the same time, a 0.22F/5.5V ELNA supercap left to discharge with no load had decayed to 1.28 V, and that's while starting a week later at 4.85 V. While its time constant had finally overtaken the electrolytic, it looks like voltage may not catch up until the 1 V vicinity, still enough to sneak past in practical memory retention but hardly awe-inspiring. I was expecting that to happen eventually, but it's turned out to be a much closer race than anyone would have expected.)

In case of of Onkyo T-4300 I was faced with a different conundrum: Even getting the front panel PCB unclipped proved a challenge, let alone accessing the supercap stuck under the VFD. Meanwhile, memory retention of only about a day got really annoying. Under these conditions, can you just tack on the wires from a new supercap on the solder side and call it a day? Yes, yes you can. It's probably not recommended, what with the dead part's leakage current and potential physical leakage, but it does work. Some people have also implemented approaches like slightly enlarging the PCB holes and sleeving the old cap's legs for insulation, or cutting its legs as short as possible and gently bending the PCB so it has enough space to drop out, both of which worked for them.

In terms of electrolytics suitable for memory backup, your best bet are probably going to be general purpose types or high endurance series with decent but not necessarily super-low ESR. If in doubt, you'll have to do some testing on what you can get – form thoroughly and let them sit.

↑ Back to table of contents

Entry last modified: 2026-10-23 – Entry created: 2026-10-09

My Capacitor Measurements

My collection of capacitor measurements over on diyAudio has gotten slightly out of hand, and I've had a few run-ins with the 20,000 character limit per post (how's that for arbitrary?) lately, but until I can figure out a sensible way of formatting and presenting the data, that's where it'll stay for now.

As you can see, you do not need particularly fancy gear to take useful measurements – a lab power supply that can output up to 60 V, a decent multimeter or better two, a basic ESR meter (which is really an impedance meter, but that's plenty useful as-is), some metal film resistors of 10k and 100k (I also like to keep a 1-2.2k/5W cement resistor at hand for discharging unruly capacitors), and then some test and clip leads and odds and ends. A temperature-controlled space will be helpful given that leakage and ESR can be highly temperature-dependent, but if you don't have one at least keep a record of ambient temperature.

The main thing you need is patience (or at least something else to do in parallel). Measuring leakage current in particular is an ideal hobby for those enjoying such thrilling activities as watching the proverbial paint dry or grass grow. I would always start high (up to 25% above rated voltage, at times I have even gone close to 50% higher if it seemed promising) and work my way down, with brief discharging pauses when changing voltage to give DA a chance to bleed off. Even so, you may find that getting down to rated voltage alone takes a long time for readings to stabilize, depending on how leaky the capacitor is to begin with. I am not one bit surprised to see that with 63V or 40V types at 10 V leakage current just kept on decaying exponentially even after more than a day.

After doing this for a while, you start noticing some quirks. For one, high voltage stress can lead to elevated levels of leakage current (maybe +25-50%) that refuse to go back to normal, until you thoroughly discharge the capacitor that is, which seems to "reset" things. Don't ask me what that's all about.

Here's an attempt at some semblance of capacitor series "pecking order" based on my measurements:

  1. Very good:
    Nichicon UPW (high endurance, decently low ESR, low leakage but takes a while to fully reform).
    Elite ED, Teapo SY. (Both similar to UPW, a hair short on capacitance but otherwise very good parts.)
  2. Good:
    Chemi-Con LBG (a bulk storage series of average voltage handling and average ESR, but capacitance beating spec by a fair bit).
    Nichicon PL (an older low ESR, long life series)
    Samwha WL, CapXon GL (lowish ESR, higher endurance types, average leakage / voltage handling).
  3. Fine:
    (NOS) Thomson-CSF THRS (good general purpose series caps – low leakage / good voltage handling, quick to reform and low self-discharge rate, capacitance very close to spec, mediocre ESR, 85°C only).
    Teapo SH (quite similar to THRS electrically, capacitance falls a bit short, leakage / voltage handling average to good, feels a bit lighter/cheaper)
    Jamicon HS (decent ESR and voltage handling / leakage, overrated for capacitance, holds voltage quite well, average endurance, a solid if unremarkable snap-in type)
  4. S'alright:
    Jamicon TK (a basic 105°C series that tries to be half-decent at least, arguably a bit short in capacitance and voltage handling but not half-bad ESR)
    Samwha RD (a general purpose type that doesn't set any records for leakage / voltage handling but sports good ESR in smaller sizes, nothing special once you get into the triple digits though)
    Elite PF (an unremarkable general purpose series of average ESR)
    Aishi WH (arguably overrated in both capacitance and voltage handling but also fairly small, decent ESR; treat the 2200µ/35V like a 2000µ/29V and you should be just fine)
    (NOS) Matsushita ED (seems like a decent higher voltage type)
    (NOS) ROE EKA (a solid lower ESR type of yesteryear, leakage rather uninspiring)
    (NOS) BC Components 037 (sort of like EKA, 100V not really holding voltage)
  5. Substandard:
    (NOS) axial Philips "L.L." (decent enough leakage, mediocre ESR)
  6. Exquisite Junque:
    (NOS) axial BC Components (ex Philips) of no particular repute

So far I've only seen higher-than-advertised capacitance (not caused by excessive leakage) with Japanese makers. The likes of Elite and Teapo can make very solid caps as well these days but they often underread by around 6% at room temperature. (Mind you, from a longevity standpoint I'd rather be seeing a cap that's exemplary for leakage and a bit short on capacitance than vice versa.) Thomson-CSF seems to have had a knack for getting it just right.

It seems that capacitor makers seem to be quite aware of the oxide layer's tendency to be gradually consumed when the part does not see any voltage, leading to excessive leakage and potential failure. This would obviously be more of a problem for low rated voltages. Two different general purpose 4700µ/16V types I tested could run at 20 V almost indefinitely, and sustain 16 V all day every day.

And now for some "retro charts":

↑ Back to table of contents

Entry last modified: 2026-10-19 – Entry created: 2026-10-10

Can You / Should You Reform 45-Year-Old-Capacitors?

The short answer: Maybe.

The process of reforming an electrolytic capacitor aims to chemically restore the Al oxide layer by gently applying voltage (generally through a highish resistor, I would keep 10k and 100k at hand for typical sizes) and having the chemistry do its magic. You generally bring things up to nominal rated voltage but applying 20% more for an hour after initial charging results look promising and then going back down should speed things up nicely if you're in a hurry. Reforming is recommended for any capacitor that's been sitting for more than 2-3 years, with specifics depending on chemistry (expect ultra-low-ESR types to need more of a helping hand) and storage conditions, and the longer they've been sitting the more thorough the treatment will generally have to be.

Now electrolytic capacitors are electrochemical devices whose construction and composition may very much vary, and which can fail in a number of ways. If the electrolyte goes off and becomes overly acidic, the oxide layer will never be able to regenerate. Likewise, watch closely for any signs of bulging that could be indicative of hydrogen buildup inside, again a sign that the chemistry is toast. If the part has developed internal arcing and parallel resistance, there'll be carbon tracking internally, so again forget about it. If the part has dried out because the seal may be compromised or just rock-hard (indicated by excessive ESR – caps around 1980 weren't that much worse than standard modern ones, maybe 50-100% higher but that's about it), it's a lost cause as well. If it's got a leg rotted clean off, well I guess it might still be good for a Monty Python skit or something, but not in its original function.

I did conduct a reforming attempt with the Elna 220µ/10V controlling turn-on time constant in my 1980 Kenwood KT-80 that I noticed had been giving me trouble lately, resulting in an excessively long turn-on delay on the first start in a while. It had not seen more than about 1 Vbe (~0.65 V) since 1980, and a lot of the time 0 V. Removing it showed an ESR of 0.13 ohms, so clearly plenty of electrolyte left. I should just have paid closer attention to the minor bulge at the bottom…

While the part initially proved quite leaky, I could eventually up the voltage and finally got 0.024 µA at 10 V out of it. Alas, the bulge seemed rather noticeable now, and after resoldering the part its ESR went up to 0.21 ohms, so some of the electrolyte had clearly called it quits. So… half a success, I guess? A vintage low-voltage capacitor was always going to be a long shot, but given its operating conditions it seemed an ideal candidate.

I had the "higher ESR after resoldering" problem with a small orange low-leakage ELNA of similar vintage as well, in that case I think it was just a seal problem (possibly gone rock-hard).

The best candidates are likely to be larger caps that are less prone to drying out to begin with.

One thing that is not going to work is recovering capacitors that have developed excessive leakage through wear in use (heat stress and ripple current). Reforming does well at patching up minor weaknesses in the oxide layer, but it's not at the same level of what they did at the factory, so it may not fully restore parts sitting for a long time either. Such parts may still be useful when derated somewhat, so you will need to measure leakage current at a range of voltages to determine a sensible new nominal rating. I would aim for about 0.1 µA per 1000 µF and 10 V (after 36-48 hours) for standard demands. Then normal derating practices can be applied to the new rating. (For example, you might find that a 25V part is still good as a 20V, so applying a 75% derating means you could run it at up to 15 V continuously.)

↑ Back to table of contents

Entry last modified: 2026-10-26 – Entry created: 2026-10-26

Quick links

Quick navigation

© Stephan Großklaß 2026. Commercial duplication of this content, including eBay descriptions and similar, with prior permission only.

Contact me