More bad caps

Different processes, Read the wiki on "Capacitor plague"

I've replace the CapXon brand caps in Dell monitors, there are some other brands that have similar venting issues.

Cheers

No sympathy:

Good observation. I've seen the same thing. It's probably a conspiracy.

Two guesses:

1. The parts are mismarked for a higher working voltage. It may say

25WV but is actually a mismarked 16WV cap.

1. The manufacturer rates their caps at an artificially lower temperature. The caps may say 85C or 105C, but those are the maximum recommended operating temperatures, not the temperature at which the rated working voltage is specified.

What I usually do is install the next higher voltage rating and hope that they last. The usual recommendation is 2x or 3x the operating voltage, but caps that big would never fit. I also check for self heating with a thermocouple probe, just in case I inserted the parts backwards (again).

I hope not. If you compare the volumes of the various caps, my guess(tm) is about a 2:1 volume difference, which is probably not the processes or construction, unless the manufacturer makes radical changes to the internal construction.

Lately, I'm seeing more shorted MLCC caps than bulging electrolytics. Stress or heat shock them with a soldering iron and they're as good as destroyed. I have to solder paste and a hot air SMT gun. Grrr...

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
Santa Cruz CA 95060 http://802.11junk.com 
Skype: JeffLiebermann     AE6KS    831-336-2558

Yeah, I've taken to checking caps any time I have the cover off a piece of kit. Rather replace (caps or the entire item) than have to deal with a failure in *use*.

I hadn't considered that. I've found (other) appliances with obviously incorrect component selection (e.g., 10V caps on 12V supplies) but, for the most part, the "(physically) smaller" caps I'm seeing *claim* to be appropriate for their actual use.

I don't go chase down the original manufacturer's data sheets to verify the "intended" size but, rather, note that the replacement (panny, nichi, etc.) tends to be noticeably larger for the same rating. To the point where I often have to install it "crooked" to avoid interference with other adjacent parts.

I usually just use similar voltage/temp rating but, depending on the perceived quality of the monitor, may opt for a 5000hr part instead of a 2000hr, etc. The components and time come out of my pocket ("pro bono" work) and, realistically, these sorts of devices (monitors) tend to get replaced periodically, anyway. If I can squeeze another year or two of life from it...

Dunno. It's been happening "enough" that I've wondered if there wasn't a deliberate reason. I see no *need* for physically smaller components in most of these cases. I.e., the (larger) replacement parts could easily have been accommodated had the components been spaced an extra tenth, etc.

Lots of bad FETs, here (in the inverter, of course). I probably have four or five of the identical model display all in need of FETs. I'll have to look at the circuit more carefully before I select their replacements. FETs tend to be harder to replace than eCaps...

Just had a power supply open this week. I checked the electros and they all were in spec.

Turns out that 0.1 ohms was also the DC resistance--near dead short.

Found out in short order when put back in service that I needed to replace all the caps.

Frequently I find very old caps (80's vintage) electros still well within spec. It's the 2000 and later caps that are failing. Usually.

I checked the ESR of the electros and they all were in spec.

Smaller means less heat dissipation anyway, so ESR bites harder.

NT

I've gone a step further and replace all the caps of a specific type or brand, even if only one of them shows obvious signs of bulging. The major time burner in doing this is NOT replacing the caps, but rather extracting and replacing the PCB or motherboard. Might was well replace everything rather than have to go back in again later.

I don't have any easy way to test the voltage rating on the caps. That's why it's a guess. I tried to compare the sizes with a reputable manufacturer (Panasonic). That catches about half of them. I have to guess on the rest. With the deration curve, I try to judge the correct voltage based on how hot the caps get. If they're near the CPU of have a high ripple current, I go for the biggest part that I can cram into the PCB. However, some areas are tight and I have to use expensive polymer capacitors. Those never seem to fail, but are not available in large values or high voltages:

Also consider installing the caps on the bottom of the PCB. I've also installed two caps, with half the capacitance. One goes in the holes while the other goes on the bottom of the PCB.

There is, but it's not really a conspiracy. The problem is that the lifetime of an electrolytic capacitor can easily be predicted if the ripple current, operating temperature, and working voltage are known. Manufacturers design to spec where one of the most important specs is expected useful life. Anything that lasts beyond the useful life is considered a waste of money and needs to be "economized". Under ideal circumstances, every component of the device blows up simultaneously. So, the working voltages are carefully selected for this goal. As an added bonus, this also provides a minimum cost solution for the choice of electrolytic caps.

When I see FETs blow, it's usually because a downstream electrolytic cap had shorted. I just hate replacing the FETs, only to have it blow again because I missed a shorted cap.

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
Santa Cruz CA 95060 http://802.11junk.com 
Skype: JeffLiebermann     AE6KS    831-336-2558

On motherboards, I usually won't waste time recapping. Too many units, often no thermal reliefs, etc. Just dump the machine. Monitors, OTOH, have few caps so easy (inexpensive) to replace them all.

Often not much space beneath.

*Lots* of "surplus"/faulty equipment to choose from so its usually not worth spending much time on any one piece. E.g., the ones with the bad FETs I would normally just skip over. But, having several of them with the same (apparent) problem makes it worth the time/effort to order replacement FETs. For "one off", it's just not worth the time.

Yup. But inverters typically only have two caps. So, a no-brainer to replace them. "Gotcha" is making sure to check the fuses (so you don't have to place a *second* order for "fuses" after ordering the FETs).

[Monitors take up a fair bit of room so I'm not keen on having more than one or two "in process"/disassembled at any given time]

AFAIK, electrolytics don't age in a voltage-dependent way; it's all about temperature, ripple and dissipation. Indeed, running them derated might have consequences, like capacitance drifting up (due to, er, de-reforming) or excessive ESR.

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com

Nah, just confirmation bias.

The old ones are well into their bathtub curve, and will eventually dry out from sheer time (whether in operation or in storage). The sample is also strongly selected towards professional equipment that's 1. survived this long to begin with, due to very conservative ratings, and 2. worth repairing rather than dumping in the pile (so who knows, maybe some of the things you find of that vintage are already on their second set of caps -- but how would you know?).

The ones that are actively beaten up, by being used at ratings, are the ones accumulating wear. But you'd never see such equipment from the 80s or 90s, because it was too cheap to bother servicing. All the equipment in this bracket is new, 2000s and up, because that's what's new or still surviving.

Depending on market segment, there may also be holes in the service life of these things, like, that bad batch of caps that was going around in,

2006 or whatever it was, would be highly selected against so you'd only see products in that series from 2005 down, and 2007 up; etc. But unless you're making intricate notes about what products you're servicing, when they were made, etc., and had a big enough sample size to be able to do this kind of analysis, you'd also never know.

I suppose it would be interesting, not only to record that kind of information, but to have *all* servicemen recording it, into a huge database; so one could study frequency of service across products, brands and production dates; reliability of service (repeat trips from bad repairs?), what was at fault, how it was replaced, and with what; and on and on. But that would be a pretty big overhead just for poking a soldering iron at some boards and seeing if it worked or not. Such a database might be useful for more expensive things (cars?), and is often used to generate metrics within a business (production and service).

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com

Warning, this can screw up a lot more than it helps. The ESR is so low (comparable to film capacitors, if they were available in these values!) that traces, even power planes can resonate; switching controllers will lose compensation and oscillate; etc.

The impedance is so low that large values aren't needed anyway, so -- in a design -- you can easily save 10x the capacitance. But in an existing product that's not designed for it, it can cause problems.

On the upside, modern computer hardware (say since 2010ish?) typically has such demanding current requirements that polymers are the natural choice already. Then the problem is, was what you picked to replace it with right or not...

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com

I'm not sure that's the case. I would expect all/most of the caps (on a particular board, from same manufacturer, etc.) to then be similarly "smaller" (of course, there are points at which the volume of the component can't effectively be scaled down to the "next smaller STANDARD package).

And, the differences in packages seem rather dramatic -- looking for any manufacturer with that particular size package for that particular cap typically turns up "no hits".

I am starting to wonder if Jeff's "mismarking" (etc) claim may be more to the point.

Of course, I only *see* the components when they've failed. So, I can't tell if there are similarly "(physically) smaller" components that have had no problems...

There was a time when I tried to track brands. Too many. Now, if the item is "worth recovering" (parts cost + time estimate + how easily and cleanly I can disassemble/reassemble), I just pull all the caps that are manufactured by the manufacturer of the noticeably bad cap.

An unnamed French online vendor of PC parts provided their product return stats to a French website, hardware.fr, which published two articles about it. Each article covers 6 months of data, from spring

2012 to spring 2013. Somebody translated it into English here:

or or even

I think this is just for initial returns from consumers; it doesn't talk about whether the faulty products were then serviced or what the results of that were. The wording implies that this is a regular feature on hardware.fr, but I only know about this translation off the top of my head.

tl;dr version: For best results in the time period covered in the article, buy a Gigabyte motherboard, Fortron/FSP* power supply, Kingston RAM, PNY video card, Seagate hard drive, and a Samsung SSD. For worst results, buy an ASRock motherboard, Corsair power supply, Corsair RAM, Sapphire video card, maybe a Western Digital hard drive, and an OCZ SSD.

• I think FSP used to sell PC power supplies under the "Sparkle Power" brand, but somebody finally convinced them not to do that.

Matt Roberds

The percentage numbers are useless. Without data on the number sold, the number returned, the time interval between sale and return, and the reason for the return, there's nothing on which to base any kind of reliability estimate. For example, one product could have a fairly high return rate, but only because the French distributor may have included returns for non-electronic reasons such as credit failures, sold factory refurbished product, resold repackaged defective or returned products, or included products damaged in shipping.

Having working with a local small distributor of computers, boards, and components, I've found that most returns have little to do with any manner of component failures, especially bulging caps, which usually fail well after the warranty expires. The main reasons for returns were "I didn't like it", "A better model was just released", "I couldn't make it work", or "I got a better price elsewhere".

I've also played forensic detective and tried to guess what caused various return failures for a manufacturer or networking devices. Subtle clues, such as a return without a power supply, usually means that the buyer lost the power supply, plugged in an unsuitable substitute, and blew up the device. Chronic overheating is fairly common and easy to spot. Corrosion caused by immersion or condensation is common and obvious. Static electricity blasts are difficult, but can sometime be seen as a lightning bolt pattern on an LCD panel. Overheating electrolytic can be found with a thermal camera, but in my limited experience, are rarely the inspiration for returning a product within the warranty period. Impact or drop damage is obvious and amazingly common.

The overwhelming number of returns work just fine and are often the result of RTFM failure. However, I didn't get to see any of those. Want to really reduce the return rate? Just design in a data logger to collect evidence of abuse and incompetence, or maybe add a built in lie detector for those that swear they read the instructions.

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
Santa Cruz CA 95060 http://802.11junk.com 
Skype: JeffLiebermann     AE6KS    831-336-2558

Good point, but I haven't any of that, or had any problems with substitutions. My 150 MHz scope isn't fast enough to see ground bounce, which is where I would expect to see most ringing. If possible, and before I replace any caps around the CPU, it put my scope on the power supply line and look for switching regulator noise. If present, I get to replace all of the electrolytics. (I have not needed to replaced any polymers yet). After the caps are replaced, I check it again. I don't have a minimum noise specification, but 50 mv p-p is my estimated maximum, but that's a guess because of variations in power supply quality, onboard switcher design, probe location, and coupling.

Incidentally, I've been replacing the usual 6.3v electrolytics found on the CPU VDD line (about 1.3V) with 4v polymers. This worries me a bit, but so far (about 25 caps), no surprises.

As you mentioned, ringing might be a problem. I don't see the PCB manufacturers using loose tolerance electrolytics in some manner of resonant filter circuit. I also don't see them using the variations in ESR with temperature to some advantage: I suppose I could add a small resistor in series with the capacitor string to make it look more like cheap electrolytics.

Oh, I've had my screwups, like installing replacements backwards: I've been fairly creative when replacing electrolytics. Higher voltage ratings, "close enough" capacitances, and different lead spacings are common. Mounting the caps in different locations, such as on their side with longer leads, and on the circuit side of the PCB, are also common. I have to do these when the component hole size is too small to safely remove the lead without also ripping out the plated through hole. Incidentally, my worst problem is lack of sufficient clearance around component pads, which often results in difficult to see soldering shorts.

From my warped perspective, board manufacturers use polymer caps because the CPU's are getting small, the associated heat sinks are becoming larger, and there's not enough space for large electrolytics. They would also get too hot near the CPU. As speeds increase, the maximum trace distance between the CPU and filter caps decreases. Also, they can use fewer polymer caps than electrolytics.

I haven't done many polymer cap replacements. The few that I've done were all electrolytic substitutes and have all worked. But admittedly, I haven't spent much time and effort beyond looking at the VDD line to see if there are any anomalies. I run an overnight memory test (MEMTEST86 or MEMTEST86+). If it passes, that's good enough.

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
Santa Cruz CA 95060 http://802.11junk.com 
Skype: JeffLiebermann     AE6KS    831-336-2558

On a sunny day (Sun, 22 Mar 2015 10:20:14 -0700) it happened Jeff Liebermann wrote in :

I had an electronics repair shop for many years, and also did service repairs for some large electronics companies.

Some equipment has standard issues, you get those back in droves with the same problem. Manufacturers usually provide a standard note how to change it so it won't happen again, or some instructions what to replace (cause of problem and collateral damage). Electrolytics do go wrong, but bad soldering, especially in the tube days, or soldering failing due to temperature differences was an other regular.

These days I have had one bad switch mode AC / DC adapter with bad caps after the other... Those are throw away items for some I guess, OTOH I would expect something to work at least some years. But .. much of it is bad design. So, it seems electrolytics _are_ a frequent source of trouble...

I'd expect it in the 10s of MHz range, actually -- we're talking big fat capacitors and modestly low inductances (~10nH). Mainly, the ringing is against ceramics. Which depends on what you're looking at, but for computers specifically, there are usually arrays of 1210 ceramics beneath the CPU, along the memory and such. Which probably total on the order of

100-200uF, which is a fair challenge even for polymer ESR (sqrt(10nH / 200uF) = 7mohm), and unlikely to be a problem with electrolytics on hand. (Also, 1 / (2*pi*sqrt(10n*200u)) ~= 8.9MHz.)

Probably about right. Most controllers aren't even stable if the ripple is high (not directly so, but because high ripple indicates a short loop time constant), and all but the worst equipment will be made to such awful tolerances. Onboard might be more critical (low voltages, even more so), but ATX spec is something like 50mV. Maybe 100mV on the 12V supply, I forget.

AFAIK, both can be ran at ratings; electrolytics might simply be unavailable at those voltages?

I've got an old video card (2006 era I think, but it was pretty fancy back then) which is populated with SMT can and chip style polymers. At least, I'm fairly confident that's what they are.

Hard to tell what the architecture is, since it's a pretty dense multilayer, including blind vias. Sneaking suspicion they have +5 and +12 coming in from the extra power connector (Molex 1x4), dropping the 12 to 5 in some load sharing method, then using the 5 to power the board. There's a pair of converters at the back end, probably for the memory and/or VCCIO, and a pair at the front, probably for VCCINT. There's one 16V

180uF polymer, two 6V 470uF (8mm cans), four 4V 1200uF (12mm cans), and a bunch of chip types ("33 C41" and "330 e44", whatever they mean).

Oh nevermind, it's 2004 era. As I recall, it still puts modern mobile chipsets to shame. It was hot shit back when Doom 3 came out. :-)

Hmm, those are interesting... widely varying tempcos. Curious what brands/formulations have the more stable characteristic, and if it persists down to even lower temperatures. Larkin was looking for something like that a while ago.

Well, poop. :o)

Yeah, sometimes you have to do what you have to do... not to mention multilayer boards and such.

Could be. There's definitely a strong incentive to avoid polymers; they're expensive. And apparently guilty of price gouging, from time to time. Hard to argue with the low impedance though, and with core voltages lower than ever...

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com

AFAICT, the figures *excluded* "NTF" (no trouble found) returns.

"How is a product declared defective? There are two possibilities. Either a technician will consider through an exchange such as a phone call with a customer shows that the product does not work, or if there is a doubt, the product in question will be tested to validate the failure of said product.

Note that this implies the products *are* "defective". I.e., later figures don't have a category explicitly covering those items that were NTF. (presumably, any found are just omitted from the tabulations of specific failures -- i.e., total may not == 100%)

As to quantities:

"All statistics by brand are based on a minimum sample of 500 sold products and statistics by model have a minimum sample rate of 100, with the largest samples being tens of thousands of sales per brand and thousands for specific products."

So, this is just above "garage shop" production levels (e.g., 500 units sold) and a significant number (>100) of returns/defects to, hopefully, eliminate some of the noise from small sample size.

Yup. I think almost 50% of "short term" returns come from "I got impatient trying to make it work. It made me feel stupid".

I see failures at the far end of the bathtub curve -- just after warranty expires, etc. (or, perhaps LONG after!) I.e., something is "worn out" -- either from planned durability or unexpected.

It's not always the fault of the user failing to read the manual. Often, the manual is shit. Or, it may be 100.0% accurate -- and describe a product that is effectively unusable!

I can recall adding a "mode 99" to a navigation product for *fishermen*... just to save an extra button on a keypad overlay, etc. Of course, if that invokes some *advanced* functionality, then you can argue that a user will already be sufficiently proficient with the device to be able to deal with that idiosyncrasy.

OTOH, the remote on one of our TV's is so incredibly NONintuitive that it is frustrating to use even after prolonged exposure: "Didn't anyone ever sit down and actually *use* this thing? BEYOND just verifying that each button worked??"

Actually, it's good practice to design with one or more black boxes as part of the basic design. Esp given the availability of nonvolatile stores! How else would you be able to ascertain what was happening in the product prior to it arriving at depot?

(I suspect most consumer/SOHO kit just gets discarded rather than repaired/refurbed)

There is a dirty little secret about most filter caps in switched mode powe r supplies.

The value doesn't matter. The frequency is so high that if you replace a 4,

700 uF with a 47 uF that has low enough ESR it will work fine. I started do ing this on LCD TVs because apparently our clientele there would watch a sc rewed up picture, like have a bad Tcon or panel and not even think about ha ving it fixed until it is completely dead. And they don't tell you. And som e of them will claim it was not lke that before.

So I got 1,000 uF / 50 V that'll do almost anything temporarily, just to te st it. If the picture is not right then the job is blown off and that is th at. I don't even cut the leads, I just had my stack of a few caps and reuse d the same ones over and over.

And I don't care if they used 3 2,200 uF in parallel, that good old 1,000 W ILL make the set work. One time I ran out of 2,200 Uf and they had three of them in a bank, I had 3,300 uF and two of them add up to the same 6,600 uF , so they got what they wanted. But it does not have to be so. Close enough is close enough. Three 1,000s would have been fine as long as they are goo d caps.

My sister's PC monitor decided on bad caps. The major culprit was IiRC a 1,

000 uF / 35 V. This is at the house and I don't stock that much in the way of caps here. I put in I think it was a 100 uF / 160 V and it worked. Needi ng the monitor I threw it together and said "this is not going to last, but within a week or two I should have the right caps for it and then I'll put them in".

Well things happen and it got blown off. The monitor is still working after about four years ! Now that must have something to do with the fact that t he cap I put in was NOS from about 35 years ago. Apparently it CAN handle t he ripple current. It is noticably larger than caps of its rating today of course, and I mean by quite a bit.

Fact is, you can go a little higher in value but not too much because of th e surge at turnon. You can go lower but if you go too low and the leg of th e supply is in the feedback loop it can cause trouble. But 2,200, 3,300, 4,

700. 1,500, 1,000, don't worry about it. Do not go up to double the value a nd unless you know the circuit and don't go less than half the value. And d on't upgrade the voltage if possible. You can but I recommend against it.

This works for me. Back in the old days when you had a regular power transf ormer and rectifiers, those values were chosen to reduce ripple voltage. To keep the headroom for the regulators at the lowest specified line voltage. Caps in an SMPS are a whole different story. I had to relearn it.

The reason is that instead of working at 60 or 120 Hz, they are more like 1

00 kHz and up.

All this only applies to the secondary side on a usual run of the mill SMPS . In old audio equipment for example, it is better to stick closer to their values. Like 33s and 47s, they don't use lytics where the capacity matters much.

When you are talking about big PC mmotherboards and things like that, same applies almost. The thing there is to get the damn best caps you can get. I mean Digikey and sort by ESR and ripple current capacity. If the value is a little different it will be alright, see the little coil off the output o f those regualtors feeding maybe a bank of 6 2,200 uF / 6.3 V ? Well those can be 1,000 uF just as easily and do the job just as well, for just as lon g. Even when a bunch of processes start all the same time and the uProcesso r pulls the juice, the regulator will just respond faster.

In fact I do motherboards the same way. I just determine which caps are in which bank and put one in each. That WILL make it run, but the problem is t hat the fault may have corrupted the BIOS to a non-bootable state. The manu facturers sometime call it "hang permanently" which happens sometimes when a BIOS flash goes bad, like the power goes out or something. Like on one I had it clicking out the PC speaker and the LAN lights were blinking. I stuc k just a few caps in there and then it didn't do that anymore, but it still did not boot. The BIOS is corrupt. So by not cutting the leads and nor put ting in $25 worth of caps I know it is not a viable repair. I'll use that s ame cap in the next one. If it boots then I'll put in something more approp riate, maybe even what was in there originally. :-)

Funny though, back in the day when we didn't have a cap checker we took lik e a 47 uF / 250 V and just started bridging caps to find the bad one in a T V. In a solid state set, 250 volts is high enough and 47 uF is high enough to get an indication but low enough not to blow everything with inrush curr ent as it charges.

That was a LONG time ago... No wonder all the way it is now, I am old !

I cannmot see, I cannot pee, I cannot smell, I look like hell