Capacitor mounting methods (lead length) vs performance

[amberwolf]

This thread is to discuss (and test) the difference between mounting the main electrolytic capacitors in controllers vertically vs horizontally, relative to the PCB, so that their leads are the shortest possible, causing the least resistance (and inductance) path between the PCB power planes and the capacitors' internals, enabling them to more easily do the job they're there to do--fill in for the battery during the instantaneous switching loads of PWM motor control.

I started the question over on this thread:
http://endless-sphere.com/forums/viewtopic.php?f=2&t=13638&start=15
and am copying my original comments plus a couple of relevant quotes from others here in this opening post.

If there's any way you can seat those main filter caps vertically on the board, you should try it. (yeah, I know they won't fit in the case if you do this, unless you drill holes for the caps to stick out)

Those cap leads are almost certanly only steel, plated with copper and tinned, so they don't conduct electricity *or* heat very well compared to the copper traces on the board. At high power draws in the controller, having the caps' leads a fraction of what they are now should make a difference to performance.

If you can get caps that have formed copper/alloy leads not made of steel, it'd be even better. Or at least ones with thicker plating on them, and thicker leads overall.


When I have tested my 2QD brushed controller with various cap configurations, it performs better under heavy loads with the cap vertically right on the board with the leads as short as possible than mounted as I typically see in photos of ebike controllers, sideways, parallel to the board with long leads down to the PCB.

The waveform at the controller output is smoother, a bit, using the same size cap with vertical mounting, at the higher current draws. AFAIK it's because the cap leads aren't creating as much resistance when they're shorter, so not preventing them from doing their jobs as much.

I first even looked into this because of info at the 4QD site about it, and finding that under some conditions it's actually possible to heat the cap leads enough to melt the solder holding them into the board, when they're that long (hasn't happened to me, but did at 4QD to the designer during early versions of his controllers).

I suppose a quick hack solution leaving the caps in place would be to solder some fat copper wire along the leads. Kind of like how people sometimes build up the traces on the boards...

Doc has tried that. I dont like it for a few reasons. One reason is that it makes it so that you cant move the caps around anymore - makes them stiff which makes them brittle. It is also a lot of work and somewhat risky and difficult in the tight confines.

I suspect that it is the inductance of the long leads more than the resistance that affects performance.

Solutions include mounting smaller caps vertically...
I suppose I could mount 470uF caps vertically and then try to jam a few 1000uF in there to make up the total difference.

The reality is this is now I specified the controllers to come in so I probably wont be re-working them.

The way I see it there are ceramic caps in parallel and I would rather have 4,700uF of slightly higher ESR electrolytics than 2,000uF of lower ESR electrolitics.
I think the ceramics in parallel really help.

Good suggestion though. Maybe someone can do a test and if the results are astounding (like 5% or 10%) improvement then we can think harder about it.

Another thing to consider is using smaller uF caps, but more of them, so that the current flow in and out of each one is less. They'll heat up less, and so will the leads. It will still matter for the leads to be as short as possible, but less than if all that current has to flow thru fewer sets of leads.

Of the few larger controllers I've looked at, such as my Curtis 1204, all use multiple smaller caps in parallel, mounted vertically to the board. Some glue the caps down to the board at the base of each cap, and also really fit them tight against the PCB so they can't flex from vibration. AFAICS the Curtis just has glue between caps, but there are so many that they probably damp the vibrations enough to not worry about it much, plus the whole board bottom is secured to a couple of bus bars that help prevent the PCB from flexing (along with the twin heatsinks the MOSFETs are on).

I'm fairly sure it's not a cost thing, because most of the time when I've looked for low-ESR caps to replace blown ones in various power supplies, bigger ones don't cost enough more than little ones to justify using a lot of little ones unless there is some engineering reason to do so.


I'm not sure I have the capability to do the kind of testing needed to show the difference between both ways, plus my controller is just a brushed DC unit, and so would be significantly different in operation than these BLDC controllers (though I suspect not that much different in actual results between the two types for this issue).

If I do manage to be able to test this, and find anything different between the two methods, I'll post my results in this thread. Anyone else willing to methodically test this is welcome to post their results here as well.

If any kind of mechanical problems result from the vertical mounting, I'd like to see those reports, too.

My own (modded) 2QD has only the single cap on it right now, vertically mounted flush to the PCB, with the cap partly outside the controller box and tightly held in place by the hole it passes thru (is a good friction fit, so no vibration movement of the can).

 

[amberwolf]

Also from the above thread:

I agree that multiple caps in parallel would lower the ESR significantly.
That would really need a new board spin though.
Too bad I dont speak Chinese or else I would suggest to them exactly that.

Yeah, I guess we'd need to make our own PCBs for that or something. It'd cost more for them to make them with multi-small caps, so it's unlikely they'd go that route since they don't actually *need* to, to deliver a functional controller.

FWIW, you could send the suggestion to them in English and there's probably *someone* there that reads it well enough to figure it out. Might require drawings, though.

After I get the rewound cieling-fan back from Karma (still waiting on enough money to ship it to him first), then I'll have incentive to build a BLDC controller to test it with. Maybe I should take those images posted previously of the bare board, and work on creating an Eagle layout for them that uses many smaller caps, vertically-placed. Catch with doing it that way is the board isn't probably going to fit into the same casing. I'll try to keep the same outline, but not sure if it's possible.

If I do get it done, I could test it out if I could get someone to send me a fried controller's IC chips (as long as they work) so I could transplant them to a test PCB. Well, first I'd have to succesfully make a correctly-registered-and-drilled double-sided PCB, which I'm not entirely sure I can do. I'm sure I have 18 FETs of the same type around here, although they're not as good as the 4110s, and I don't remember what they actually are.

 

[ZapPat]

I don't think that the small bit of extra cap lead length will make a huge difference in these particular controllers. As Methods mentions, there are film caps in parallel with the electrolytics that handle the high frequencies and these are the most important ones to have close to the top FET's drain and the bottom FET's source (across the + and - power buses). These high frequency caps would suffer much more from having long leads and/or from being far from each phase's FETs. Electrolytics already have terrible inductance from their construction, so a bit more doesn't matter nearly as much for these. We also have to consider that these ebike controllers don't have super-fast PWM switching anyways, and this is why they can get away with their mediocre PCB layout in the first place. I also don't think the caps leads are made of steel, but more likely tin-coated copper (they are quite soft and only slightly magnetic), but I'm not sure of this.

 

[fechter]

The ESR of the capacitor at the switching frequency is going to be substantially more than the resistance of the leadwires in most cases, even if they are like an inch long. What seems to work best is to use many smaller capacitors in parallel rather than one large one. This gives you a more compact profile and all the lead resistances and ESRs are in parallel, so the effective ESR is better. The caps I tested here (1000uf, 100v) measured around 18 milliohms at 20khz. I think this is fairly typical for aluminum electrolytics. Inch long leadwires might be another milliohm or 2.

A combination of aluminum electrolytics and multilayer ceramics might improve things at higher frequencies.

 

[liveforphysics]

Picking the right cap is the biggest factor. When I was researching for ULESR caps, I found some pretty awesome stuff out there. Often at $15-40 per cap though, but I found some single 220Mfd caps with lower ESR's than all 4 of the 1000Mfd caps used on the standard infinion controllers. All the ULESR caps come with robust legs standard.

 

[amberwolf]

dnmun says in the other thread that it was the inductive impedance rather than resistance:

notes about the capacitance thing. the inductace is a function of di/dt so the less current flow changes and the longer the time it takes to change reduces the back emf generated in the wire, so the voltage spike is reduced too.

the inductance relationship to length of a cylindrical conductor is exponential and is 2 orders of magnitude higher within 2 cm as i recall. shoulda saved that graph.

that is why the stand them on on the pcb and keep the legs short. it is not the resistance, it is the empedance from the inductance that is the reason.

With all these varying replies, I'm more curious than ever how much difference it actually would make, especially after the physical problems seen by 4QD with longer leads on brushed controllers, yet obviously not having much, if any, problem with them on these brushless ones.

I suspect that the wideband RF generated by the brush arcs is causing a lot more inductive heating of long leads like this than the simple PWM harmonics do in the brushless controllers.

Another possibility is that the frequency changes of the brushless controllers make a difference as well, while the fixed frequency of the (majority of) brushless controllers happens to be at the "right" frequency to cause a problem with some lead lengths and/or lead diameters. This is less likely than the RF theory above.


As for the leads being steel, AFAIK these days the leads on pretty much every electronic component tend to be steel that is plated in copper and then tinned in some fashion. High-performance components like ULESR caps might be an exception to that. Some of the high-current MOSFETs I have used are an exception, using a solid copper alloy of some type instead.

It's easy enough to tell: just pass a magnet over the trimmed lead bits leftover from your installed components--if it picks them up they're almost certainly steel-cored. It's how I usually clean up after doing a board-stuffing.

 

[rhitee05]

A few comments:

What dnmun says is right, it's more about lead inductance than resistance, even for a (relatively) crappy conductor like steel the resistance of 1" of leads is trivial. Inductance is a bigger issue, especially at higher PWM frequencies and harmonics. I believe your theory about the higher RF harmonics of the brushed motors is probably valid and would partially explain why lead length seems so critical.

The best approach is to use multiple caps, rather than one huge cap. A good approach for brushless controllers might be to put one cap as close as possible to the FETs for each phase. An even better approach is to use multiple caps in a range of values. This is standard practice for higher frequency applications, and a less-intensive version should work well here. The idea would be to supplement the big electrolytics with one or two smaller values of capacitors, say a 10uF and 1 uF in a lower-ESR technology like ceramic or tantalum. Those smaller caps will lower the overall ESR at mid-to-high frequencies. A common technique is to use values spaced a decade (10x) apart to get good low-ESR performance across a broad range of frequencies.

 

[ZapPat]

Picking the right cap is the biggest factor. When I was researching for ULESR caps, I found some pretty awesome stuff out there. Often at $15-40 per cap though, but I found some single 220Mfd caps with lower ESR's than all 4 of the 1000Mfd caps used on the standard infinion controllers. All the ULESR caps come with robust legs standard.

I found some caps that sounded like the ones you describe here, Luke, quite expensive and low capacity, but very low ESR. I thought I had the best solution... Then I found out about surface mount MLCCs, and now I wouldn't consider *not* using them in PWM switchers anymore. And did I mention they are really cheap, have massive ripple current ratings for their size (meaning very low ESR/impedance).

Just one of these tiny chip caps will eat up higher frequency noise and provide a very rapid current source to each phase's power connections. The noise generated by the FET's transitions from state to state is actually of much higher frequency than the 20kHz PWM frequency. The FETs change state in tens to hundreds of nanoseconds, but lets say 300ns as we may see in typical ebike controllers. This means a frequency of about 3Mhz for the large pulse of energy that is needed during the FET state change, plus some higher harmonics too no doubt. Ceramic caps with very low inductance are very good for this, but they have to be *very* closely placed across each phase to be uselfull. If they are too far, even fairly small lengths of PCB traces will effectively isolate the caps because of the inductance between them.

The larger electrolytic caps can be further away from the FETs as these caps handle pretty much only the PWM frequency itself, much lower (~20kHz), but also needing more actual bulk capacitance than for the HF switching events. And since the frequency is about 100 times lower, PCB and lead inductance is less of a problem (thankfully).

As others have mentioned, using multiple smaller caps in parallel is very good, but only partly because of the lowered ESR and lead inductance. The optimal place for caps is right across each phase, or at least not too far away for the bulk caps (electrolytics), so having three banks of caps in a BLDC controller is a good idea or else the PCB trace's inductance becomes a limiting factor. User rithee05 had a good suggestion of putting different values of caps in parallel to handle a wide range of noise, *but* when doing this it is essential that the smaller, higher frequency caps use smaller and lower inductance packages than the larger caps, or else there is no benefit.

 

[dnmun]

the reason for the large electrolytics on the input is because that is where the largest current surges take place, not just that the legs of the electrolytics need to be short, but the entire conductor is generating the back emf when the current changes. that is the reason for the caps to begin with, to reduce the current fluctuations in the main power lead.

also there are ceramic high frequency caps right on the S/D busses, and then there are larger electrolytics to help smooth out the current spike on the S/D busses.

so the length of the legs on the electolytic is secondary to the length of the conductor they are attached to, but the capacity and ESR of the electrolytic can help to reduce the size of the current change in the main conductor, and spread out the current change over time, both of which reduce the back emf generated by the inductance di/dt. the lower ESR is most valuable in higher frequency switching where there can be significant heating of the capacitor by the fluctuating charge flowing in and out of the cap.

all imho.

 

[rhitee05]

ZapPat,

Your analysis is right on. The extremely high frequency ringing from the FET switching is a major issue and ceramic caps, especially surface-mount MLCC's, are great for this. Placement becomes much more critical the smaller the cap value gets. If the board layout permits, one great technique is to solder an appropriately-sized MLCC cap directly between the traces as close to the FET as possible. This has very low parasitic inductance and will work well for higher-frequency bypassing. Electrolytics by themselves really won't do anything for this high-frequency energy.

One of the major cap manufacturers, Kemet, has a neat little tool you can download from their website. It has SPICE models for most of the capacitors they make, mostly ceramic, and tantalum. The plots show ESR and overall impedance vs. frequency so you can see how a particular capacitor will behave. All caps have a resonant point where the internal inductance takes over and it stops behaving like a capacitor, so it's helpful to know where that is to choose the right values and packages. It's neat to see the effects that package size and even dielectric material have on the frequency response. Can be found here:

http://www.kemet.com/kemet/web/homepage/kechome.nsf/weben/kemsoft