A number of you expressed an interest in the outcome of my search for an alternative to the solid tantalums that we've had so many problems with, so here is an update as promised.
Whilst MLCCs looked promising at first, their susceptibility to failure during soldering and under vibration and thermal stress has effectively excluded them for our application.
I couldn't find polymer tants at a higher working voltage than 25V, so the choice came down to either solid polymer aluminiums or Multi-Layer Polymers. Based on data gleaned from the net, I've opted for MPLs, subject to us being able to procure them to the requisite spec. These parts claim to have very low esr, formidable ripple-current rating, 'true' voltage rating that doesn't require derating to achieve reliability and no aging mechanism.
As I stated earlier, the performance of the tantalums in our application is dominated by their esr limitations, (110m-ohms), and we can achieve the required performance in our application with a total capacitance of only 100uF if the esr of the replacements is sufficiently low. (Quite an interesting calculation involved to derive that - peak dV/dt ripple doesn't occur at peak current).
Paktron's 10uF/50V CS4 Capsticks have an esr in the low m-ohms at our operating frequency and occupy slightly less board area than the tantalums they are to replace, so we will have room for ten of them to give the required 100uF. (We shall keep some, if not all, of the ceramics in parallel to handle the higher frequencies - as I have already said, we have had very little trouble with them). I've been in touch with Paktron in the States and they are sending some samples.
On a general note, tantalums have been around for decades and have always been used for supply filtering/decoupling in the military environment where significant amounts of capacitance is required, so I find it somewhat surprising that tantalum manufacturers haven't sorted out the problems with their reliability by now. The problems we have fallen foul of can't be peculiar to us, so I hope some of these discussions have been of benefit to others out there.
I managed to forget to mention one of the most important features of the MLPs - they don't fail short-circuit. In a system where there is a number of parts in parallel, this is a very significant advantage.
I was attempting to repair a board which included a little DC-DC converter with a tantalum cap on the front end, which had been mistakenly connected to
28V instead of the expected 12V. I was placing DMM leads on various points while the program manager bent over my shoulder, closely examining my technique. Suddenly one of the tantalum caps lit off, and burned like a matchhead to completion. I was more started by the shout in my ear than by the smoke and fire.
Sure, if you pump enough energy into any part you can get it to explode. But an MnO2 tantalum has explosive power gain; millijoules of energy can start a chemical reaction that blows it to bits. A wet-slug can only deliver as much explosive energy as the circuit pumps into it.
A note on wet slug capacitors: Early wet slugs had siver cases. A common failure mode was silver migration, in which silver from the case would migrate into the electrolyte, eventiually causing a catastrophic short circuit. Wet slug capacitors with tantalum cases are now available. These capacitors do not exhibit the migration problem, and are very reliable, but they are very expensive. . Another recent alternative to tantalum capacitors is niobium. These capacitors have characteristics similar to tantalum. They are available in both solid and wet slug configurations. They have low ESR. High reliability versions (0.2%/1000hr) are available.
I consider a small explosion to be classed as a "nasty failure mechanism". Witnessed in final testing of an airliner autopilot some years ago.
Subsequently witnessed en masse with some military computers on final Environmental Stress Screening when some idiot connected them to 200V Ph-Ph instead of 115V Ph-N.
In both cases a lot of damage was done to surrounding circuits, PCB's and chassis.
Aren't the low voltage (50-100vdc) part ratings limited to 85degC operation? The derating for voltage above this temperature is something like 50% at 125degC, so that's around 37V @105deg.
Your problem is that you're trying to retrofit an smd package size provide by an original device, the suitability of which was not practically proven in the application and environment specified.
Wet tantalums in hermetic tantalum axial cases. larger radial ceramics and axial film caps are typically used in mil and aviation power conversion devices. Sometimes they require deviation clearances, where no standard mil-grade devices are available. These will be physically larger than the parts presently used in your circuit and mechanically incompatible to the physical layout presently available there.
Your other alternative, of course, was to identify the fault-inducing stress and to prevent it from occuring.
Patient - "Doctor, my caps blow up when I do this." Doctor - " Don't do it."
Changing part package sizes or actual circuitry were not considered viable options of your retrofit. Testing either option could have gone part way towards confirming the character of the fault-inducing stress mechanism. Currently this is only an assumption.
Based on this assumption, you've decided to replace a moderate-ESR large-C filtering component with a low-ESR small-C one - something that may invalidate a signifigant proportion of your previous design verification record.
Add-on circuitry or similar part type substitutions need not do this, particularly if their failure modes simply duplicate previously known conditions (series dv/dt limiters become short etc)
Yes, the 50V parts derate at 1.25%/deg C above +85C. Ours is a cockpit application, so upper operating temperature is +70C. Allowing for internal heating, this brings our maximum operating temperature up to about +80C, so I don't consider this an issue.
(For the record, the SFE parts are through-hole). Regarding the suitability of the devices for our application, they are specifically aimed at the area in which we have used them and, amongst other things, their technical spec states:
"In the past problems have arisen with the use of solid tantalum capacitors in low impedance circuits only obviated by the use of protective resistors in series. New techniques, burn-in and test procedures enable the SF and SFE ranges to be used confidently without such protection, thus considerably increasing circuit design possibilities".
As is so often the situation in military designs, space is at a premium, so we do not have the luxury of using some of the more bulky, but more robust devices that are available - a colleague has identified some 'Supermatallized Polypropylene' parts that look ideal, but we would have to have a separate bolt-on box to put them in.
We have investigated this problem ad-nausium and much of our investigation is described in my original thread.
Numerous failed parts have been returned to Arcotr "There is no apparent common failure mode either by date or region of the internal anode where degradation/short-circuiting occurs. The application appears benign in terms of the environmental/electrical stress on the component, both voltage and temperature are well within parametric maxima."
The parts are comfortably derated in terms of operating voltage, being 63V parts running at nominally 28V and our operating temperature range is within the +85 deg C at which voltage derating starts to apply.
The one area that might have been considered marginal is ripple current rating: 2.7A rms per cap @ 20C, derating to around 1.75A rms @ +80C.
Our peak ripple current requirement is a 20A rectangular current waveform, so at +80C, nine caps sharing this peak curent would indeed be overated - and that's without taking current-sharing into consideration.
However, the issue with ripple current is entirely one of heating within the cap. Our peak requirement has a maximum duration of around 10 to 15ms, with a background current ripple less than 0.5A between the 9 caps.
The absolute maximum, (and operationally bizzare), duty cycle for the peak requirement would be of the order of 2%.
Furthermore, not one of our failures has occured during the high ripple current events, but have always been associated with testing at elevated voltage, albeit still comfortably within the voltage rating of the parts. We have also taken steps to reduce inrush current during supply voltage transitions, but with no noticeable reduction in the rate of failures.
The point at which I lost faith in the SFEs was when we tested a batch of unused parts by charging to 60V through a 10k-ohm resistor.
Short of leaving the damned things in their box, I can't think of a much more benign test regime, but around 2% of the parts tested, (which incidently is about the same as the failure rate we get in use), went spuriously short-circuit at some point during the charging and then recovered and recommenced charging up to the 60V. On several units, the shorts occured numerous times at progressively higher voltages until the unit finally charged to 60V. Once these units had charged to 60V, the fault could not be repeated. (I have numerous plots of these failures, but they are far too bulky to stick on here as an attachement).
We have since had Arcotronic introduce additional screening along similar lines, but again with no discernable improvement in failure rate during use.
The 'D' case is the largest package in the SFE range, but we don't have room for a larger package anyway, so I'm obliged to use a part that will fit in our limited space.
I submit that, on the basis of all the evidence, it is the Arcotronic SFE parts that are the problem, not the manner in which we are using them.
The issue with the supply reservoir was always one of impedance to the nominally 80kHz, bi-directional, rectangular PWM current waveform, the requirement being to reduce supply voltage ripple within the equipment to a level at which the series inductor in the +28V line can keep exported current ripple on the 28V supply to within prescribed limits.
As I have stated at several points in the original thread, it was the esr of the SFEs that dominated the supply voltage ripple due to PWM current, not the dV/dt component, therefore a part with significantly lower esr will not require as much capacitance to achieve the required result.
We will obviously conduct tests to confirm this on the bench, but I believe I can put forward a solid case for Qualification by Anaology with the previous design on this count.
Not sure what you're getting at here. I'm not proposing that we introduce add-on circuitry, (although I was thinking about it at one stage when we considering using the Solid Polymer Aluminium parts, with their o/c failure mode). On all the evidence available, the MLP parts do not suffer anything remotely like the failure modes we've seen with the SFEs.
That said, since posting this update, I've received an e-mail from someone who's been following the thread, indicating that there are problems specific to the proposed MLPs under some environmental conditions that may well mean we can't use them. I'm currently looking at a suggested alternative manufacturer's MLP which does not have this problem. (If anyone wants further details, contact me direct by e-mail).
Thank you for your observations and comments - usefully thought provoking. It's always a good idea to re-evaluate earlier decisions in a situation such as this.
-- Ted Wilson Senior Circuit Design Engineer BAE Warton
snipped-for-privacy@baesystems.com
"The gods do not deduct from man's allotted span the hours spent in fishing."
If ripple export is the concern, just a thought: Do you have headroom in the thermal and power consumption budget? If so, maybe you could use active ripple cancellation. Essentially these would be transistors that add load during phases where the unit doesn't draw full load. I know that this wastes energy but it has worked nicely in applications where conducted noise was the issue. In our case it was mostly noise caused by RAM banking.
We're limited by a mean supply current requirement, so we don't have a lot of headroom when it comes to wasting current and, on the face of it, with
20A peak winding currents, this approach does not look viable.
If I were starting again from scratch, I would have a long hard think about possible alternative approaches, but that's not an option here - all I have room to do is find ways of making the existing design more robust, whilst in the process of addressing several obsolescence issues.
That said, I've reviewed the design many times now, in the light of these problems, and I don't see anything wrong with the approach I have adopted. We have passed all our qual testing and all that is required for a quiet life is for the reservoir caps to do their job without failing.
If only life were that simple! :-/
Regards
Ted
-- Ted Wilson Senior Circuit Design Engineer BAE Warton
snipped-for-privacy@baesystems.com
"The gods do not deduct from man's allotted span the hours spent in fishing."
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.