Film Capacitors

Intuition

A Swiss roll of plastic and metal

Cut a Swiss roll and the section shows what it is: two thin layers rolled together, an enormous amount of surface packed into a small cylinder by winding rather than by stacking. Nothing in it is thick, and every part of one layer is close to a part of the other.

A film capacitor is built that way and for the same reason. A plastic film a few micrometres thick is the dielectric, a metal layer is the plate, and the whole thing is wound into a roll or flattened into a slab. Two ways of doing it exist. Either the metal is evaporated straight onto the film in a layer tens of nanometres thick, which is a metallised part, or separate aluminium foil is wound alongside the plastic, which is a foil part. Everything that separates one film capacitor from another comes from those two choices and from which plastic was used.

What the family is good at follows from the plastic. It holds its capacitance across temperature and it does not care about applied voltage at all, which is what a class 2 ceramic cannot say. It loses very little energy per cycle. It has no polarity. It does not dry out, so it has no endurance rating in the sense an electrolytic does.

What it is bad at is being small. A 470 nF film part rated 400 V is a brick beside a ceramic chip of similar value, and above a few microfarads the family stops competing on size at all. That is the whole trade, and it is why film capacitors are chosen deliberately rather than by default.

Practitioner

Fast edges, and the current they draw

The specification that appears on a film capacitor's datasheet and on almost nothing else is a limit on how fast its voltage may change, quoted in volts per microsecond. It looks like an odd thing to limit until you write down what it costs.

The slew-rate limit is a current limit in disguise, and the line is straight through zero.

That is why the two constructions have very different ratings. A metallised plate is tens of nanometres of evaporated aluminium and the current has to reach it through a sprayed end contact; a wound foil is a real sheet of metal with a real tab welded to it. Foil parts take the fast edges, cost more and take up more room. Metallised parts do everything else, and they bring one property foil parts do not have.

Engineer

The fault that repairs itself

What happens when the film is punctured

Every dielectric has weak points, and a large enough transient finds one. In a foil capacitor a puncture welds the two plates together and the part is a short. In a metallised capacitor something else happens.

The cleared radius follows from its area. The fault burns its own metal away and the part carries on.

The plate is only 30 nm of aluminium. When a puncture forms, the discharge current concentrates at the hole and the metal around it vaporises before anything else can happen, in microseconds. The fault is isolated by a small clear ring, the short disappears, and the capacitor keeps working with a little less plate area.

How much less is worth computing. Reaching 470 nF across 8.0 µm of film with a relative permittivity of 2.2 and a vacuum permittivity of 8.854 pF/m takes 0.193 square metres of plate. Clearing an illustrative 1.0 square millimetre of it removes 2.43 pF.

It would take 1.93 k clearings to lose one per cent of the value.

It would take 1.93 k clearings to cost one per cent of the value. So a metallised part does not merely survive a puncture; it survives thousands of them without measurably changing. That is why metallised film is what goes across the mains in a safety capacitor, where failing open is the entire requirement.

Which plastic, and what it decides

Three plastics do most of the work, and they are not interchangeable.

Straight lines, and one of them slopes the other way. That is what separates film from class 2 ceramic.

Polypropylene has the lowest loss and the highest voltage capability, and its coefficient of -200 parts per million per degree runs negative, reaching -1.20 % by 85 °C. It is what a resonant tank, a snubber or a mains-frequency part is built from.

Polyester, usually called PET or mylar, is cheaper and smaller for the same value, and pays with 600 parts per million per degree, or 3.60 % over the same rise, and considerably more loss. It is the general-purpose choice.

PPS holds close to flat, at an illustrative 100 and 0.60 %, and survives reflow soldering, which the other two struggle with. It is the surface-mount film part.

A factor of 125.0 between the quietest film and a class 2 ceramic.

The loss figures separate them further. At an illustrative 0.020 % for polypropylene, 0.050 % for PPS and 0.50 % for polyester, against 2.5 % for a class 2 ceramic, the spread between the ends is a factor of 125.0, and in a tuned circuit that ratio is the difference between a sharp resonance and a soft one. ESR is where the loss mechanism is taken apart properly.

Professional

The charge that comes back

A film capacitor charged for a while, then shorted out and released, does not stay at zero. Over the following seconds a voltage reappears across it, and the part charged itself from nothing.

The charge was never removed by the short, because it was not on the plates.

The mechanism is dielectric absorption. Some of the charge does not sit on the plates; it sits inside the dielectric, held by polarisation that relaxes slowly. A short removes the charge on the plates in microseconds and does nothing to the charge in the material, which then reappears at the terminals over seconds or minutes.

At an illustrative 0.050 % for polypropylene, a part charged to 400 V recovers 200 mV after a short, settling over about 30 s. Polyester, at 0.20 %, recovers 800 mV. Both are small, and both are enormous compared with what a sample-and-hold circuit or an integrating converter can tolerate, which is why polypropylene is specified in those places and nothing with a ferroelectric dielectric is.

There is a safety consequence too, on parts large enough to matter. A big film capacitor in a mains-connected assembly, discharged and left, can present a real voltage at its terminals a minute later. Testing capacitors treats proving a part dead as a procedure rather than a single measurement, and this is one of the reasons why.

Where the family earns its volume

Four jobs go to film almost automatically.

Snubbers and pulse work. Fast edges, large currents and repeated stress. The slew-rate rating is what is being bought, and the self-healing is what keeps it alive.

Mains-frequency AC. Across the line, in motor circuits, in power-factor work. A film part tolerates continuous alternating voltage with very little internal heating, which is exactly what an electrolytic cannot do. Motor run capacitors are this application.

DC link and energy storage where life matters. Film has no wear-out mechanism in the electrolytic sense, so an assembly meant to run for twenty years without service uses film and pays in volume.

Anywhere the value has to be a value. Filters, timing, integrators, oscillator tanks, audio paths. No bias dependence, small and linear temperature drift, low loss and low absorption, in that order of importance depending on the circuit.

What film should not be asked to do is bulk decoupling or supply reservoir work at high capacitance. Above a few microfarads the part is large and expensive, and the job wants an electrolytic with a ceramic beside it. Choosing the right capacitor sets the four families against each other in order.