MLCCs: Acoustic noise and how to solve

How often have you heard about 'singing' capacitors? Have you ever encountered one? Don't worry, it's quite normal: if you are sure that your circuit works well and you are not exceeding any maximum absolute rating of the component you should look for a solution that only solves the acoustic problem, with no impact on the electronics and its performance.

There are mainly two types of noise generated by capacitors: clicking noise and buzzing/humming noise. The first one happens when a fast voltage transient is applied to the terminals of the capacitors; the second one instead is caused by an AC voltage superimposed to a DC bias potential difference. Clicking noise can be heard especially on supply capacitors when the power supply net they are mounted on is enabled: in many cases clicking is made inaudible by an inrush current limiter, nominally used to reduce the current on the bulk capacitors at the system power on, but, in this case, helps from the acoustic point of view too. Buzzing/humming, instead, is a permanent phenomenon once the system is supplied. To understand why capacitors actually 'sing', let's take a look at their structure and involved mechanism; keep in mind that such a noise is generated by oscillatory/vibratory motions which are typically in the pm-nm range, so visually imperceptible.

Capacitors are typically made of a dielectric material and two electrodes in a single sandwich or a multi-layered fashion when higher capacitances are required. This is not an article about capacitor structures but here a quick look is mandatory.

Electrolytic capacitors are usually made of three rolled-up sheets: cathode, anode and dielectric foils. The dielectric is a thin oxide strate on the cathode while the paper is meant to keep a certain amount of liquid electrolyte (the actual anode) in strict contact with the dielectric and the anode metallic plate. This structure is then encapsulated in a metallic (typically aluminium) box and sealed at the bottom with a rubber plug. The internal roll is semirigid and this means that movements between armatures and other motion can be mechanically absorbed or damped.

MLCCs are

If a capacitor has semirigid armatures, like an aluminium electrolytic capacitor, it can vibrate when subjected to a particular stress, especially if they are not stacked with a solid dielectric like in MLCCs. In this case, vibration is mainly caused by the electrostatic force between the anode and cathode. The only way to remove it is to reduce the ripple voltage across the capacitor terminals. If you are experiencing a high ripple on the main supplies, maybe you don't have only noise problems but also insufficient input capacitance in your system. If you increase the capacitance by adding other bulk capacitors in parallel you will improve the filtering on the supply line, reduce the thermal stress of the single capacitor (the total RMS current will be partitioned), eventually improve EMC and finally, you will see that the noise decreases a lot. Let's do some math, imagining an aluminium electrolytic capacitor in which the two metallic armatures vibrate under an AC voltage and the paper layer partially absorbs such amrature motion. Under static conditions (zero ripple voltage across the capacitor) we have: $$Q=C*V , C=\frac{\varepsilon A}{D} \rightarrow C=\frac{\varepsilon A}{D}$$ $$F_{es}=\frac{\Delta V Q}{2 D} = \frac{(\Delta V)^2 \varepsilon A }{2 D^2}$$ where F_es is the electrostatic force, A is the capacitor area (assuming cathode and anode have the same), and D is the distance between the armatures (so it is the dielectric thickness). Now, considering that the dielectric is a soft elastic material (no plastic/permanent deformation - typically paper), then we can use the Young modulus of the paper to extract the change in D when an AC+DC voltage is applied on the capacitor: $$F_{el} = k \Delta D = \frac{EA}{T_0}\Delta D$$ where E is the Young modulus, A the capacitor area, T₀ is the initial paper thickness under zero tensile or compressive stress. $$F_{es} = F_{el} \rightarrow \frac{(V_{DC} + V_{AC})^2 \varepsilon A }{2 d(v(t))^2} = k d(v(t))$$ Let's linearize (first order); V is V_DC and v is V_AC*sin(2pift); v(t) is V_DC+V_AC*sin(2pift), considering a sinusoidal AC component (usually the AC voltage is full of harmonics and this could lead the mechanical system to amplify some of them and attenuate others). $$\frac{(V^2+v^2+2Vv)\varepsilon A}{2 d(v(t))^2} = k d(v(t))$$ $$\frac{(V^2+v^2+2Vv)\varepsilon A}{2k} = d(v(t))^3$$ $$d(t) = \sqrt[3]{\frac{(V^2+v^2+2Vv)\varepsilon A}{2k}} = \sqrt[3]{\frac{(V_{DC}^2 + (V_{AC}sin(2\pi f t))^2 + 2V_{DC}V_{AC}sin(2\pi f t))\varepsilon A}{2k}}$$ Now let's put some numeric value to understand the entity of the armature oscillatory motion

Are 5nm to 10nm enough to generate noise? Yes, of course! Consider that a 10pm displacement can still be heard by a human ear with perfect hearing; this value can be even higher in the case there are other conditions damping the sound. The surrounding environment could also attenuate frequencies and amplify others, so 10pm is a rough reference value.

Please note that ripple voltage reduction may work in many applications and circuits but there are other circuital configurations in which the other parallel capacitors could resonate with the first one because they have mismatches and different non-idealities! This could happen especially with capacitors with a high capacitance change with respect to applied voltage!

Another acoustic noise source could be the piezoelectric effect; some capacitors are affected because of their technology and the dielectric materials used in their construction. Mainly MLCCs (especially Class II) and Tantalum capacitors can generate noise when used in switching or AC circuits, so pay attention when selecting the right technology for your design.

Could PCB layout cause or worsen an acoustic issue? It can worsen or amplify the noise but it cannot be an acoustic noise source: PCBs could resonate with the noise source or they can even amplify the noise itself, acting exactly like a loudspeaker vibrating membrane.

Two types of solutions can be found when trying to remove the noise generated from a capacitor: the electrical one and the mechanical one. Typically one wants to solve a mechanical problem using mechanical methods but this is not always possible, so, other solutions are listed below: they can be appliable or not, depending on the type of circuit, board design, technology used ...

Leaded capacitors are good for noise reduction since they are soldered on the board through leads; you will not reduce the effect by working on the noise source but reduce the 'speaker' effect due to the proximity of the capacitor to the PCB. If you are not using PTH components, you'll not be allowed to use such capacitors.

Megacap are just more than one MLCC soldered one on top of the other, sold with two very short leads. They can be easily soldered on a PCB through an SMD process. The principle behind the use of a megacap is the same as leaded capacitors: lifting the MLCC from the board and connecting it through longer leads you can decouple the component from its sounding board (PCB).

Soft termination capacitors could be a good theoretical solution but if the movement happens directly inside the body of the capacitor, mechanical solicitation is too fast or high in modulus, and the internal structure of the MLCC leads can't absorb the movement. Soft termination is thought to absorb slow movements and (partially) mechanical stress due to PCB deformation (PCB bending due to a not perfect installation or vibrations which can happen during component operation). Again, you will not reduce the noise acting on the source but reduce the coupling between the source (MLCC) and its speaker (PCB).

Electrolytic capacitors have moving parts inside but are typically larger and heavier than the ones of an MLCC; the noise will be reduced in this way thanks to the technology of the component.

Polymer capacitors are not affected by piezoelectric effect at all so you will not experience any (at least) NVH issue when using them.