DRSSTC design guide updated with MMC design guide

It only took me a mere 4 years from I first started this article about the MMC until it is now done for public released 🙂 …


WARNING!: Working with electricity is dangerous, all information found on my site is for educational purpose and I accept no responsibility for others actions using the information found on this site.

Read this document about safety! http://www.pupman.com/safety.htm



If you do anyway, be aware of large switching transients that may damage nearby electronics, read this entire article before proceeding.

The idea to this coil came with Steve Ward showing off his first QCW DRSSTC that used a buck regulated DC supply to ramp up the supply voltage along with a long on-time to grow straight and very long sparks compared to the secondary coil length.

I thought it could be done simpler, yet with less control, by using the rising edge of 50 Hz mains supply voltage. From start of the sine wave to the top it corresponds to a on-time of 5000 uS and to be able to use large IGBT bricks the frequency would have to be kept down. Sword like behaviour of sparks is however mostly seen at above 300-400 kHz, where as lower than that results in more branched sparks.



A high impedance primary circuit is needed to keep peak current at a level that the IGBTs can handle to switch for very long pulses, for a DRSSTC, up to 5000 uS. In order to get enough primary windings, I went for a upside-down U shape primary as a regular helical coil with high enough coupling would quickly get as tall as the secondary coil itself.

To use 3 IGBT bricks in parallel it is important to ensure as even current sharing as possible, this is done by mounting them close to each other on the same heat sink, drive them from the same gate drive transformer with individual gate resistors matched as close as possible.

Steve Wards universal driver 2.1b only has a robust enough 24 VDC section to run up to about 300 uS on-time with large gate capacitance, when trying to run with longer on-times than that, the 24 VDC 1.5 A regulator is now longer enough to supply the needed current. A external 26 VDC 8 A power supply is used instead and output stage will have to be reinforced to conduct higher currents and dissipate more heat.



Bridge 3x  SKM145GB123D IGBT bricks in a parallel half bridge configuration
Bridge supply 0 – 260VAC through a variac
Primary coil 21 turns of 8 mm copper tubing in a up-side-down U shape
MMC 10 strings in parallel of 10 in series Kemet
 capacitors for 0.047 uF at 9000 VDC, 280 A peak and 40 A rms rating.
Secondary coil 160 mm diameter, 330 mm long, 1500 windings, 0.2 mm enamelled copper wire.
Resonant frequency Around 100 kHz.
Topload 100 x 330 mm spun aluminium toroid.
Input power 10BPS, 500 cycles, 50A limiter: 750W at 260VAC at 3A.
Spark length Up to 500 mm long sparks.



Bridge section

Driver section

Same as Steve Wards universal driver version 2.1b. Just made on single sided PCB without SMD components and the 24VDC part has traces reinforced, MOSFETs heat sinked and uses a external power supply. A external 26VDC / 8 Ampere power supply is used to ensure that under voltage will not be a problem, at least not before something starts to smoke.



31st October 2011
I put the bridge together on a heat sink with 3 phase rectifier used with all inputs in parallel for 1 phase supply and connected all 3 half bridge IGBT bricks in parallel with 3 straight bus bars. All recycled components from a DC link inverter.

Designed staccato PCB as the old layout used in my VTTC I was on a vero board. A optical output was added to use the interrupter with a standard DRSSTC driver.

Started construction of the secondary coil.

2nd November 2011

Etched PCB board for staccato controller.

Finished winding the secondary coil. It was made with a total of 1500 turns of 0.2 mm wire and dimensions 160 x 330 mm. Varnished the secondary coil with polyurethane varnish.

3rd November 2011
The secondary coil was given a second thick layer of varnish, not the most pretty job as I tried to pour as much varnish on as possible and let it rotate and settle it around the coil by itself. As the secondary coil was not completely in level it result in a little running, but overall a fair result of adding a lot of thin varnish.

Finished assembly of the staccato controller and bench tested it.

The project got shelved due to starting on a new job, after a long rest of over 3 years the box with parts was once again brought out in the light and construction could continue.

7th March 2015

Etched driver PCB and started populating the board with all passive components.

6th October 2015

Construction of a very cheap MMC from capacitors that was bought from a 10$ ebay auction for 100x Kemet R474N247000A1K, rated for 900VDC 28 A peak and 4 A rms. A easy and uniform construction, with current sharing in mind, is to construct it around a round piece of wood or plastic tube.

The resulting MMC has a capacitance of 0.047 uF at 9000 VDC, 280 A peak and 40 A rms rating. Which is spot on for this coil to be running with design goal primary inductance of around 100 uH, 300 A peak, 5000 uS on-time and maximum 10 BPS.

11th October 2015

Construction of acrylic primary supports, that has 21 slots and is formed for a up-side down U shape primary coil. A way of getting a large primary inductance and still maintain a certain distance to the topload as the secondary coil is very short.

The supports are made by hand using a saw, file and drill press.

16th October 2015

Getting the coil winded from the inside and out was no easy task, the whole large roll of copper tubing is heavy, easily bends too sharp and is like a spring. It will lock itself in the wrong slots and it can be a very frustrating piece of work. The complete result is however worth the effort, it looks smooth and even.

As the slots was not made to snap the copper tube in, from shear fear of cracking the acrylic, I made a small hole behind each slot that made it possible to tie each turn at each primary support, with a little piece of copper wire it is secured from deforming the coil.

As water cooling of the primary coil is going to be a must with the long on-times, simple clamps was made from copper sheet and two screws the fastens the 4 braided copper flexible wires to the tubing. The same 4 braided copper wires was soldered to the MMC terminals, as even as possible distributed around the circular copper wire terminal.

Having made the MMC on a wooden stick makes it easy to mount with two wooden blocks with holes in for the extra length of the round rod.

29th December 2015

Driver board populated with all active components. 24 VDC regulator is left out as this will be supplied externally from a 26 VDC, 8 A power supply. All traces related to the 24 VDC is reinforced by soldering a 0.5 mm2 copper wire along them. Four 2200 uF 35 VDC capacitors was added to the underside of the board, one at each N- and P-channel MOSFET. All output stage MOSFETs have heat sinks mounted.

All these precautions are hopefully enough to ensure no under voltage or over heating situation is possible when running at 5000 uS on-time.

25th July 2016

The coil have been put together with power supply, bridge, driver, platforms, secondary and topload. Two fuse holders for large bussman fuses was used at the end of the flexible copper braids for primary tap.

7th October 2016

First test of the coil after all the components have been put together, there are still a few things not in place, but it is good enough for initial testing.

The first test is without power on the DC bus. This is solely to test if the driver and power supply is good enough to drive the 3 IGBT bricks’ gates in parallel with a satisfying gate waveform. The following oscilloscope shots show the coil being driven at 5 ms on-time at 100 BPS, corresponding to every rising edge of the half wave rectified 50 Hz mains supply.

The following months, where I had only little time to do more testing, I could not get the coil to run properly with power on the DC bus. I could measure oscillation at the resonant frequency, but nowhere near enough for the coil to actually produce a output ever so slightly or just so small as to light up a close by fluorescent tube.

I tried running the coil with added 6000 uF capacitance, with normal interrupter, with and without DC bus snubber capacitor, with much fewer primary windings etc. I did a lot of changes to it, so it would be more like a regular DRSSTC, than a QCW, little did it help and I did not get much further before putting it away for Christmas holidays.

26th March 2017

After some online discussions as to what the problem could be, and showing off the coil for the first time, it was suggested that the regular 1:1000 cascaded current transformer for feedback was simply not delivering a strong enough signal as a high impedance coil naturally works with a much lower peak current. From the thread on the forum here https://highvoltageforum.net/index.php?topic=24.0 it is decided that I will try to make a new 1:50 turns ratio current transformer to see if increased feedback will help the coil oscillate.

The only capacitance on the DC bus was a 0.68 uF snubber capacitor, this was also increased to a 10 uF capacitor bank of MKP film capacitors. A small amount of energy storage is needed to make the phase correction driver able to run stable.

First light was achieved with some 20 cm long sparks, remember that the coil is far from tuned for maximum performance and the input voltage was only 200 VAC.

1st April 2017

I made a wide range of secondary and primary circuit measurements to find the resonant frequencies, as the U-shaped primary makes it difficult to simulate with tools like JavaTC.

The measurements was done with secondary in place inside the primary. But secondary ground was unconnected and primary circuit was left open loop by removing the tapping point. I am not sure if this is the correct method, as resonant frequencies are much different when measured with secondary ground or primary loop closed, this is because energy is then transferred between the two resonant systems. The results could however vary with 10-20 kHz compared to the open loop measurements…

Secondary circuit test results
Setup with a 80 cm long wire with 3 bend wires hanging over and pointing down to be “branches”. Signal from signal generator connected to ground terminal on the secondary coil, ground left floating. Signal into oscilloscope captured from open loop probe hanging next to secondary coil.

Unloaded result: 101 kHz, 80 cm wire result: 91 kHz and 80 cm wire with branches result: 88 kHz.

Primary circuit test results
Setup with signal generator and oscilloscope connected across the primary LC circuit and with a jumper across the IGBTs to have a closed loop. Signal generator is connected through a 10K resistor.

Primary resonance with secondary ungrounded – 7th turn from bottom 102kHz, 6th turn from bottom 96kHz, 5th turn from bottom 90kHz and 4th turn from bottom 86kHz.

Rest of the measurement results did not have individually saved oscilloscope shots, so here is a overview of the primary tapping frequencies.

I recorded video from oscilloscope and spark formation (dark dark video, blerg, sorry). There are 4 tests where primary is tapped at 96 kHz, 90 kHz, 86 kHz and last at 65 kHz, where it for reasons I still do not fully understand, performed the best! This is a huge detuning compared to the loaded 88 kHz secondary measurements. I also tried all the taps between 86 kHz and 65 kHz, with only increasing performance until I could detune it no further.

The staccato interrupter is not particular stable and does not really give a good clean 5ms on-time, it starts conducting before the zero crossing, possibly due to non-adjusted phase correction on the driver, this will get looked into next time its running. Waveforms are highly distorted, peak currents are low in the magnitude of 50 A peak. (Blue 100V/div inverter output – Yellow 100A/div current – 5-6 ms on-time)

and also some close up pictures of the beautiful sparks, still much shorter than what I expected, but maybe the very high impedance primary circuit just limits the current way too much, future experiments would be with a step up transformer for a higher primary peak voltage.

4th April 2017

From last nights experiments, I think that this idea might work on a small scale, the peak currents drawn by this large coil simply creates too large switching transients.

I tried tuning the coil at 120 kHz and 130 kHz, way above the estimated loaded secondary resonant frequency of 88 kHz. It performed better than ever before at 120 kHz tap, which properly makes a little sense compared to the better performance at 65 kHz, it certainly does seem to be a harmonics pattern here. But I do not think I can tap it any further down on the inner side of the primary coil right now.

There was however also much higher current draw, loud clunks from the variac and lights dimming! The voltage spikes on the mains supply are at levels where my voltmeter was damaged in my variac. This is also why I call quit on the project as it is, its future will be rebuilding it to a conventional, properly PWM controlled, QCW.

I had sparks fly out to about 50 cm as it can be seen in the video


So far the prototype has worked and shown that the concept works. The spark formation is more straight than first anticipated, as most QCW coils operate above 300-400 kHz to get long sword like sparks. It is however clear that the sparks produced by this coil, resonating below 100 kHz, is swirling a lot.

Tuning is very different from a regular DRSSTC where the sweet spot that produces the longest sparks at the lowest current can be within a few centimetre on the primary coil. Here I could get the same performance over a wide span of 60 kHz, tapping the primary anywhere would give me around 30 cm sparks, but it was easy to recognize when a true sweet spot was found, as the very abrupt current draw could be heard clearly from the variac clunking loudly and lights dimming slightly.

The switchings transients are however a great danger to nearby electronics and is of a magnitude where filtering is properly not enough, certainly it is not a solution to add more passive components to counter a problem that can be completely eliminated by using a different topology and have a control scheme that can control a ramped voltage from a capacitor energy storage, like the class D amplifier, phase shifted or PWM controlled QCW coils demonstrated by other Tesla coil builder.

I wanted to try this method, to see if simplicity could do the same job, it could not.


There is not a overall demonstration video yet, but the 3 videos from research development above.

Musical SSTC/DRSSTC interrupter

All credit for this article goes to Martin Ebbefeld from ctc-labs.de, after he closed down his website we made an agreement of me hosting some of his content. I translated his article as best as I could from German to English and added a few more details or information. The pictures are unfortunately in a low resolution, but it is for now all that is available.

A first few thoughts on how music can be played with a DRSSTC. The high current primary waveform that is used in a DRSSTC can not just be frequency modulated to amplify a analog signal as played back from a regular music source. The stress would be too great, the heat dissipation would destroy the IGBTs.

So what do we do?

A lot of clever and experimenting amateurs around the world have implemented different methods to be able to reproduce tones with a DRSSTC, tones that are equal to what a piezo speaker can produce, but almost all music is not that simple.

The high frequency resonant signal that is used to drive the bridge comes in small packages of 100 to 300 us length, this does not leave us much head room to modulate in. What we can do instead is to modulate the time that lies between the pulses, the so-called off time. Either by MIDI signals or from a simple analog to digital conversion signals as it is done in this article. (To be able to understand the following text and piece it all together, it could be necessary for the reader to study basic operation of a DRSSTC)

The simple circuit used here is a development of a circuit earlier made for other purposes, it could now be customized to work as music interrupter. We can not simply play back our favorite piece of music, we either need it in MIDI format or as a raw melody where we will have to edit it by hand.

Although there are already programs freely available for micron controllers to process MIDI signals in real time to a series of pulses to drive a DRSSTC over fiber optic cable, that method can be too complex for some experimenters. This article uses old and easy to understand analog components.

A normal DRSSTC will operate with a break rate of around 120 to 200 Hz (off time). This means that the time between pulses is 1/200 = 0.005 = 5 milliseconds. The characteristic buzzing sound is a product of the 5 millisecond off time and if we adjust the off time to be less, the distances between the pulses will be shorter and the sound will rise in pitch and appear as higher tones.

The idea is to match the input to a list of real tones, a awesome and brilliantly simple method. So how is this implemented in the modulator, the following examples and diagrams will explain this.

Tabel 1: Almost entire piano key list. Middle C (C4) and Tenor C (C5/treble C) with their respective frequency in Hz
Note C1 C2 C3 C4 C5
C 32.70 65.41 130.81  261.63 523.26
Cis 34.69 69.29 138.59 277.18 554.37
D 36.71 73.42 146.83 293.66 587.33
Dis 38.91 77.78 155.56 311.13 622.25
E 41.2 82.41 164.81 329.63 659.26
F 43.65 87.31 174.61 349.23 698.46
Fis 46.25 92.49 184.99 369.99 739.99
G 48.99 97.99 195.99 392 783.99
Gis 51.91 103.83 207.65 415.3 830.61
A 55 110 220 440 880
Ais 58.27 116.54 233.08 466.16 932.33
H 61.73 123.47 246.94 493.88 987.77

Middle C, called C4 is the octave called one-lined and Tenor C (treble C) is the octave called two-lined. The above table shows the frequencies in Hz for the different scales and are within those that a Tesla coil could play without going into extreme conditions. The inability to play higher tones lies with the ratio of on time and off time that becomes very unfavorable for the system.

A proportional limit between on- and off-time should be kept to maximum on-time to be 10%. This means that the highest playable note that a Tesla coil should play to avoid any damage would be maximum 10 times the maximum allowed on-time.

To understand this, lets look at the following illustration.

The simulation above shows a interrupter signal with 10% duty cycle which represents pulses that are 200 us long, which approximately corresponds to the pulse length in a normal DRSSTC, this is the period of time where the coil operates at its resonant frequency. Some coils operate with shorter pulses due to faster ring up time in low impedance primary circuits or some need longer pulses if they have a primary circuit with a higher impedance.

With a duty cycle of 10% and on-time lasting 200 microseconds, the length of the off-time ,until the next rising edge of the on-time, is 1.8 milliseconds.

\mbox{1.8 ms}=\dfrac{1000}{1.8}=555.55 Hz

555.55 Hz almost corresponds to the tone Cis in C5.

The reason to not play notes higher than 555 Hz has to be found in power dissipation limits, tolerable temperature rise and overall stress of the IGBTs, read my IGBT design guide for more details on this. The closer together the on-time pulses are, the more often the bridge is on feeding energy to the resonant circuit of the DRSSTC. More heat is dissipated, it is a larger load and power consumption rises dramatically. If the off-time is short enough, it may happen than that pulses start to overlap the ring down period of the damped oscillation and the current in the primary circuit no longer settles back to zero, but oscillates between the lower and upper limit which eventually will lead to failure in one way or another.


We need to use a source that can feed our modulator with tones according to the sheet in table 1. This can be a synthesizer, whether it is software or hardware does not matter. We can however not use a normal music CD, it would only result in a chaotic mess of pulses that would not produce any good results on a DRSSTC. To get the best result we have to take out a part of the song, preferable the melody or lead guitar, something that is simple and recognizable for the song. With the method this modulator uses it is possible to convert the length and volume of each tone to a pulse length and exact pitch that the can used to drive a DRSSTC into playing audible tones.

Another example will follow.

In the above illustration the signal can be seen, from which the modulator has its pitch and tone length. Using the internally controllable on-time of the modulator it can vary the volume.  The highlighted area in the illustration is about 2.154 milliseconds long and that corresponds to:

\mbox{2.154 ms}=\dfrac{1000}{2.154}=464.25 Hz

464.25 Hz almost corresponds to the tone Ais in C4. With a deviation below 2 Hz.

These pulses was created in a music software where it is possible for a synthesizer to output square wave pulses. The tone lasts for 65.034 milliseconds and thus corresponds to a 1/16th note, since 1000 divided by 65.034 gives 15.376 and it roughly is 16 bangs per second, if you include the short breaks between each stop note.

In the above illustration we will take a look at a low tone for comparison.

The length of the note corresponds to the earlier illustration, it is also about 1/16th note, but with a much more coarse structure.

The length of one period of the pulse is here 8.66 milliseconds

\mbox{8.66 ms}=\dfrac{1000}{8.66}=115.47 Hz

115.47 Hz almost corresponds to the tone Ais in C2. With a deviation below 1 Hz.

This is a very low tone and DRSSTC does actually handle low tones quite well, also the so-called base line notes known from base guitars.

However these square wave pulses lasts for too long and the duty cycle is up around 50%, which is much higher than our initial design specification of a maximum 10%. So we need something in the modulator to make a more useful signal out of the very high duty cycle signals.


Generating the tone sequences described above by the means of software is the most difficult part of this modulator. Without properly prepared signals from soft- or hardware there will be no singing Tesla coil from this modulator.

The simplicity of the circuit makes it easy to debug and possible also to expand with more features. It consists of a operational amplifier that works as a pre-amplifier so that it is possible to use low level audio signal from a CD, PC or keyboard.

The input audio signal is filtered and amplified by the LM741 operational amplifier. It is further amplified by the two BC547 transistors and converted to a square wave equal to the output of a Schmidt trigger. So it is possible to feed this modulator with sinusoidal signals and get a proper square signal to drive the DRSSTC with.

The most import part of the circuit is the 555 timer, the main task of the modulator is handled here, which is to ensure a correct length of the pulses as to not exceed the 10% duty cycle.

In the above illustration we have the input sound signal in green, the signal that is fed to the LM741 operational amplifier. This signal is conditioned, amplified, inverted and inverted again before it lands at the trigger input of the 555 timer.

If the trigger input of the 555 timer is held high, nothing will happen on the output and the oscillator circuit output made with the 555 timer is on standby. When a series of pulses are put into the modulator, the second BC547 transistor connects the 10 nF capacitor to ground and the trigger input of the 555 timer is momentarily pulled to ground. This causes the 555 timer to output a pulse of precise length determined by the potentiometer and the 22 nF capacitor. The signal shown in red is the output signal from the 555 timer.

The distance between the rising edge of the red pulses corresponds exactly to the distance between the rising edge of the green pulses, that was our input, so the important information as pitch is followed through. Everything that is done with this circuit is to cut off the unnecessary part of the square wave. The on-time, the width of the red output signal, can be regulated with the potentiometer to determine how much power the coil is allowed to use.

In the following illustration is can be seen that varying pitch and sequence of the input signal is used to generate a series of pulses that can be used to drive our DRSSTC with.


A last sensitive issue needs to be addressed. Very pleasing and cool sounds can be made by playing more notes simultaneously, also known as chords, but it does present some interesting issues with this simple modulator.

A square wave in itself is not able to express a chord as it can only represent one state. Different synthesizers overcomes this problem in different ways and some uses a intermediate frequency where it alternates between the two notes used in a chord so that for each X’th part of a chord it plays the two notes like this A, F, A, F, A, F, A etc. This method can also be simulated in some software synthesizers.

Chords does however also last longer and a short allowed on-time might give interesting results where the sound is not faithfully reproduced to the source. Chords are not recommend with this modulator, but experimentation could get interesting.



Music is played from the speaker output of a childrens toy keyboard connected to the input of the audio modulator. The Tesla coil demonstrated in these videos it is the Kaizer SSTC 1 in a state where it was being rebuild to the Kaizer SSTC 2 where the implementation of the modulator is shown in the schematics.

Good MMC Capacitors

Published December 26, 2013. Updated February 24, 2021.

Here is a list of capacitors tested by the high voltage community to be known to withhold the use as primary capacitor in Tesla coils.

Capacitor specifications are taken from data sheets at 100 kHz and some values for peak current, RMS current, ESR and dv/dt are estimates(* marked) from similar capacitors and graph read outs.

Product  IpeakIrmsESRdV/dTRth
data sheetVμFAAV/μS°C/W
Aerovox RBPS20591KR6G1000285422742715*
Kemet Arcotronics

Capacitive reactance Xc = 1 / ( 2 * π * f * C)

ESR can be calculated from the tangent of loss angle given as TANδ in the data sheets. ESR is frequency dependant. Capacitance is given in Farad, frequency in Hertz. ESR = (1 / (2 * π * f * C)) * TANδ = TANδ * Xc.

Thermal resistance (Rth) when given in data sheets are either Watt needed to raise the temperature by one Kelvin or degree Celsius the temperature raises by one Watt dissipation. Conversion from W/K to °C/W is to divide one by W/K dissipation factor.  °C/W = 1 / (W/K).

Ipeak is calculated from the dV/dT rating times the capacitance of the capacitor. Capacitance given in micro Farad times pulse rise time given in micro seconds will give a result in Ampere. Ipeak or Ipulse = C * dV/dT.

As a rule of thumb ESL is about 1.6 nH per millimeter of lead distance between the capacitor itself and the rest of the circuit. This also includes the leads of the capacitor itself. This only applies to well designed capacitors.

MMC calculator

MMC tank design calculator for SGTC, VTTC, DRSSTC and QCWDRSSTC Tesla coils. Results are guidelines to designing a MMC and should always be double checked in your final design! Most importantly is that voltage rating is the DC voltage rating, from experience this can used for good quality capacitors, AC voltage rating with frequency derating would be much lower.

Capacitor specifications are taken from data sheets at 100 kHz and some values for peak current, rms current, ESR and dv/dt are estimates from similar capacitors and graph read outs.

Inputs are in green. Outputs are in red. Formulas used can be seen below the calculator.

Basic MMC configuration – List of good MMC capacitors
Capacitance uF
Voltage rating VDC  
Capacitors in series
Strings in parallel
Price per capacitor  
MMC voltage rating VDC  
MMC capacitance uF  
Total capacitors  
Total MMC price  
Advanced options
MMC capacitor parameters
Peak current rating A  
RMS current rating A  
dV/dt rating V/uS  
ESR rating Find correct ESR rating
for your resonant frequency
dissipation factor
Tesla coil parameters – Examples are small, medium and large
Frequency kHz
Primary inductance uH  
Primary peak current A
On time uS
Advanced results
Primary impedance
(single cap)
Power dissipation
(single cap)
Temperature rise
(single cap)
ºC 0-5 very good, 5-10 good
10-15 not good, 15+ bad
  Actual values MMC rating
Peak voltage MMC VDC VDC
RMS current MMC A A
dV/dt for MMC V/uS V/uS
Peak current for MMC A A

Theory used

MMC voltage rating: MMC voltage rating = DC voltage rating * capacitors in series.

MMC capacitance: MMC capacitance = (single capacitor capacitance * amount of capacitors in parallel) / amount of capacitors in a string.

Primary impedance: Zprimary = SQRT(Lp / Cp).

MMC Xc, reactance: Xc = 1 / (2 * PI * F * C). F is frequency in Hertz. C is capacitance in Farad.

MMC Zc, impedance: Zc = SQRT(ESR^2 + Xc^2). ESR is the combined ESR for the MMC. Xc is the MMC reactanse from above.

Peal voltage MMC: DC peak voltage over MMC = Zc * primary peak current

RMS current MMC: Irms = 0.5 * primary peak current * SQRT(on time * bangs per second). Steve McConner.

dV/dt MMC sees: Actual dV/dt in V/uS the MMC sees = (2 * Pi * V) / F. V is peak DC voltage over MMC and F is frequency in Hertz.

dV/dt rating MMC: dV/dt rating in V/uS = Primary peak current / MMC capacitance.

Peak current for MMC: Peak rating = capacitor peak rating * amount of capacitor strings in parallel.