https://www.pes.ee.ethz.ch/uploads/tx_ethpublications/Keynote_Presentation_ITELEC_15_FINALFINAL_as_published_251015.pdf
Do you have something like that from any other teams?
I never put together a prototype unfortunately, but I enjoyed the analysis a bit:
http://seventransistorlabs.com/Images/Proposal_2014-07-26.pdfI didn't think it would be very much worthwhile to use different semiconductors (which were all but unavailable at the time -- and over a year later, are still hardly to be found!), the semiconductors being a vanishingly small part of the package (< 5%), even if you have to do crazy things with them to get the efficiency (multi level inverter perhaps?).
Naturally, I was having some trouble getting a simulation to show reasonable results. I tried a number of approaches, e.g.
http://seventransistorlabs.com/Images/Simulation_Screenshot1.pngfrom which I discovered some glaring discrepancies in the models (it's rather hard to get 45 +/- 15W dissipation out of 2000W of power, when the effect of your models' errors is not like 1%, but 40%!). The general plan would've been ZVS with "too-small" filter inductors. Of course, this puts a large burden on the inductors, which need to be rather beefy, and extremely low loss.
ZVS is basically the only possibility for silicon. You need enough semiconductor to get conduction loss at least low enough to work (e.g., 70A transistors for a 7A load), and then you have to deal with the rather massive junction capacitance (which accounts for upwards of 100W, if it has to be charged dissipatively via hard switching or snubbers).
As for filtering, I find it interesting that the above review shows the ripple module taking up almost as much space as capacitors, anyway. Their capacitor multiplier (in effect) is rather sizable, more than doubling the volume of their capacitors alone. This is about what I expected. It might end up strictly smaller than using electrolytics, but the advantage for all that added effort is very, very slim.
I didn't know of those TDK ceramic link capacitors at the time, so I didn't have them tabulated. Anyway, it seems like C0G get better energy density anyway (and hell, about the same cost, considering they spent some $5k on capacitors in that damned thing!). But you still have to deal with it, which is a real stinker.
Another analytical oddity that may be of use: minimal volume filter designs. As filter order goes up, attenuation goes up, so the cutoff frequency can be closer to the attenuation limit (i.e., the frequency where attenuation must be e.g. 60dB). As order goes up, the number of components goes up (which means more energy storage), but as frequency goes up, the size of components goes down (which means less). I derived the formula assuming asymptotic behavior, which should be okay for modest filter designs (Butterworth to 1dB Chebyshev, say) and high attenuation (>40dB?). The ratio of cutoff to limit frequency, for this condition, is actually dependent on the desired attenuation only.
(Another aspect of minimum volume design would be relative component size. Inductors and capacitors don't carry the same energy densities. The ratio of L/C corresponds to filter impedance, so you will have the function of combined volume versus impedance, with a more or less parabolic curve, having a minima somewhere.)
Tim