I saw this earlier and briefly considered it. 50W/inch^3 is soldering iron level heat dissipation. And my take on it was that it really isn't possible unless you can cheat and have the "inverter" be the thing on the end of a solid copper bar that is sitting in ice water on the other end :-). So really they are looking for a 10X improvement in efficiency. Which is to say to take something which is 90% efficient and make it 99% efficient. Even looking at the wide bandgap semiconductors they reference on the web site I'm having a hard time getting more than a few percentage points more efficient.
I think the power density they quote (50 W/in^3) isn't for dissipation, it's inverted power. I'm not an expert in electrical engineering, but I can't think of a fundamental physics limitation that's a blocker here. If there is one, I'd guess that it'll involve the electromagnetic radiated power of the device. From the specification document [1], they're asking for 95+% efficiency.
Thanks for the link, you're correct they want to pull 2KVa out of a box no bigger than 40 cubic inches. These guys (http://www.apparent.com/products/) have what I consider a very workable technology, basically 2W/cubic inch but one inverter per panel. Since you have to have panels anyway, having the panel produce power directly in the form you want eliminates the need for size as the panel is compelled to be a certain size anyway.
But instead of changing the question (and age old trick of engineers to arrive at a feasible solution :-) The push here seems to be about efficiency. 95% efficient would be a huge improvement.
Whatever they go with, the tech is going to make a really killer subwoofer amp.
Looking at the requirements (mostly the high DC input) it seems that an IGBT class-D would be the logical starting point, but those are hampered by a 150kHz-ish max switching rate. I think Don Lancaster's Magic Sinewaves (http://www.tinaja.com/magsn01.shtml) meet the distortion requirements while offering the slowest switching requirements.
>50W/inch^3 is soldering iron level heat dissipation
...we'll 50W/inch^3 is the throughput. If you assume 96% efficiency (which is common for COTS inverters), you'd have 2W/in^3 in waste heat to dissipate. And 50W/in^3 isn't very high on just a per-piece level. Here's a Vicor DC/DC converter that has 1240W/in^3 power density:
I think that is likely the winning path. When I first looked at this that was my approach, basically a flat sheet of copper with the inverter laid out on top of it. Basically a 10 x 16" copper heatsink with holes in it where vertical components sat, capton tape on the back flex circuit and D9 packaged semiconductors on the other. My battlebots motor controller (which really was just a 24DC to 24V PWM converter at 200amps[1] which was in the ball park (power wise) but FETs don't really like operating at really high voltages so I was looking at an IGBT version.
As I was more concerned with power dissipation early on, more surface area is great for convection.
