The mosfets are from IXYS Corp, which can be purchased from Digikey Corporation. The microcontroller is from Arduino.
Ten kilowatts is a lot of power, and this is what is needed to levitate molten copper and steel. At this power level high currents cause oscillations on the gate and the PLL topology is not good enough to maintain a tight resonant lock. I wanted something that could find the resonant frequency with any coil and lock onto it without any manual adjustment.
Precise resonant locking and tracking was accomplished with a microprocessor-PLL circuit. I use the PLL to find the phase difference between the inverter and tank capacitor. Ninety degrees is the correct phase difference. I use the microprocessor to monitor the PLL output and develop a DC voltage that corresponds to the phase difference. I use this DC voltage as the input to the PLL's VCO in order to maintain the correct frequency. Here is an important point: tune to a slightly higher frequency so the current slightly lags the voltage. If you are too close, or the current leads the voltage, the mosfets will heat up. I got my mosfets to get hot with small currents if I had the tuning too close to ZVS. When I tuned the current to slightly lag I had 10x the current going through them and they still remained cool (with forced air convection).
Next, I had to move to a 240 vac line @ 30-50A. At this power level heating becomes a very real issue. I have two 100 cfm fans blowing on each 5"x5" heatsink for the mosfets. I switched to mosfets because they work a better at the 100 khz frequency range, and they have less switching losses. I am currently using the IXYS Polar HiPerFET series IXFN56N90P mosfets. I am using two in parallel for each leg of the half-bridge. At 25c each mosfet can handle 56A. I am figuring that I should keep each under 30A, which is why I have two of them. They are rated for 900v.
Running mosfets in parallel can be tricky. First, you want to make sure they all come from the same lot. The problem with running more than one device is unequal heating and oscillations on the gate. Fortunately, because these mosfets have positive thermal coefficients, the hotter they get the less current they conduct. This way, one mosfet does not run away and carry more and more current as it gets hotter than its partner. Make sure the mosfets for each leg are on the same heat sink. Second, you need the have enough resistance on the gate to prevent oscillations. Five ohms is enough. I used 10 ohms because the Ldi/dt was too high with the former, resulting in ringing during the transitions. I originally had ferrite beads on the gate leads, but I eliminated them in order to shorten the lead length. This resulted in even less ringing to the point of it almost being non-existant. The gate resistor was sufficient to prevent oscillations on the parallel devices.