An LLC converter for the rest of us.Of all the converter topologies out there, the LLC converter has been considered to be among the hardest of them all to design, and this may well be true, specifically so, if it has to be designed on paper, based on math equations, as well as to be stabilized with a feedback network.
In this post, I will present a different, more practical approach to the design, partly by viewing the LLC converter like more of just an operating point than an actual topology, partly by avoiding the feedback network, and rely on the regulation inherent by the internal impedance of the supply.
To be specific, the topology is a half bridge driving a resonant tank comprised by a series capacitor and the leakage inductance in the transformer. The ratio between the leakage inductance and the magnetizing inductance should ideally lie between 5-10, and to obtain that ratio, the magnetizing inductance has to be lowered. This is done by adding an air gap to the core.
When this is done, the half bridge is driven at a frequency that is higher than the tank resonance, and with around 200nS of deadtime.
It is really that simple: Do this, and you will land in the sacred LLC mode, where switching transitions happen at zero voltage, and this means: no spikes, no overshoot, no oscillations, no need for snubbers, just pure clean transitions from Vbuss to Vgnd.
With a variable transformer as the front end, you can control the output from 0V to whatever the turns ratio of your transformer amounts to, you get a drive capability of a couple of kilowatts in a reasonable form factor, in this case 80x100mm.
I had my eyes on the Ti UCC21520 Gate drivers which appealed to me with their sub 20nS propagation delays, and I wanted to try SiC mosfets too.
I chose the UF3C065030K4S with 27mOhm on resistance and turn off delay of less than 50nS. To drive this bridge, I initially chose the IR2086 and this became the basis for the first iteration.
I am putting emphasis on the devices delays, and this is because I view delays much as I do backlash in a mechanical system. If it is in the rack and pinion on your car, any slack would be downright dangerous, otherwise it is mostly an annoying nuisance. In my day, I have had to put up with devices that had more than 1 uS of delays and find it a great relief to finally be able to buy devices where the delays can be ignored.
This first iteration was based on a full bridge PCB, originally designed for gate drive transformers, where I placed the gate driver chips next to the switches, not paying any attention to the fact that the traces leading out to the tank passed by right under these chips, on the PCB back side. This gave me the ultimate lesson to never route power under a chip. They may have great low impedance outputs, and clamps on the inputs, but internally they are of course wonderfully high impedance, and any amount of high frequency current right next to them, will induce voltages inside them, and they will fail by the numbers, until the power is routed away from them.
I also found out that there is little reason to use a full bridge, half will do nicely.
The driver chip that I started out with, the IR2086 is more or less just an oscillator with gate drivers included, but I wanted to add safety measures and the search for a better choise lead me to read app notes, and inevitably, the idea of using the LLC operating point.
The best chip I could find was the FAN 7631 from Fairchild. It has several useful features, one being that the frequency of oscillation is set by a single resistor, and so is the deadtime. On many of the other chips, these two key parameters are selected by the combination of a resistor and a capacitor, meaning that for experimenting, these components have to be soldered out and exchanged too often.
The FAN 7631 has soft start up, several current limit schemes, brownout protection and the list goes on with a useful selection of latching and auto restart features.
Here is the latest iteration:
I use LM35 temperature sensors on the switches and the transformer core, since these elements are the most likely to overheat, although I must say, those SiC mosfets do not get hot at all! Incorporated in the driver pcb are 5V outlets to power such thermometers.
I like even numbers, and was looking to get 5kVp out of the transformer, however my basket winding machine started to fail when the turns count approached 500 turns, and this should lead to problems. With the cores I could get, an UY22A with 22mm diameter, I really could not sustain 10v/turn. With ferrite cores in high frequency converters there is a bad tradeoff, because above 20kHz, the core is limited by losses leading to heat. This means that although the higher frequencies could lead to more volts per turn, the delta flux to be lowered to keep the core cool, and you are back to square 1.
In this case, the operating point was finally selected for me, by the rectifying diodes in the multiplyer, 2CL2FP which are 30kV, 100mA, 100nS types, because between 90-100kHz, they just plain folded in and fried within 10-20 seconds.
At 75kHz they would pass 33mA and at 30kHz, they work fine even at 50mA, so that settled it: 30kHz it is!
At 30kHz I can get away with 0.32T delta flux for 7 volts per turn, and so, with a bit of gain from the LLC operating point, I can get 4.2kV from the transformer leading to 8.4V per stage.
7 full wave stages, and I land at 60kV
The design of the voltage multiplier was based on a crate of capacitors I had from a tesla coil that never made it past the hoarding stage, and they work fine, of course. 68nF 1600V, I work them a bit higher at 4 per stage for 2.1kV per cap, they are known to start popping at 3.6kV so.
The multiplier was prototyped under oil, which is a mess, and the first intermediate is potted with Robnor PX439N which is a heat conducting epoxy potting compound. This works fine so far, even though quite a bit of heat is generated in those series resistors. I use the resistors to keep peak current in the diodes below maximum level, but also because they calm the output waveshape.
At the present stage, I have 30kV to drive 50mA into 600k, and nice textbook waveshapes:
Now the next stage is to get the 60kV 50mA with a 7 stage multiplier, but I am waiting for delivery of the potting compound.
Cheers, Finn Hammer
