Off-line flyback uses HV transistor in common-base configuration
OK, so I get a press release (pdf) from those wonderful folks at CamSemi, AKA Cambridge Semiconductor. It talks about their C2163 off-line switcher chip (pdf). The press release says: “…the BOM cost is much lower compared with any of today’s more expensive MOSFET-based solutions.” So unlike the editors of most publications, I actually go and dig out the datasheet for the C2163. Imagine my surprise when I see it uses a power MOSFET as well as a HV (high-voltage) transistor. So I ask to see real pricing that proves that a HV transistor and LV MOSFET is cheaper than a HV MOSFET and they say that they can’t give me Digi-Key pricing since it does not really apply to these high-volume products, where you would be buying hundreds of thousands of transistors at a time. I see their point, but I still am a little annoyed that the press release implies you don’t need a power MOSFET when you really do need a power MOSFET.
So then I ask my pals what is going on and the really clever thing about the circuit comes out. See, they are operating the HV transistor in common-base mode. They hang the base up at some lower voltage, say 10 or 20 volts. Then they use a high-current low-voltage MOSFET to yank the emitter down to common and turn on the transistor. This is cool for two reasons. First, you can use a low-voltage MOSFET, which is much cheaper than a HV one. But the really cool thing about operating the HV transistor in common-base mode is that it means the driver circuit is not subject to Miller Effect. Miller effect is really nasty with HV MOSFETs. You are trying to stuff charge into the gate and get the voltage to go up. But turning on the MOSFET means the drain of the transistor is rocketing down in voltage, after all, you are turning on the FET so that drain voltage is getting connected to common. But all MOSFETs have a capacitance between the drain and they gate. So despite the drive circuitry trying to charge up the gate, the drain of the MOSFET is dropping by hundreds of volts, and doing it fast because you want to minimize linear conduction losses in the FET, so you want it turn it on a as fast as possible. So even though the capacitance between the drain and gate is small, the drain is slewing so many volts in such a fast time, it really slows down the ability of the gate to take charge from the driver circuit. So now you beef that up, but remember, that charging current is all “wasted” it just gets stuffed into the gate-source capacitance and then discharged when you turn the FET off. So yeah, the CamSemi part is cool, and you should all go out and buy a million, but I figured you would much prefer to see the interesting principle they applied to use a HV transistor instead of a FET, and how they operate it common-base so that the Miller effect (which also affects bipolar transistors) does not matter.
When I showed the circuit to my power pals, they pointed out the common-base does not eliminate Miller effect- Power guru and recent Maxim new-hire Jon Dutra noted:
- There is still a Miller effect even if the bipolar is driven common-base, unless the base impedance is zero (Which it is not). But the bottom FET does not see the high dV/dT so it is a faster way to drive, at the expense of part count and V*I losses when on. To Dobbys point, some clamp diodes are likely in order, as when Williams did the cascode in figure 6 of AN25. The corner cases, and line and load transients, will tell the tale of reliability. The data sheet of this device tells you more about their algorithm. It looks like it could work. I am surprised they ac coupled the FB pin. They don’t give a part number for the cascode NPN in any of their data sheets. At low power, an NPN cascode on top of a FET or NPN, should be fine.
And yes, when Jon referred to Dobby, he was commenting on a reply I got from Linear Technology founder Bob Dobkin. Dobkin warned:
- I have seen another problem with HV transistors. They are slow. If you turn off the emitter, all the current in the transistor has to go somewhere. Some comes out the base, some comes out the emitter. If you turn it off fast enough, the emitter will fly high and blow out the fet below it.
This jived with a comment I got from National Semi consulting scientist Bob Pease, who noted that you must have a diode in parallel with the dropping resistor to the HV transistor base (the way CanSemi does above). That way, if the base flies up, the diode will hard-clamp the base to the bias voltage, as opposed to letting it rise high enough to blow the transistor.
So I guess the points all my pals are making is that 1) it is a nice circuit. 2) whether it is cheaper than a single HV FET depends on your purchasing agent. 3) Make sure you hammer the circuit under line and load transients and put in clamps where necessary to make sure nothing blows up. As always, analog design is as tricky as it is elegant. Thanks to Jon and both Bob’s for helping me understand the subtleties the circuit.
Oh, and in a game-changer to the whole expense issue of HV MOSFETs, one of the great presentations I saw at the Fairchild Power Seminar last month was about their SupreMOS FETS. They use trenches in the die to round out the e-field, so the parts have a higher breakdown voltage. The big game-changer here is that this means that the relationship with breakdown voltage and die area is no longer based on a 2.5 exponent. It is a linear relationship, so that means we can expect HV FETs to drop in price over the next decade.
In addition to allowing HV FETs to become as cheap as HV transistors, this might also finally make 42-volt electrical systems in cars practical. Years ago, I got a great explanation from a smart fellow that works at Freescale in Arizona. Sorry I can’t remember, it might have been Aristide Tintikakis or it might have been Claude Escoffre. He explained that the problem with 42-volt automotive electrics is that you can’t really use relays switch circuits. Anyone who has played with 48V power will see it is more like a welder than a power supply. So the 48V dc will burn up contacts so you really have a hard time using relays. Thing is, MOSFETs are fairly cheap when they are 30-volt MOSFETs meant for a 12-volt car. But his point was that a 36-volt car would need MOSFETs rated close to or even over 100V. Those FETs are so expensive they have meant it is cheaper for carmakers to use 12V power and more copper, rather then all switch to 36V power (42 is the charging voltage, similar to a 12-volt car that charges and runs at 14V). So now cars can have LED lighting and switching power supplies that operate fine off of 42 volts. And if Fairchild and others can reduce the cost penalty for HV FETs, well then we might have 42V cars one day soon.