I was looking at power transistors to use to control a load of some tens of watts (a tiny arc furnace) from a microcontroller, driving a couple of flybacks with something like 10 amps at 5 volts at a few kHz. But obviously the microcontroller can’t drive 10 amps, so you need a high-power buffer, and it should probably be a transistor.
What buffer transistors would you want to use?
In MOSFET-land, a pricey power MOSFET like an IRF540N is one option when you need low on-resistance; (0.044 ohms, 33 A, Vds up to 100, 145¢); another would be a parallel pair of IRLML6344s (0.029 ohms, 5 A each, 30 V, 36¢).
An interesting figure here is the expected load impedance. An IRF540N can control a 3300-watt load, but if it’s resistive, that load has to be 3 ohms. If the load is 10 ohms, at 100 volts it can only draw 10 A, and so it’s only 1000 watts. Similarly, if the load is 1 ohm, we can only use voltages up to 33 V, and so it can only be 1089 watts. So even though the actual impedance of the IRF540N is only 0.044 ohms, from the point of view of efficient power transfer, it’s kind of like a power supply with a 3-ohm internal impedance. I’ll call this the “virtual impedance”.
In general if the virtual impedance is too low, you can stack up multiple switches in series to get the voltage you want. It may be inconvenient but it’s probably not that bad. But if it’s too high you may be in for trouble; BJTs need a lot of ballast to avoid current hogging.
Because you can parallel MOSFETs, too high a virtual impedance figure is less of a concern; paralleling them drops your virtual impedance just like it drops real impedance. So one IRLML6344 has “6 ohms” of virtual impedance, but two in parallel have “1.5 ohms”, and four have “0.75 ohms”.
The modern choice would probably be a GaN FET like the EPC2036. Damn, those things are sweet. 100 V, 1 A, 0.065 ohms, and only 0.91 nanocoulombs of Qg, compared to the IRF540N’s 71 nC and the IRLML6344’s 6.8. And its threshold voltage is lower, too. So you can switch it on or off much faster and with less energy. The 245¢ EPC2016C is 100 V, 18 A.
You can’t use a TIP120 darlington for this kind of thing; though it can deal with 60 V, 5 A continuous, 8 A peak, its saturation Vce is 4 volts at 5 amps, so it would eat basically all the power you were trying to feed the transformer.
60 V / 5 A is a virtual impedance of 12 ohms.
So are there bipolars that would work better?
Digi-Key suggests the 55¢ 2STN1550, the 141¢ MJB41CT4G, the 232¢ MD2001FX, the obsolete 21¢ 2SD23210RA, or the obsolete 32¢ 2SD250400A.
Starting with the cheapest, the 21¢ Panasonic 2SD2321 can switch 5 A (8 A peak) at up to 20 V with a beta of at least 150 at 2 A and a typical saturation Vce of 0.28 V when you’re running it at 3 A, zooming up to 1 V at 8 A or so. It has a 150MHz “transition frequency”, which I think means its beta is guaranteed to be not more than 15 if you’re running it at 10 MHz.
20 V / 5 A is a virtual impedance of 4 ohms.
So, you could feed it 40 mA through a 120-ohm base resistor from a microcontroller GPIO pin, grounding the emitter, and running the flyback primary winding through it from 5V. The current through the coil starts to climb, and keeps climbing until we turn off the transistor, at which point the current leaps to the secondary and energizes the arc. If we don’t buffer it further we probably won’t get more than 6 amps out of it. But then it’s dropping a whole volt, so it’s dissipating 6 watts, briefly exceeding its 0.4 watt rating 15 times over. And it probably spends a fair bit of time dissipating more than half that. So it’s probably going to overheat in its little bitty NS-B1 TO-92-like package.
But, even before that, quite likely at 5 V 2 A we drive the poor little transistor into second breakdown.
The max you could theoretically switch with this transistor, if second breakdown wasn’t a consideration, is 100 W. In a flyback setup you won’t get more than 40 W; the flyback waveform is an interrupted sawtooth, so its RMS value is half of its mean value, a quarter of its peak value of 160 W. With a 5 V supply, though, you’ll be lucky to control 10 W with it. And because of its lousy power dissipation you can only control a tiny fraction of that continuously.
This is slightly more promising, specced to switch 5 A (9 A peak) at up to 10 volts, and dissipate 750 mW from its TO-92-B1. But its saturated Vce crosses 1 V at only 4 A; at 8 A it’s up to 2 V (and thus 16 W). So it’s just going to dissipate way too much power for this, even if it doesn’t hit second breakdown (Panasonic forgot to include the safe-operating-region plot in the datasheet this time).
10 V / 5 A is a virtual impedance of 2 ohms.
This is a little bitty surface-mount SOT-223, which entitles it to dissipate 1.6 watts, and it’s rated to switch up to 5 A (10 A peak) at up to 50 V; at 5 A 5 V it says its (non-small-signal) beta is typically 95, so you’d need 53 mA to avoid saturating, which is a bit much to ask from a microcontroller. It turns on in 90 ns and off in 700 ns, so you can switch efficiently at near-MHz rates.
50 V / 5 A is a virtual impedance of 10 ohms.
ST omits the performance curve plots entirely, as it turns out, only specifying a 0.26 V saturated Vce at 3 A.
I feel like this transistor would probably work in a 12V system! But in a 5V system we’re just asking too much current from it. Say we can drive its base with another transistor so we don’t have to worry about the 40 mA limit on AVR pins. 5 amps at 5 volts is a peak of 25 watts; in a flyback setup we can never get more than half of that, since if the mean current of the sawtooth is 5 amps, its RMS current is only half of that. So we’re talking about 12.5 watts, which is not much of an arc furnace. If we were using 12 volts, though, we could do 30 watts with a switch like this.
I don’t know, I think we’d need to go for something a lot beefier to get hundreds of watts of power into our stepdown flyback arc power supply.
This is a TO-263-3 surface-mount version of the TIP41 power transistor, specced to switch 6 A (10 A peak) at 100 volts, but with a beta of only 15 and a transition frequency of only 3 MHz (which the onsemi datasheet helpfully explains is the gain-bandwidth product), which limitations would be fine for this application. You’d have to use some kind of a driver circuit to drive its base: another transistor, a step-down pulse transformer, something.
100 V / 6 A is a virtual impedance of 17 ohms.
It’s rated both for 2 watts and 65 watts dissipation, depending on whether you’re holding the case or the ambient air at 25°. The junction temperature max is 150°, junction-to-case thermal resistance is 1.92°/W, and junction-to-ambient thermal resistance of 62.5°/W (or 50°/W “when surface-mounted to an FR-4 board using the minimum recommended pad size”). If you divide 150°-25° by 1.92°/W you get the 65-watt number, while 62.5°/W gives you the 2-watt number. So if you heatsink this guy well enough, you could dissipate tens of watts.
It says its saturated Vce is 1.5 V at 6 A, which would be 9 watts, so that’s really all the heatsinking it needs if you’re using it as a switch. At 5 volts it would be grossly inefficient, controlling a 3.5 V 6 A (21 watt) load at a cost of 9 watts. But if it were controlling, say, a 70-volt load, it might be fine.
This is a monster 700-volt bipolar 12-amp (18-amp peak) 58-watt NPN BJT, marketed as “High voltage NPN power transistor for standard definition CRT display”! It’s a transistor specifically designed for the horizontal deflection output for a CRT, but astonishingly it’s not marked as “obsolete” and it was only introduced in 02007. Its beta is only 4.5, and it’s sloow, 2.6 microseconds storage time.
700 V / 12 A is a virtual impedance of 58 ohms.
Suppose you connect the flyback primary between the positive power rail and a capacitor to ground and put a triac (or just an SCR) across the capacitor. Initially when you plug it in there will be a flyback pulse as the capacitor charges up, to twice the input power rail, I think, but then when the cap starts to discharge, the flyback secondary’s diode will go forward-biased and rapidly drain most of the energy out of the circuit, so fairly rapidly the cap will be charged to the power rail voltage, and everything will be quiescent. But if you tickle the SCR gate, the cap dumps to ground and the whole cycle starts again. You can control the amount of power delivered to the output by doing this more or less often. A triggered spark gap could maybe substitute for the triac in a pinch. I mean basically this is just a Tesla coil.
This doesn’t seem like it is going to be very efficient, but it would definitely work.
I am going to assume that the 46¢ Ween (formerly NXP, formerly Philips) Z0109MN0 is a typical triac. It’s an SOT223 four-quadrant triac that starts passing 1 A at 600 V (dropping 1.3 V) if you tickle its gate with 1 V and 10 mA, until the current drops below 10 mA (or maybe 30 mA?).
600 V / 1 A is a virtual impedance of 600 ohms.