Build by Kittan           kittan |at| gekkoscience.com
FET Forward Transfer CharacteristicsThe guts of this dummy load is a large N-channel MOSFET. Initial testing was done with an International Rectifier part IRLB8743PBF, which is actually better suited as a logic-level switcher for high-current loads. It'll handle a Vds (drain-to-source voltage) of 30VDC and some ridiculous number of Idc (continuous drain current), at least 30A or so, in a TO-220 package.
The reason I say this device is better as a logic-level switcher than as an adjustable dummy load is the forward transfer curve. This curve relates the drain current to the gate voltage. A good switching FET will have a steep curve from zero to the threshold voltage, which ideally will be fairly low. Past the threshold voltage, the drain current is fairly flat at a fairly high value. What this means is, it doesn't take much voltage at the gate to turn it fully on, and it passes through the transition region quickly which minimized switching losses (as the FET isn't operating in the linear region for very long, going from fully-off to fully-on very very quickly).
This is undesirable in an active load. What you really want for an active load is a wide gate voltage range over the FET's linear region. What that means is, a small change in gate voltage will have a small change in drain current. Going from 2.5V to 2.55V doesn't mean drain current goes from 1A to 10A. A FET with a higher threshold voltage and flatter transfer curve below the threshold is a better choice, as it's a whole lot easier to control stably.
I ended up setting with the IRF540 partially because I already had about 70 of them running around, and partly because the forward transfer curve is a lot friendler.
Take a quick look at the curves there. On the 8743 the start of turn-on happens around 2.6V, and by 3V you're 11A. By 3.5V it's pushing 100A. Compare that to the much shallower curve of the 540 - turnon starts around 4V, by 5V it's a bit over 10A, 6V is seeing 12A.
Mind you the graphs are not created equal, as those are pulsed-current measurements with different voltage levels and pulse widths. But the general trend is, the 8743 turns on really gosh darn fast with a really low voltage, where the 540 requires a higher voltage and has a much lower I/V slope. The higher voltage part of it isn't necessarily the best, but the shallow slope means it'll be much easier to make small changes in drain current without requiring microvolts of precision.
A shallow transfer curve isn't necessary for automatic control but, as mentioned, it does reduce the need for extreme controller precision in order to get extreme output precision. If you're using manual control however, it makes things a lot easier. If you're using a potentiometer to determine gate voltage, you'll smoke a lot of 8743 by accidentally adjusting from 2.75V to 2.80V and accidentally quadrupling your drain current (which, at a 12V load, takes your power dissipation from about 30W to well over 100W almost instantly).
Current MonitoringOne necessary feature of a dummy load is knowing how much current is flowing through it. A piece of test equipment isn't terribly useful if you can't actually measure anything with it. There are a couple different ways to measure current in a circuit, but the easiest is to insert a known fixed resistance and measure the voltage drop across it. This is pretty easy to do in a dummy load, since it's just a high-powered variable resistor. Putting a small fixed resistance in series with the FET will allow you to measure the current without greatly affecting the operating range of the circuit itself.
If you're looking to put a resistor in series with a single part, you can either put it above or below in the circuit. Below the load (in this case, the FET) would be low-side measurement, as opposed to high-side measurement where the sense resistor is placed immediately at the inlet of the circuit, above the load. Both means have benefits - low-side allows you to measure directly, as the voltage will be ground-referenced already, and at a fairly low total voltage. High-side requires measuring the differential voltage (so measuring both sides of the resistor and finding the difference), which requires two leads and higher voltage handling. A benefit to high-side sensing is the FET is directly tied to ground, so the resistor's offset voltage doesn't add any offset to the FET's Vgs signal voltage.
I tend to go with high-side sampling, partly because I like having the load ground-referenced without worrying about current-dependent offset voltage. Another good reason I do it is the MAX4373 family of ICs, which are totally handy high-side current sense amplifiers with internal precision gain and a 28V input tolerance. This project will make use of the MAX4373FESA+, which has an internal gain of 50 (or Fifty; other options are TESA at Twenty and CESA at 100). I also have a bunch of 5W 0.003ohm resistors running around which will come in handy.
5W at 0.003ohm means a maximum current of sqrt(5/0.003) = 41A. I'll probably never ask that much out of it, since I'm testing at 12.8V and don't really feel like asking my FET to handle 525W.
With a sense resistance of 0.003ohm and a gain of 50, the MAX4373FESA will report 150mV output per amp throughput - if I ran the whole 41A through the circuit, my FESA would tell me 6.15V
Direct Gate ControlThe first thing I tested, which proved to be somewhat disastrous partially because of steepness in transfer curves, was direct manual gate control using a potentiometer. If you have very good dexterity and are watching everything very very closely it can be possible - within reason.
My first test I had 3 FETs in parallel and was watching the gate voltages and currents closely. At some point though, I must have accidentally twitched and rotated the pot an extra degree, which sent me shooting up the transfer curve and ended up almost instantly roasting all three FETs. Gotta hand it to the PSU though, it was trying so hard to keep up with an almost-dead-short load that my 12V lead wires were actually bouncing up and down from the current bursts.
There must be a better way of controlling a FET than manual gate drive. How about automatic gate drive? Surely that's a real thing.
Basic Characteristics of an Op-AmpIf you don't already know, Op-Amps are just almost the best electronic thing since sliced bread. I mean, electronic sliced bread. These guys have been around since the days of vacuum-tube computers, and are basically a differential-amplifier input coupled with some other sexy drive circuitry that make engineers' lives so much easier.
The basic characteristics of an ideal op-amp are two (differential) inputs with infinite input impedance, a single output with zero output impedance, and infinite open-loop gain. A good op-amp will actually be a fairly close approximation of these things, oftentimes seeing 10^7 ohms input impedance, less than 20 ohms output impedance, and an open-loop gain of around 100K. Pretty cool, huh? Sure, but what does that actually mean? Go learn that somewhere else, I guess.
The two inputs on an op-amp are differential, meaning the amplifier actually acts on a difference in voltage between the two. They're called the inverting (-) and noninverting (+) inputs because of their effect on the output. An increase in voltage on the inverting input will tend to cause a decrease in output voltage, and an increase on the noninverting will cause an increase in output voltage. That is, the noninverting has a direct effect on the output, and the inverting has an inverse effect on the output. Makes a lot of sense.
The main goal of an op-amp is to make the two inputs equal. If you put a voltage on the + input and tie the output back to the - input, the output will be exactly equal to the + input because that's all it requires to make the - input match it. You can do all kinds of crazy feedback loops if you want to, which make op-amps some of the most useful and versatile circuits in electronics history.
The MAX4373 is actually an op-amp at heart. The op-amp measures the voltage across the sense resistor with its two inputs, and the feedback loop on the output is wired up as a voltage divider so a fixed portion is fed back to cancel out a part of the input voltage in a specific way. The result is the output gives us exactly 50 times the voltage difference between the two inputs, no matter what that voltage difference is. Pretty nifty.
Op-Amp Gate ControlOp-amps, as we have seen, shift the output voltage in order to make the two inputs equal. If you feed the output back to one of the inputs using some circuitry, the op-amp will drive that circuitry such that the result is its two inputs are equal. In the case of a common voltage amplifier (like the MAX4373 or a simple audio circuit) the feedback is most likely just a couple resistors. If the circuit is a switch-mode power supply, the feedback loop might be a PWM comparator, a drive transistor, transformer, rectifiers, resistors and an opto-isolator. It could be even more complex than that, so long as the output from that op-amp directly affects something which directly affects some other thing which eventually directly affects the voltage on one of its inputs.
We can use the op-amp's input-neutralizing ability to increase our precision control of the FET gate. By using the current sense measurement as one input and directly driving the FET's gate with the op-amp's output, we now have a feedback loop. The op-amp increases its output voltage, which increases gate voltage, which increases drain current, which increases the drop across the sense resistor, which increases the sense amplifier's output voltage. A decrease in the op-amp's output causes, eventually, a decrease in curent sense output. If we put our potentiometer on the other input, and use its voltage as a scale of the desired drain current, we now have direct control. We set the pot for a particular amount of current, and the op-amp changes its output until that much current is flowing, as reflected in the sense amplifier's output.
But the op-amp has two inputs, both of which have a different effect on the output. What do we do?
The key to stable feedback loops with an op-amp is "negative feedback". You want the loop to work such that, if something changes to inadvertently increase the output, the feedback loop will cause the op-amp to automatically decrease it and pull it back to where it needs to be. Positive feedback is self-reinforcing, which means it can spiral out of control really quickly; negative feedback has a "restoring force" which tends toward maintaining a stable level even under fluctuating conditions. So what we want is negative feedback, which since everything in the feedback circuit has a directly proportional effect (increase in output means increase in current means increase in sense voltage), we need to use the inverting input. If we want a constant current and the voltage increases, the current increases so the sense reading increases. An increase on the inverting terminal will cause a decrease in output voltage, lowering gate voltage and therefore lowering drain current back to the desired level.
Our "reference voltage", the adjustment we want the circuit to match, is therefore tied to the noninverting input. If you increase the potentiometer voltage, it's indicative of a desired increase in current. Increasing the voltage on the noninverting terminal will increase the output voltage, increasing gate voltage and increasing drain current and increasing sense voltage. Lowering the potentiometer voltage results in lowering the sense current. The extent the gate voltage has to change to make the appropriate change in measured current is something the op-amp automatically takes care of, which removes the need for your fingers turning the knob to have the required precision to match the transfer curve. The op-amp does it for you, all you have to do is set the current - a very linear operation.
Delay and StabilityOp-amps, as mentioned previously, can have very very high open-loop gain. This gain is limited by the operation of the negative feedback loop, but a consequence of it remains. Open-loop gain is related to slew rate, or the rate at which the output voltage can change in a short period of time. Usually it's very high, measured on the order of Volts per microsecond. If your feedback loop is pretty big, you might run into issues with, you might say, propagation delay - how long it takes a change in the output "signal" to show its effect on the input "signal". In our case we have the output pushing current into a capacitive load (the FET gate), which then has a voltage-field affect on the FET, which the current through it ramps up over time. That rampup is measured in the sense amplifier, whose output changes to reflect that measurement, which finds its way back to the input. Not much time passes, at most a number of microseconds before a change at the output is noticed on the input.
If the output of the op-amp has a high slew rate however, a large voltage change can happen in those microseconds. What that could mean, is the effect on current was quite a bit larger than the op-amp expected. The current will overshoot the target value, that information eventually making its way back to the input where the op-amp notices and starts to drop its output voltage. That drops the output current, but again the loop delay can cause an undershoot which needs to be fixed. This can cause ringing or oscillations in the current, sometimes pretty big ones.
One way to take care of this is to reduce the slew rate of the op-amp using a lowpass filter. A lowpass filter will buffer out rapid changes in the output, allowing only more gradual voltage changes to reach the FET gate, reducing the speed at which drain current readings can change. This reduces the circuit's overall responsiveness to external changes (like input voltage) but also makes it much more stable overall.
I ended up, partially through calculations and partially from testing, at a 470ohm/47uF single-pole filter. The cutoff frequency is (1/2piRC = 1/(6.28*4.7e2*4.7e-5) = 7.2) 7.2Hz. That's probably lower than it needs to be, and I might change it up before going too much farther into design as that could substantially limit the ability to compensate to rapid input voltage changes and current selection changes.
PrototypingHere we have what I built so far, complete with test leads and lots of mess.
Mounted to the large heatsink is the IRF540, point-to-point wired to a 12V input cable, the 0.003ohm sense resistor, and a 1000uF capacitor across it all to buffer out any oscillations. The sense resistor is measured through the grey and white twisted pair, and the orange wire is the gate lead.
From left to right we have: 1) the current-setting potentiometer (which in this configuration will swing from 0V to about 420mV, or 0A to 2.8A desired); our red lead is sending its value to my scope
2) the gate-drive lowpass RC filter
3) The gate-drive opamp, a TLV2460 rail-to-rail CMOS op-amp built to operate well on a single-sided 5VDC supply
4) The MAX4373FESA current-sense amplifier; the grey wire is sending its measurement output to my scope
5) 5V input from a bench supply; I could drive the circuit off the load voltage but it's easier to use an external source
I might post schematics someday if I feel like it, though I think there's enough information already presented to derive the schematics from the writeup content without a lot of effort.
Test Results and Future StepsSo far I haven't done a lot with this guy. It's hooked up to a 12.8V high-current supply (capable of sourcing 80A continuous) which is, admittedly, kinda dangerous. I did the initial control-loop testing with some 5ohm wirewound resistors in series with the FET to limit maximum current until I could work out some reliability issues.
Once the thing was proven with small loads, I mounted the FET to a bigger heatsink (yes, it's an Intel I-series heatsink turned upside-down, I got boxes of those things running around) and cranked it up to about 2.4A drain current - about 30W total dissipation. I let it sit on that load for about 30 minutes with zero issues.
The next test will probably involve two FETs operating in parallel, which I'll have to verify temperature curves for the specific part but it should be alright. If that works well enough, we'll move on to larger power stages and, eventually, digital control - a microcontroller's DAC will provide the current setting signal instead of a potentiometer. When we move on to multiple power legs in parallel, the DAC will be multiplexed with some nifty sample-and-hold chips so we can actually set each leg to draw a different current if we want to.
Eventually we'll probably end up with multiple FETs comprising a single power leg, with its own op-amp control and measurement circuit. Several legs will be driven in parallel, and will be fluid-cooled. In order to keep the legs separate we'll need to electrically isolate them from each other, so we'll probably use oil instead of water and directly solder the TO220 tabs to lengths of copper pipe connected to each other with rubber hose. That'll allow for very solid mechanical mounting and excellent thermal conductivity of the FETs without electrically connecting different legs. With each leg given separate current measurement and control, we should be able to maintain a constant precision from below 1A all the way to the maximum around 240A.
Definitely stay in tuned for more updates as this thing develops further. If it's solid and resilient enough, and there's enough demand for it, we might build some to sell.
|If you like what you saw and you want to contribute to future awesome projects, feel free to donate some BTC: 1GekkosciLeaey8Na9siC8oD5HcMtLnWwd|