Is a MOSFET really a good current source???

Depletion mode FETS (MOSFET or JFET) make good current sources. Just tie the gate to the source. Since virtually all JFETs are depletion mode, this is probably your best bet. Most MOSFETS are enhancement mode.

Your data, IIUC, is wrong. MOSFETs in fact have a highly constant current vs drain voltage, for a fixed Vgs. If you at first measure otherwise, recheck your setup until you get it right. For example, be sure you that don't have Vds = 0, or under say 1 to 2 volts. Natch. :-)

You are well down at the bottom end of the scale for such a device. Have a look at the dirty sheet for the set of ID vs VDS plotted WRT Vgs and you will see that you are operating the device in it's linear (resistive) region rather than its saturation region...... where the curvy bits are bending down to zero rather than being flat.

I think that's right.

Down there it behaves as some sort of square law resistance depending on something to do with Vth, Vg and a possible K.

DNA

Ah, perhaps I'll retract that. Mr Win is on the case and I seem to have got it wrong..... :-(

DNA

Channel-length modulation is the culprit.

It's also not clear to me that your definition of impedance is deltaVDS/deltaID... DC numbers are meaningless.

...Jim Thompson

--
|  James E.Thompson, P.E.                           |    mens     |
|  Analog Innovations, Inc.                         |     et      |
|  Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
|  Phoenix, Arizona            Voice:(480)460-2350  |             |
|  E-mail Address at Website     Fax:(480)460-2142  |  Brass Rat  |
|       http://www.analog-innovations.com           |    1962     |
             
I love to cook with wine.      Sometimes I even put it in the food.

Genome a écrit :

For this time it seems Win did read a bit too fast. (believe he took the 'M' for milli). The figures seems pretty reasonable to me and 15K at 7mA Id and 75V Vds is to me a good figure.

To the OP, if you want to increase the dyn impedance you'll have to add a source resistor and set the bias such that Vbias = Vgs + Rs*Id This would be necessary for good thermal stability and some reproducibility too.

And since it's for toobs a few volts don't matter much.

You'll have to select low capacitance mosfets too if you want to get past a few kHz BW.

WRT this, PNPs might be a better choice.

--
Thanks,
Fred.

He said he was using AC current for the measurement.

--
Thanks,
Fred.

How are you measuring Zd? You have to measure partial Vds/partial Id for *fixed* Vgs. From a quick squint at the numbers, it looks as though you're calculating Id/Vs as you vary Vgs, using about a 150V supply that is sagging as you pull current out of it. That's a misunderstanding of what you're trying to measuring. You're basically measuring the large-signal *source* impedance, which is 1/transconductance, and should show the behaviour you're seeing.

A simple way to measure it properly with one DVM would be to put a resistor between the source and ground, apply a floating Vgs using a pot and a battery, and compare the 120 Hz ripple current you measure in the resistor to the ripple voltage on the drain. (From the sag, I'm assuming the supply is unregulated, which means it probably has enough hum for the measurement.)

Cheers,

Phil Hobbs

everyone else:

Didn't you look at the numbers, and compare them to some data sheets?

daceo:

10-50K is about right for a small-signal MOSFET, so these numbers look good to me. If anything they're better than I would expect since you're using (I think) a power MOSFET. Considering that 15K at 7mA would require 105 volts from a resistor and much less from your MOSFET I think it's not too bad.

You'll get a higher impedance if you use some source degeneration, with the drain impedance increasing by the proportion that the apparent source impedance increases (i.e. if you have a 20 ohm source impedance then a 20 ohm degeneration resistor will double the drain impedance). If you keep the degeneration resistance significantly higher than the source impedance it'll also make the drain a more linear current source.

o | | | ||-+ ||

One-point values are not useful. Apply Vgs of some amount, say 3.1V and sweep the drain voltage from zero to near breakdown; repeat for 3.2V, 3.3V, etc to say 4.5V. The idea is that one will see a flat Id VS Vds above some voltage; that "flatline" being the hi-Z area that you are looking for. As the Vgs increases, the current "flatline" will increase, but will still be flat. Granted, the IV curve will look like a low value resistor (value decreases as Vgs increases) until it curves to the "flatline", but that is the area you do not want to use. A curve tracer is rather useful here, as the drain current during "flatline" will change as a function of temperature (it is sensitive, like Vbs in a bipolar) - so that at higher powers (meaning over 10mW in a TO-220 package) point measurements become useless. Handling the device will mess up the measurements (hot hands, cold heart?).

A Jfet does *not* have a "flatline" current VS voltage curve anywhere! A depletion mode MOSFET is far superior!

And do not heat it up with hands or "large" IR drops.

Actually, MOSFETs are good to the (tens of) nanoamp region; just beware of the temp sensitivity (at all currents).

Since a PNP is the compliment of an NPN, does that mean we look for the Late effect?

Hi and thanks for al the replies,

Yes JFETs are quite good as current sources, but in my set up the mosfet sort of adjusts its current to whatever is required at the voltage I want it to be at, and this works quite well.

Yes I am right down in the noise and off the scale with regards the data sheet, and I feel that I am in uncharted territory for me.

Yes I suppose I am getting a benefit with the mosfet over a resistor in increased impedance, I am sure I have seen people quoting Meg ohms for similar circuits (frustratingly never kept the link) under similar conditions..... and I did not expect the impedance to change with current so much either, never mind, you live and learn!

I initially found the problem when looking in to the AC impedance (low capacitance mosfet comments noted), so I set out as a sanity check to do steady DC measurements, from which I calculated the results above, which roughly reflect my AC measurements.

Anyway, I measured and plotted my own data using lab psus and dvms, for increasing steps of Vgs of 0.1 volt and Vds of 10, 20, 50 and 100 volts (assuming no current is flowing at 0 volts), and these are the results.

Vgs 0 10 20 50 100 Volts

2.61 0 0.0032 0.0034 0.0039 0.0049 mA 2.71 0 0.0072 0.0078 0.0093 0.0119 2.81 0 0.0175 0.0195 0.0235 0.0303 2.9 0 0.0444 0.0499 0.0606 0.0794 3 0 0.1144 0.1295 0.1589 0.2101 3.097 0 0.3025 0.338 0.4268 0.5584 3.193 0 0.7781 0.8782 1.0949 1.58 3.289 0 2.06 2.33 2.91 3.99 3.386 0 5.06 5.86 7.37 10.69 3.48 0 10.83 14.57 20 30.32 3.58 0 28.36 33.65 50.66 140

I think the channel length modulation looks like a good match to what I am seeing, I will have to read up on it, a brief look at data sheet however, and I can not see it quoted (lambda?). Is this the kind of thing that may change with manufacturer even though the device is nominally the same?

I am quite interested in the floating battery gate drive method of measurement. Sounds fun!

Yes I am using a power mosfet, IRF730 and it is interesting that you think I am doing quite well with this type of device, I do actually have a bit of degeneration included in the way of a current sensing resistor, 43 ohms and some other bits which probably help.

Yes, thermal effects are what I think I am seeing when I get the Vgs above 3.4 volts, I took the precaution of screwing the device to the biggest lump of heat sink I could find (about 1kg). I have had a run in with the negative, ehem or is that positive temp co? in these devices at low currents.... Another headache, from the past!!

Curve tracer, sounds nice, found a simple circuit for one that I was going to build, but is on hold at the moment... (it started attractively simple, and then I thought what if, and could I do that, and it rapidly became less simple!!....)

With regards to a PNP it could work but it would make my circuit more complicated, and its quite nice with no base current. May ultimately be the way to go...

So my situation now is to do some more measurements to see what happens at higher currents, just to see if it flattens off, build (borrow or buy) a curve tracer, and try and find the channel length modulation number for the devices I would like to use. Always something new to learn!!

Cheers For all the input, lots to think about

Daceo

No, I stand by my remarks, based on the theory for ideal MOSFETs and on many hopefully-precise measurements I've taken. But Fred, you've thrown down the gauntlet, so I'll drag out my data, and perhaps add some new measurements on that old part, the IRF730.

One caution when using high-voltage MOSFETs in the linear region: they love to oscillate at high RF frequencies! This can certainly cause changes in their observed low- frequency characteristics. A small series gate resistor to isolate the recommended gate-source protection zener, and other precautions are in order, along with checking their activity in the 20 to 80MHz region with a scope.

I agree. As a longtime MOSFET fan, who loves to use them in low-current linear applications, and who is happy with their sub-threshold theory and measured properties, I'm constantly struggling to find small enough parts (small die, that is), and have to admit that BJTs often show superior performance.

I have rechecked my data, and stand by my assertions. However, let me remark, my data was for high-voltage MOSFETs operating at 50% or less of the maximum voltage rating. My measurements above 50% show continued textbook low-leakage performance for many power MOSFET types, continuing nearly to their avalanche voltage, and abysmal leakage performance for some others. I had poor 1000V mosfets (the manufacturer shall remain unnamed) that had 50x higher leakage at 500V than a 600V mosfet. But we can't blame such things on "poor current-source behavior".

I did not have time to test any IRF730 parts. But anyway, given all the different manufacturers, and the different fab facilities over the years, as they strove mightily to probe the bottom floor of production costs, I imagine it'd be a real crapshoot to see what leakage results one came up with.

Let see the figures presented by the OP and compare them to a simple resistor giving the same current:

Vds is 75V.

That's an already awful current source and there's little room to improve that awfulness (except under current multiplication-avalanche the dynamic resistance can't be worse than a true resistive behavior).

--
Thanks,
Fred.

Interesting addition to the data.

I agree the OP's data looks horrible. I'm saying that's either his particular "quasi-defective" MOSFET, or a measurement error of some kind. Even the poor 1000V mosfets I mentioned weren't that bad!

Sheesh, I can't have been supremely lucky in the measurements I've taken over the years? Actually, if the parts I've used routinely to make my high-voltage amplifiers, five or six different types from four or five manufacturers, several thousand MOSFETs in total, were that bad, surely I would have noticed. But I'll double check.