Adjustable current mirror

Hide quoted text -

th

q.

d

Yeah it's those 'little' ones that I remember. Dang useful thing to have around the lab. You can still get the little ones with analog meters.

George H.

It isn't a Gilbert cell, it's a transconductance multiplier. Four- port current multipliers like Raytheon/JRC/Fairchild RC4200 are exactly what this calls for. Buy a mass-produced one, they're cheap enough.

This is the old data sheet:

It may be an obsolete part number, but one can scrounge 'em in onesies. I'm not finding any clear answers on searching the web for a modern replacement.

The 'opamp thing' current mirrors use, alas, a current sense resistor, and are limited to maybe four or five decades before offset voltages or power supply limits cause nonlinearities. Regular old matched-BJT current mode devices can get six to eight decades of linearity. The important thing to remember, is that the matching of transistors to 1% or so on the transconductance at 1 mA is also matching of those transistors at 1 uA, maybe even 1 nA. Matching I*R drops doesn't scale so well.

Don't they use Gilbert cells inside?

Four-

He wanted a 20:1 mirror ratio, which may be hard to find in a monolithic transistor mirror. With an opamp having a 5 uV offset - not hard - and a 5 volt max drop in the resistors, the offset error is 1 PPM of max output. The 20:1 ratio then just depends on the resistor ratio.

Do classic transistor mirrors really hold up at uA and nA currents?

John

ide quoted text -

h .

hack something together with an opa549?

8A, thermal shutdown and current limit

-Lasse

Bipolar mirrors flake out, due mostly to the ISE parameter Kirwan asked about (recombination). CMOS mirrors can get down there, but physical offsets start raring their ugly heads. ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
I love to cook with wine.     Sometimes I even put it in the food.

Well, in that triangle wave generator I posted, I observed reasonable (i.e. not ludicrously bad) results. It did 1Hz to 4.5MHz on 680pF and about

3Vp-p, what's C*dV/dt on that? The duty cycle was perhaps 40-60% at the lowest frequency (nearly 50% higher up), which means the currents were still matched to the tune of +/-10%.

Since everything is leaking, that could just as well be the difference between leakages of the 2N4401, 2N4403, 1N914 and 2N4093. I guess I would expect leakages to be more mismatched than that, but I don't really know. At any rate, it looked good *near* cutoff, so even a discrete current mirror (at least with "lucky" parts choice?) can work quite low.

I should go put that circuit back together. I have access to all the parts in the EE lab. I don't seem to have any scope shots of the low frequency waveform from back when.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

No, just matched transistors and op amp. The RC4200 datasheet is worth a read. I think their op amp input currents are the low-I limiting factor.

Yes, with low-noise BJTs (high beta discrete transistors); the failure of Ebers-Moll is only due to leakages at the low currents, and due to parasitic resistance at high currents. That emitter-resistor is a sneaky culprit, all general purpose transistors have some emitter degeneration to keep 'em happy with high currents, and it's NOT as nicely matched as one would like. Rbb likewise has some effect.

Turnoff is quicker with leakage/base recombination high, so switch transistors don't do the low current well; high current drive is better with emitter degeneration, so 2N3055 isn't stellar, either. Matched quad transistors are usually the low-noise type, though.

Sounds like about 4 nA to 20 mA.

That's pretty impressive for a circuit that experts have panned. I guess the frequency is exponential on pot rotation at the low end.

If you have leakage, it's probably the 1N914s. Transistors, specifically their c-b junctions, make much better diodes than diodes.

You don't need the diodes; the schmitt+PNP could short out the upper current source.

John

Jim Thompson a écrit :

Some behave really well. While my next prototype is baking at 125°C I just pulled a random BC548C out of the junk box and tested it.

VCE=5V, IC Ib Beta

5.3m 10u 530 1m 2u 500 489u 1u 489 90.5u 200n 452 43.2u 100n 432 20.5u 50n 410 3.44u 10n 344 1.55u 5n 310 523n 2n 262 228n 1n 228 112n 500p 224 38n 200p 190 117n 80p 146 62n 50p 124 1.8n 20p 90 740p 10p 74 330p 5p 66 130p 2p 65

It's not that bad.

--
Thanks,
Fred.

Fred Bartoli a écrit :

BTW it's fun having a silicon BJT working with something like 12mV base bias :-)

--
Thanks,
Fred.

Repeat at +125ºC ;-) ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
I love to cook with wine.     Sometimes I even put it in the food.

Yes, of course, but not too many got that high. And the oven is occupied to something that earns money. But I'll do it when it idles...

--
Thanks,
Fred.

Picoamps don't perform work. :-P

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Tim Williams a écrit :

Hey, 12mV*2pA=24fW Since the begin of universe that's about 10kJ. Not nothing :-)

Giving me some idea. I might try to build the world's lowest current, lowest supply voltage, hence lowest power, low frequency astable oscillator, just to set the record and use those drained batteries to something really useful, for still a long time.

--
Thanks,
Fred.

Sounds maybe like an RC relaxation oscillator that uses some super-low-current Schmitt sort of thing to discharge the cap, sort of a mosfet-based PUT or something.

1 nA, 1 Hz should be easy. 1 pA would be a challenge.

John

I seem to recall JT had posted something like that. I think it had ~20G resistors, so current ~nA.

Funny thing about "0.7V" is it's just because we like Manly currents, like one miliampere. A bunch of decades down, and it's a lot teensier, roughly ln(1e9) times. Mmm, that's not quite correct, and 12mV sounds way too low. Lesse. If Vbe = 0.7V at 1mA, then Is = 1mA * exp(-0.7V / 26mV) = 2fA. So then at 1pA, V = 26mV * ln(1pA/2fA) = 0.16V.

Did you actually measure 12mV? If so, then I guess that means it'll have to be underbiased a bit -- like germanium. In fact, same physics, just a wildly different order of magnitude. Pick the right current range and temperature, and you can "zero bias" a BJT, just like in the germanium days.

I'd like to see an LC oscillator at this current. It's gonna need about a million turns, of course, but maybe that wouldn't be too awful to print. Let's see, for Fo = 1Hz, Rloss ~ 1G and Q = 10, parallel resonant, so Xc = X_L < 0.1G, or let's say L = 16MH and C = 1.6nF. If you can make a small core with A_L = 1uH/T^2 (I'd recommend sputtered permalloy), then you'll only need 4 megaturns. At 1nA and a typical current density of 155A/cm^2, you're looking at about 650 nm^2 cross section, or a 25 nm square hunk of copper, which is well within modern process capacity. Let's say you "wind"

100 planar layers, each containing 40,000 turns. Including a generous 25nm of insulation in all directions, you need 2mm width and 10um height, plus whatever size the ferrite gets to.

You probably don't have to worry about saturation too much. Magnetization is 1nA * 4Mt = 4mAt, and if the path is about 5mm, you get only 0.8 A/m. Even supermalloy takes 1-2A/m to get into saturation, so that's not too bad. Oh, as long as I'm diddling with geometry, I might as well verify inductivity: if B = 1T from H = 1 A/m, and the core cross section is maybe 2 x 2 mm square, then the flux*turns is 16 V*s, which should be plenty. The inductance is 16 V*s / 1nA = 16GH, so a smaller core or fewer turns could be used.

This seems surprisingly doable, if rather expensive. The sheer number of layers would be expensive, but at least the middle layer masks can be identical. I wonder what the best construction method is -- I suppose you wouldn't use plastic on the 25nm scale; then again, what's wrong with directly using the UV cured mask resin as printed? Deposit some, sputter some copper, cut off the peaks, deposit some more, etc. You'd start with a "start winding" mask which takes the start lead out, and this layer is deposited on top of a permalloy base (might as well make the whole thing out of permalloy, plastic and copper). Same idea on the topmost layer: take the winding out, then cover with a heavy layer of permalloy. Nice thing about deposited metal, you can make the perfect pot core.

Such a device wouldn't be integrated on silicon, but it could be placed in the package nearby. You aren't getting high density integration at this current level, anyway; we're talking big (~1mm) transistor dies running at picoamps here, so the current density is miniscule and the voltage drop is small.

Speaking of low voltage stuff, it could be very interesting to use transistors at currents near Is (voltages near Vt). Hook up a couple layer thermocouple to such a chip, cool one side and heat the other, and *veeery slowly*, have computation flow out of it. A true thermal computational machine! It's too bad there's no coherence of physical states, it sounds a lot like a quantum computer -- give it a teensy bit of heat energy and wait for the equilibrium state to evolve!

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

On a sunny day (Fri, 12 Feb 2010 16:20:16 -0800) it happened John Larkin wrote in :

Just use the output of a LED exposed to outside light, it will have a 24 hour period time. No battery needed, frequency 11.57 uHz. Bit noisy, but you can always wind an LC filter as Tim mentioned LOL.

The problem then becomes Q.

PHEMTS are interesting for low (sub-volt) stuff because they have high transconductance around zero gate bias. I have some curves... I'll see if they might have usable gain in their ohmic regions. It wouldn't be low current stuff, but might be able to run from a thermocouple.

John

Yes, if you use small geometry devices whose beta holds up down there. BFG25A/X is an excellent choice, BFT25A for an earlier and slower model.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net