[volt-nuts] Precision Current Source

[Andrea Baldoni]

On Mon, Aug 16, 2010 at 10:54:12PM +0200, Dr. Frank Stellmach wrote:

> I have made the design about 20 years ago, so I did not remember  
> everything right..

I really appreciate very much the informations you gave me.
My comments are made to better understand and share opinions, but I did not
wanted to criticize your design, that anyway already revealed being optimal.

> The BS170 MOSFET has a typical leakage current of 10pA only, have a look  
> on the onsemi datasheet:  
> http://www.onsemi.com/pub_link/Collateral/BS170-D.PDF.

Hmm. I had looked at the datasheet of the fairchild BS170, and it just states
"gate body leakage, forward" max 10nA, but yes, the ON datasheet also state a
typical figure of just 10pA (gate reverse current, a definition for a JFET).

In the data sheet of the BS108 for instance, there is just the max value of
10nA and all other small signal MOS from ON I looked have 100nA.

No other MOS ever claims for a typical leakage value, only max.

It would be very interesting to test it.

> That's similar to what I have found in my old design notes about the  
> current source. I think I have tested this parameter somehow, but I  
> can`t remember any more, how it's been done.

I found in a forum two approaches. Measuring the time a known (and known
leakage) charged capacitor discharge itself on the gate (using the conduction
of the channel as detector) and a more traditional one with voltage source,
high value resistor and very high input Z and low drift AMP OP to measure
voltage drop over the resistor.

Probably you can combine the two. Very high resistor, variable voltage
source, go high until the MOS start to go in on state...

> It's mandatory to keep the FET and the OpAmp bias currents well below  
> 1nA, better around 10pA each, as those currents degrade the precision  
> especially at 10?A.

I agree, and my main target is 10uA.

> I also added a manual decade divider by ten precision resistors to get  
> 0, 0.1, 0.2, .. 1 V, so the smallest current is 100nA.

May be very useful in a lab, even if going so low force to take into account
moisture, cable insulation (teflon?)...

> thermoelectric voltages, have a much bigger impact on precision of the  
> current source.

I'll treasure this.

> I just scanned the actual Lake Shore, Keithley, and Time Electronics  
> sites and could not find any programmable current sources being that  
> precise like my old design, and costing that little.

Sure low cost is not a goal of those manufacturers ;)
..the same reason because I started with a circuit and not buying a
thermometer :)

> I can't figure out how to realize the idea of Bruce, as you need a 4W  
> connection to the sensor, and also need compensation of thermoelectrics.  
> This cannot be accomplished by sort of a ratio bridge.

If I have understood correctly, you use the force connection from the
multimeter to derive the reference voltage and place the sense connection
over the sense cables of the DUT. In the middle, the current source circuitry,
based on the reference from the multimeter, feed the DUT from another couple
of force cables. It's all 4W, 4 at the multimeter and 4 at the sensor.

To zero it, you could disconnect the power.

> I realized that you bought a PT1000, so the 4W is in some cases not  
> urgently necessary, but as you must apply 100?A max. instead of 1mA, the  
> problem with the thermoelectrics is absolutely the same as for a PT100.

I agree.
I bought a PT1000 and a PT100 in effect, just to see which one I could use
to obtain the best.

> You may use your 10k Vishay to calibrate the 100?A current source, as  
> the 34401 has about 35ppm in DCV / yr. and the resistor is precise to  
> 0.01% max, which gives about 150ppm current accuracy, compared to  
> measuring 100?A in the mediocre 10mA range of the 34401, which is stable  
> to 500ppm / yr. only!

I used the 34970A to measure 100mA-1A range currents and immediately switched
to external shunts. Even hand made ones winding nichrome wire are much less
drifty (by self heating) than the internal ones. The 34401A is no much better
in that, yes.

> Reason for that: Current shunt R121 is a 5 Ohm resistor, used for 10mA  
> and also for 100mA.
> On latter range, this gives 50mW power dissipation, therefore the  
> resistor had to be a TK0 metal film type, obviously. But this thing  
> drifts over time like hell. So I replaced it by a TK3 wirewound, giving  
> more stability in time. They saved a lot on quality for this instrument..

Interesting modify, I like external shunts anyway more... at least for the
reason that a short would not burn your instrument. Using the 34970A with
external shunts is good also for switching relays reasons.

About the 34401A quality... well, in effect it's cheap... It could reserve some
surprises anyway. Today for instance I has been delighted with the 10.0000K ohm
stable reading :)

> agilent improved the current ranges on the 34410/11 models by reducing  
> the 10m/100m shunt to 2 Ohm, and adding 100?/1mA with an additional 200  
> Ohm shunt.

Never tried one of these, and I suppose I would never buy one. If I'll buy
another multimeter, the next one will be a 3458A or whatever they substitute
it with (after careful inspection that the latter is not worse as often
happens...).

> Well, the 3458A is superior because it has 8 different ranges, each  
> having its own, well suited precision shunt.

Ditto.

> Resistance of the 34401A drifts 100ppm in one year, even cheap wirewound  
> resistors may drift 20ppm / yr. only.

So the internal current source is not so good in the long run. Maybe is
adequate for short term, but not better than using a (good) external reference.
It would have been nice if they have provided a connector to export the internal
voltage reference to external circuitry...

> In comparison, the only component inside the 3458A, which contributes to  
> precision of all resistance ranges, is its internal 40k reference.

Definitely the 34401A is a very cheap design compared to the 3458A.
I suppose this is a trend in Agilent (you could say the same with recent
frequency counters versus older ones). Maybe this is the reason they still
don't substitute the 3458A? They have no more engineering force to make new
very-high-level products (or maybe there is no more the market for them)?

Some say that the move from HP to Agilent has marked this trend, even if
the 34401A has been born HP and the same is for 53132A.

> So you may better use you nice Vishay metal film as your resistance home  
> standard (also 20ppm/year?).
> You only have to make precise transfers to other ranges. (Just another  
> project)

Or buying other Vishay resistors.
They are not cheap, but probably cheaper than another gizmo :)
..that I would like to make anyway, just to compare them each other ;)

> Lake Shore offers a 2 point SoftCal for 0.25K precision up to 300K, for   
> "just" 214$, including the sensor.
> If your current source works fine, this bargain may be interesting..

Interesting, yes. My sensors actually are much cheaper, but obviously are
uncalibrated. I will face the calibration problem once the system'll begin
to work :)

Best regards,
Andrea Baldoni