Precision DC Current Source - DIY

[Dr. Frank]

Hello,

as promised, here are some photos of my old current source, 25 years ago used for scientific thermometry.

First picture shows the typical 4 wire Kelvin measurement, a 1k precision shunt serves as the DUT.

The source has 6 ranges from 1µA to 100mA, and an 11 step vernier.
The current can be switched off for measuring offset voltages over the DUT.
Ranges and On/Off could also be steered by a PC, using the Centronics parallel port (very oldschool and cheap!)

The assembly consists of the base PCB with PSU 5/15V, the trimmable, precision current range resistors (3ppm/K, 0.1%, wirewound)  in a shielded box, 6 reed relays for range switching, a 7650 ChopAmp plus BS170 MOSFET for low leakage current steering of the current source, and a REF02 as a stable 5.011V reference.

On the upper PCB, there's a 10 resistor precision reference divider (0.005%) for the vernier, in 0.1V steps up to 1V.

On the PCB, the different trimming resistors are shown.
Nominal f.s. reference voltage is about 1.002V, and each range resistor can be adjusted from about 1.0015 to 1.0025 times its nominal resistance value.
Trimming resolution is a few ppm for each range.

All range resistors feature 4 wire Kelvin sensing, by double throw reed relays on the upper side, and by the layout on the lower side.

One pole of the relays carries the current from the BS170/BD137, the other pole feeds back the voltage over the resistor, to the 7650 OpAmp.

The current source had been calibrated a few months ago, and the stability well below 0.01% is shown in the next picture.

Due to the REF02, the T.C. is only about 15ppm/K, and annual stability maybe < 70ppm/year.

Using a LM399H instead of the REF02, the circuitry can be made "ultra precise" with ~3ppm/K and ~20ppm/yr.
Some additional  design measures would give a better stability in the 1µA and 100mA ranges also.

I also found the schematic (hand made) and the PCB layout (AutoCad).

Sorry, as I said, old stuff, so ECAD was not so common at that time.

But I think, all components are still available, so this box can be copied easily.

Frank

[EEVblog]

Very nice, thanks for sharing!
Essentially the same as my precision 1A current source (next video, already uploaded). Haven't tried it at lower currents though with the same MOSFET.

[dannyf]

What about a mcu-based solution? PWM output drives the output + adc read-back + a pid-type control loop.

It can get reasonably good precision without calibration, or very good precision on regular resistors after calibration.

Maybe interesting to try.

[Vgkid]

Thanks for the upload Dr. Frank. I will need to try to build this, along with a lm399 based voltage source.
Who made those resistors?

[Dr. Frank]

Thanks for the upload Dr. Frank. I will need to try to build this, along with a lm399 based voltage source.
Who made those resistors?

The range resistors are called 'econistors', manufacturer not known, but they are labelled 'G.R.'.
Can be purchased at rhopoint components, UK. Around 3-5 pounds / EA.
There should be alternative suppliers in US.

The other ones were stripped from a historic KV divider, manufacturer KINTEL.

If you design a separate LM399 reference board, this chain could be added to this PCB, using 0.1% resistors plus trimming for good linearity.

Frank

[EEVblog]

Who made those resistors?

Vishay do 0.005% resistors for about $20 a pop at Digikey. Good enough for almost anything you'd want to do. But for the ultra-stable stuff they also offer better than that as well with practically zero tempco.

[Dr. Frank]

Well, another remark about the stability of the 100mA range.

This one already drifts 50ppm after being engaged, due to the self heating.

100mA across 10 Ohm are 100mW, and that rises the resistors temperature by 10-20°C.
With a T.C. = 3ppm/K, that results in those 50ppm change.
After heating up, the current stabilizes again. This state is displayed on the 3458A.

In a new design, I would use 100mV of reference voltage across an 1 Ohm reference resistor, and a different adjustment scheme.

Then I would also add a 0.1Ohm, T.C. 5 shunt for an 1A range, giving similar performance as the 100mA range currently.

Without having seen Dave's new video about an 1A current source, I assume, that this will drift strongly, on the order of 0.1%, due to its (assumed) 1.25W dissipation.

Frank

[babysitter]

@Frank+ Vgkid:

After Dr. Frank pointed me on the Econistors I deployed in my LTZ Reference guided by him, I found a lead to Prime Technology which has the Brand General Resistance and a Econistor series. I have quite a confidence in those.

[dannyf]

I went out and put together a quick program to see if a mcu-driven Iref would be possible.

The mcu generates a pwm pulse train to drive the mosfet. Differential adc samples the voltage across R2. A (rudimentary at this point) control algorithm compares the desired output (200 in this case) with the adc result and makes corresponding adjustment to the pwm duty cycle.

In this case, I am shooting for 200 (adc value, which translates into 200 / 1024 * 1100mv / 1k = 0.2ma). It takes a while for the meter to settle to that.

R3 simulates copper resistance - the Iref generator is immune to it.

To make it truly operational, you will need to put in a good algorithm (or it wouldn't be stable), a user interface and the ability to calibrate, plus potentially multi-range.

[sync]

Frank, thanks for sharing this.

Are leakage currents (external cables) a problem?
Would a driven guard a good addition?

[Ross_ValuSoft]

Thanks for sharing your design Frank.

I would have described that configuration as a current sink, rather than current source... but what is in a name? Well done.

Cheers,

Ross

[EEVblog]

Without having seen Dave's new video about an 1A current source, I assume, that this will drift strongly, on the order of 0.1%, due to its (assumed) 1.25W dissipation.

Nope. I'm using these:
http://www.vishaypg.com/docs/63116/vpr221z.pdf
0.05ppm/C from 0-60C

[Dr. Frank]

Frank, thanks for sharing this.

Are leakage currents (external cables) a problem?
Would a driven guard a good addition?

Leakage currents have to be taken into account for the low current ranges.
10pA leakage currents will give 10ppm error for 1µA output.
Therefore I estimate uncertainty for 1µA / 10µA as 0.01% (100pm) only, although a lower level is possible.

I have used the 7650 for guaranteed Ib < 10pA @ 25°C, and a typical 10pA for the BS170.
(It's very difficult to find FETs or MOSFETs with qualified/specified low gate leakage currents.)

The PCB has been cleaned thoroughly and isolated with plastic spray.

The relevant leakage currents all reside on the PCB, not on external cables. Especially the reference voltage input to the 7650 is not relevant in this aspect.
In a new design, I would implement guards around the critical signal lines.
(As a reference, see schematic of the HP 3458A Ohm constant current source).

Any residual leakage currents will be calibrated into the adjustment of the ranges, but may lead to non linearity figures for the vernier and/or medium term drifts of the output.

As a conclusion, leakage currents are relevant for a "ultra precision" version, approaching ppm uncertainty.

Frank

[Dr. Frank]

Thanks for sharing your design Frank.

I would have described that configuration as a current sink, rather than current source... but what is in a name? Well done.

Cheers,

Ross

It's really a constant current source, referenced to the positive supply voltage V+.

In principle, this V+ may be floating to higher levels, e.g. 60V, as BS170 and BD137 may withstand that voltage.

The design may be reversed, using another voltage reference (LM399), a pnp and a p channel FET, so that the current source would be referenced to GND instead.

Another possibility would be to insert a current mirror behind the REF02, to create a virtual volt reference down from V+ and then relate the current source to V+ also.
Difficult to explain, simply see HP3458A Ohm range schematic, or any of the newer 6.5 digits agilents, where this principle is used.

Anyhow, as this version of current source and the DMM used both are floating, there is no constraint or difference in the practical use.

Frank

[Dr. Frank]

Without having seen Dave's new video about an 1A current source, I assume, that this will drift strongly, on the order of 0.1%, due to its (assumed) 1.25W dissipation.

Nope. I'm using these:
http://www.vishaypg.com/docs/63116/vpr221z.pdf
0.05ppm/C from 0-60C

Dave,

I'm very curious about the real performance of your source.

I made very bad experience of the Vishay Z foil resistors regarding their promoted typical T.C., like 0.05ppm/K @ room temperature. On my five 10k VHP202Zs , I measured between -0.4 and -1.0ppm/K in reality.

The maximum values for the VPR221Z of (+/-0.2 +/-2.8 ) ppm/K  = +/- 3ppm/K - may be much more realistic.

But perhaps at 1 Ohm, those different alloys and different foil thickness will give better results.

I'm looking forward impatiently to your video, hopefully demonstrating good thermal stability of the Vishay shunt.

Frank

[EEVblog]

But perhaps at 1 Ohm, those different alloys and different foil thickness will give better results.

Vishay are the world leaders in precision resistors, so I would tend to believe their specs.

I'm looking forward impatiently to your video, hopefully demonstrating good thermal stability of the Vishay shunt.

Sorry, it's nowhere near that detailed, just basically getting the thing stable and within spec at room temp.

[Dr. Frank]

But perhaps at 1 Ohm, those different alloys and different foil thickness will give better results.

Vishay are the world leaders in precision resistors, so I would tend to believe their specs.

Well, in former times I also thought that way about Vishay.

But after my measurements and detailed discussion with the German application engineer,
I do not trust their typical values any more, but their maximum values only.

I also think, they are not able to control this technology in series production, so that you can really encounter that promoted parabolic temperature behaviour with a zero in the T.C. around 25°C.

Therefore I think, those fabulous metal foil resistors are no better than the much cheaper wirewound precision resistor counterparts.

Frank

[EEVblog]

But after my measurements and detailed discussion with the German application engineer,
I do not trust their typical values any more, but their maximum values only.

It can be tricky business to measure this stuff properly, and perhaps it's possible you could have gotten a bad batch?
Unless you have measured multiple series from multiple production runs over multiple years, such a broad claim is hard to justify I think.
But I can certainly understand being once bitten, twice shy.

[Dr. Frank]

But after my measurements and detailed discussion with the German application engineer,
I do not trust their typical values any more, but their maximum values only.

It can be tricky business to measure this stuff properly, and perhaps it's possible you could have gotten a bad batch?
Unless you have measured multiple series from multiple production runs over multiple years, such a broad claim is hard to justify I think.
But I can certainly understand being once bitten, twice shy.

Dave,
You are right, such low T.C. measurements were very delicate.
But my experimental setup was very stable and skilled in physical aspects.
I was able to measure the "real" physically defined  T.C. = dR/dT, compared to the averaged butterfly definition of Vishay.
All five resistors had a perfect linear T.C. curve around 25°C; no sign of parabolic behaviour.

Obviously, the batch I received did not show that ideal behaviour.

I claimed at Vishay, and their answer was exactly, that those typical parameters are not what they guarantee, only the maximum values.
Especially the parabolic shape of T.C., which is required for those fabulous 0.05ppm/K, was not confirmed/guaranteed  by Vishay, in an official letter to me.

But this parabolic shape shows up only, if the thermal expansions of the ceramic carrier and of the conducting metal foil can be controlled, so that they inversely match in magnitude exactly.
The special Z type magic consists of compensating all 1st order temperature drifts, so that the 2nd order, quadratic terms remain.

There were several scientific papers stressing the characteristical parabolic shape, from Dr. Zandman, see here:
http://www.vishaypg.com/docs/60108/VFR_TN108.pdf


For me as a physicist, this feature of exact matching would the core of their technology.

If they negate that feature, they do not really control that technology, and one has to account for worst case parameters only.


In a four-eye-talk, I also asked the very competent application engineer, how to order Z type parts with a guaranteed parabolic T.C. shape and with those extremely low T.C.s.

His answer was, that it is not possible for Vishay to deliver such parts.

The only way is to order VHP101 parts, which consist of a pair of the older S- and K-type resistors in series connection, having inverse T.C.s and thereby compensating the overall T.C to near zero. 
Their guaranteed mean T.C. of 10ppm over 30°C is also an order of magnitude higher than the 0.05ppm/K for the propagated single Z type.

That way, Vishay PG and their Z foil technology lost their charm for me totally.
But I gained a deep insight in this technology.

At last, I am not willing to pay that fortune (80€/EA for VHP202Z) expecting T.C. << 1ppm/K.
Wirewounds (3ppm/K) are as good as the Z foils, for one-tenth of their price.

Only the hermetically sealed, oil filled parts have an advantage, i.e. concerning annual stability.

Frank

[Andreas]

Especially the parabolic shape of T.C., which is required for those fabulous 0.05ppm/K, was not confirmed/guaranteed  by Vishay, in an official letter to me.

But this parabolic shape shows up only, if the thermal expansions of the ceramic carrier and of the conducting metal foil can be controlled, so that they inversely match in magnitude exactly.

Hello,

It would be still interesting if the TC varies with smaller and larger resistor values.
Perhaps there exists a sweet spot in the resistor range where the compensation is better.
Probably this behaviour was then claimed as "typical".

With best regards

Andreas

[EEVblog]

Obviously, the batch I received did not show that ideal behaviour.
I claimed at Vishay, and their answer was exactly, that those typical parameters are not what they guarantee, only the maximum values.
Especially the parabolic shape of T.C., which is required for those fabulous 0.05ppm/K, was not confirmed/guaranteed  by Vishay, in an official letter to me.

No manufacturer of any device is ever going to formally say anything different. Typical specs are never guaranteed.

Wirewounds (3ppm/K) are as good as the Z foils, for one-tenth of their price.

Not for inductance or capacitance they aren't IIRC, but I stand to be corrected.

[EEVblog]

It would be still interesting if the TC varies with smaller and larger resistor values.
Perhaps there exists a sweet spot in the resistor range where the compensation is better.
Probably this behaviour was then claimed as "typical".

That's kind of what I was getting out.
It's not beyond marketing to pick the best tested units to be the banner spec on the datasheet.
That doesn't mean they are fudging the numbers though, such typical performance is very likely possible within a given scenario.
Likely to be manufacturing batch differences too.

[lau190]

 :)thanks for sharing!

[gdewitte]

Dave,


Did you ever publish schematics of your final current sources (the ones for testing the µCurrent Gold)? I've searched the forum a bit, and have not found anything.

[David Hess]

Well, another remark about the stability of the 100mA range.

This one already drifts 50ppm after being engaged, due to the self heating.

100mA across 10 Ohm are 100mW, and that rises the resistors temperature by 10-20°C.
With a T.C. = 3ppm/K, that results in those 50ppm change.
After heating up, the current stabilizes again. This state is displayed on the 3458A.

My solution to this problem was to switch the currents such that the unused resistors had their currents shunted to ground like in an R-2R current DAC so their power dissipation never changes.

Are leakage currents (external cables) a problem?
Would a driven guard a good addition?

Leakage currents have to be taken into account for the low current ranges.
10pA leakage currents will give 10ppm error for 1µA output.
Therefore I estimate uncertainty for 1µA / 10µA as 0.01% (100pm) only, although a lower level is possible.

I have used the 7650 for guaranteed Ib < 10pA @ 25°C, and a typical 10pA for the BS170.
(It's very difficult to find FETs or MOSFETs with qualified/specified low gate leakage currents.)

In the past I have had good results grading the parts myself to get below 1 picoamp although not with chopper stabilized amplifiers.