Measuring nanoamps and below like a Ninja

[VintageNut]

This thread is a place to post your techniques, results, tips and tricks for forcing and measuring low current.

In my opinion the definition of low current starts somewhere between 1nA and 10nA. That is about the noise floor of unshielded wires in open air. You can test this waving your hand near some unshielded wire that is connected to an instrument that has a noise floor well below 1nA.

 

[VintageNut]

I will begin with an example of how to characterize the noise floor of a measurement instrument.

The attached plot has current on the Y axis and "time" on the X axis for a KE6430.

Since this plot is pictorial only with no spreadsheet, I will use pk-pk amplitude to define and quantify the noise of the current measurement.

The technique is to place a metal triax cap over the trax jack of the instrument, force 0V and measure current.

In this case, the pk-pk noise is 3.78 x 10^-16.

There is an offset which is defined as the midpoint of the Y axis on the graph. Removing this offset leaves the user with just the p-p noise as the uncertainty of the measurement.

[tggzzz]

Ah. An easy question to answer.

To measure current like a ninja, you just do nothing whatsoever.

OTOH, to measure current like a cowboy, you stick the two electrodes on your tongue.

[HighVoltage]

I am using my trusted Keithley 410 PicoAmMeter

[3roomlab]

The technique is to place a metal triax cap over the trax jack of the instrument, force 0V and measure current.

i am curious about the measurement cables. i got some PTFE 2 core shielded, i applied 1000v 2 any 2 of the conductors, and measured the leakage (0.04uA/m or 26Gohm/m). so im not sure, how does this compare to a good PTFE insulated cable suitable for "low low loss" ?

[Alex Nikitin]

Here is my Keithley 617 current baseline measurements for 5min, measured every 2sec (the red line is averaging for 10 readings), the bias current (about 1.5fA) is suppressed. I will try to do some wire resistance measurements in a near future, the K617 can output +/- 100V and so would be able to measure resistance up to about 10^17 ohm.

Cheers

Alex

[Gyro]

I'll throw my cheap little Picoammeter project into the mix. You could easilly de-sensititize it if you only want to go as low as nA's 

https://www.eevblog.com/forum/projects/picoammeter-design/msg790045/#msg790045

P.S: Just reading through the thread again. In terms of 'tips', there are plenty among the replies, including a number from Alex.

[e61_phil]

Sorry it is in german, but here are some quick&dirty measurements of mine with a breadboard setup

https://lowcurrent.wordpress.com/2015/09/07/messung-von-leckstroemen/

The resulting standard deviation was about 5fA and peak to peak value was about 23fA

Proper shielding will decrease the noise a lot, of course.

[VintageNut]

The technique is to place a metal triax cap over the trax jack of the instrument, force 0V and measure current.

i am curious about the measurement cables. i got some PTFE 2 core shielded, i applied 1000v 2 any 2 of the conductors, and measured the leakage (0.04uA/m or 26Gohm/m). so im not sure, how does this compare to a good PTFE insulated cable suitable for "low low loss" ?

If you are using the cables to measure mA, then a 40nA leakage is not important. If you are measuring pA, then 40 nA of leakage is much larger than your intended measurement. I think that "low" is a relative term.

One question I have is why measure leakage at 1000V? Will you be using this cable to carry 1000V?

[VintageNut]

You can test this waving your hand near some unshielded wire that is connected to an instrument that has a noise floor well below 1nA.

I once tried to measure insulation resistance of a piece of ceramic substrate. I set the bias voltage to 1kV, and my meter reads as low as a couple tens of nA. Forget about moving hands or any conductive metal parts around, even only blowing air can cause displacement current of air dielectric, which considerably alters readings by up to hundreds of nA, so one word or suggestion: shield your DUT, not only electrically, but also mechanically. Don't let air to pass by it, or any thing that can carry charges.

Exactly. Air can carry charge from HI to LO. That is why having a hermetically sealed enclosure around the DUT and energizing that enclosure to GUARD potential is an important technique. All of the air inside of the sealed enclosure is charged to GUARD potential creating a fluid that insulates against leakage from HI to LOW.

I set one of these up and measured some wire. During one of the tests I encountered a very large current which was traced to a thumbprint that spanned the wire insulation and the copper core. Cleaning the sample restored the low current. I then purposely added a thumbprint to verify the cause and the measurement was ruined as I expected.

[VintageNut]

I'll throw my cheap little Picoammeter project into the mix. You could easilly de-sensititize it if you only want to go as low as nA's 

https://www.eevblog.com/forum/projects/picoammeter-design/msg790045/#msg790045

P.S: Just reading through the thread again. In terms of 'tips', there are plenty among the replies, including a number from Alex.

DIY pA meter is very cool. Someone had to build the first one!

[Gyro]

DIY pA meter is very cool. Someone had to build the first one!

Actually credit for the first (hobby) one probably [edit: possibly] goes to Alan Yates for this one:  http://www.vk2zay.net/article/251   It inspired mine anyway, although his didn't have Vos trimming. Come to think, i've seen several based on the CA3140 too.

[plesa]

For every beginner Ninja in this thread:
http://www.tek.com/sites/tek.com/files/media/document/resources/LowLevelHandbook_7Ed.pdf
Low current measurement or electron counting is like ppm chasing in voltage standard section, sometimes more difficult I must say

VintageNut
What is your application for K6430?

[David Hess]

I sometimes hack together a transimpedance amplifier using an LMC6081 selected for low input current to make measurements below 1 picoamp but often there is an easier way.

Common good digital voltmeters have a 10 megohm input shunt resistance and negligible input current so if DC volts mode is used to make a current measurement, the sensitivity is 100nA/V or 100pA/mV.  My 3.5 and 4.5 digit bench meters have a resolution of 100uV on their most sensitive range yielding 10pA resolution.  I have some higher count meters which do better yielding 1pA resolution.

So how good are the results in real life?  Well, my ancient 4.5 digit Tektronix DM501 with 100uV resolution uses a discrete differential JFET input buffer (no autozero means no charge pumping) with a specified input current below 50 picoamps which in practice it is less than 10 picoamps (I measured it) which is only 1 count.  The noise when making this measurement with a direct connection to the input is 1 count or less which is as good as it gets.  The integration time nulls out ambient 60 Hz noise and its harmonics.

My 3.5 digit Tektronix DM502 seems to do just as well even though its autozero MOSFET front end (Siliconix LD111) probably adds some noise from charge pumping.

[VintageNut]

For every beginner Ninja in this thread:
http://www.tek.com/sites/tek.com/files/media/document/resources/LowLevelHandbook_7Ed.pdf
Low current measurement or electron counting is like ppm chasing in voltage standard section, sometimes more difficult I must say

VintageNut
What is your application for K6430?

I have participated is several; photon counting, direct current measurement of circuit boards in sleep mode, etc. Measuring a 1T resistor with less than 1V was the first project.

[VintageNut]

This is a K6430 measuring a 1T resistor using 0.5VDC. The measurement yields 6 significant digits.

The metal pans are at GUARD potential.

[HighVoltage]

What shape is this 1 TOhm resistor?
That is kind of impressive, measuring such a resistor at less than 1V
 

[Alex Nikitin]

This is a K6430 measuring a 1T resistor using 0.5VDC. The measurement yields 6 significant digits.

The metal pans are at GUARD potential.

  What do you mean by "6 significant digits" ?! The last digit in this case equals about 3 electrons a second . The Keithley is a nice unit, but it can manage only about 0.4fA p-p noise, or ~2500 electrons/sec . That means at least the last two (and more likely three) digits are noise.

Cheers

Alex

[VintageNut]

What shape is this 1 TOhm resistor?
That is kind of impressive, measuring such a resistor at less than 1V

The goal was sub-pA stable current. The resistor is pictured below. It is a Russian glass enclosed device bearing Cyrillic lettering. I bought a small assortment from eBay. The seller was from Bulgaria. You can see the black GUARD connection to the enclosure. The measurement requires a very long time to stabilize. All of the capacitance in the cables etc are being charged with less than 1 pA. All of the air in the enclosure has to charge to 0.5V as well. This is a long slow process.

[VintageNut]

This is a K6430 measuring a 1T resistor using 0.5VDC. The measurement yields 6 significant digits.

The metal pans are at GUARD potential.

  What do you mean "6 significant digits" ?! The last digit in this case equals about 3 electrons a second . The Keithley is a nice unit, but it can manage only about 0.4fA p-p noise, or ~2500 electrons/sec .

Cheers

Alex

Understood and agreed as presented. The most significant three digits are "above" the noise.

If you look at the noise plot, it is not random. If someone were so inclined, that noise looks suspiciously periodic and may likely be characterized and largely removed.

[plesa]

You are using same resistors like me I purchased few of them on eBay about 5 years ago for experiments.
I like your barbecue style shielding, when I saw it in some video/appnote made by Keithley or Agilent I was laughing, but after test I must confirm it works

[VintageNut]

The aluminum pans are qty 3 for USD $1.00. Almost for free. The edges can be folded over to make the air current almost zero.

[plesa]

The aluminum pans are qty 3 for USD $1.00. Almost for free. The edges can be folded over to make the air current almost zero.

Yep it is pretty smart, but I cannot remember when I saw it for first time
If it was invented by Keithley or by Agilent or someone else...
For normal measurement I prefer light-tight enclosure.
It can be interesting to see what is inside preamp box, can expect teardown of it? 

BTW why you are measuring resistor with such a low voltage, these resistors should withstand 200V.

[VintageNut]

The aluminum pans are qty 3 for USD $1.00. Almost for free. The edges can be folded over to make the air current almost zero.

Yep it is pretty smart, but I cannot remember when I saw it for first time
If it was invented by Keithley or by Agilent or someone else...
For normal measurement I prefer light-tight enclosure.
It can be interesting to see what is inside preamp box, can expect teardown of it? 

BTW why you are measuring resistor with such a low voltage, these resistors should withstand 200V.

The 6430 did not belong to me. It was on loan to me.

I used 0.5V because the project was to explore measuring below 1pA. The resistor was a convenient device to help accomplish the goal at a safe voltage.

Those aluminum pans will not allow any light inside if you fold over the edges carefully.

My preferred enclosure now is that KE 8101-PIV. It closes securely and will no allow air or light inside. I use the 7078-TRX-BNC adapter that makes the enclosure at GUARD potential for the HI SMU cable. The LO smu cable is converted to isolated BNC/banana. It works well for small devices. It is not safe for high voltage because the enclosure is at GUARD. I usually do not mix high voltage and low current this way. For voltages above 40V, I use an interlocked enclosure for safety.

[HighVoltage]

I also have some of these military glass resistors with high GOhm values but they never showed any stable readings on voltages below 10V. So, I have usually tested them with 100V on the SMU.

Your way of shielding is pretty clever.
Thanks for the explanation, I will give this a try.
 

[Zucca]

FYI, in case you need an electrometer... Keythley 617....

https://www.eevblog.com/forum/buysellwanted/(fs)-(eugermany)-keithley-617-programmable-electrometer-500-ship/

[Martin.M]

or with a Nixie-Nano 

greetings
Martin

[CalMachine]

These are all pretty awesome!! 

The lowest I've been able to measure with a fair amount of stability was in the pA range utilizing our Fluke 732B and a few of our TOhm standards.

[VintageNut]

I recently acquired a KE487 and a KE8002A. There is a 10G in the enclosure. Voltage source is a Fluke 731B. The resistor appears to be about 0.25% low. The 487 spec is 0.3% of reading so a better picoammeter would be required to make this measurement. The 6517B is only slightly better at 0.2% for the 2nA range. The 6430 is 0.05% on the 1nA range. The 6430 costs more than a 3458A.

This is probably why high resistance bridges remain in production. 

[TiN]

Well, it took bit more hours than I expected to get 4142B going with Ta wet-slug XTV cap as DUT.
Had to add 2KOhm in series, as SMU output is not rated >10uF load.

Connection is simple:

4142B(41421B SMU module) guarded force triax -> CNMC low-noise triax -> die-cast box -> 2K resistor -> XTV 1300uF 30V cap -> twinax cable -> 4142B GNDU sink.



As usual, linux-gpib is used to control ATE box, running on Terasic DE1-SoC FPGA board (linaro Linux with 3.18 kernel) + Adafruit BME280.
Will see if new setup able to reach 3.2nA value, achieved using K2400 and regular coax cable.

[Alex Nikitin]

I recently acquired a KE487 and a KE8002A. There is a 10G in the enclosure. Voltage source is a Fluke 731B. The resistor appears to be about 0.25% low. The 487 spec is 0.3% of reading so a better picoammeter would be required to make this measurement. The 6517B is only slightly better at 0.2% for the 2nA range. The 6430 is 0.05% on the 1nA range. The 6430 costs more than a 3458A.

This is probably why high resistance bridges remain in production.

Looking at these numbers I now more and more appreciate the Keithley 263 calibrator I've bought recently. My 617 electrometer is specified at 0.25%+500fA at 2nA range and 1.6% +10fA on 200pA, 20pA and 2pA ranges (1 year accuracy). The 263 however is specified (as a current source) at 0.065% +100fA at 2nA range, 0.25% +30fA at 200pA range,   0.375% +10fA at 20pA range and 0.425% +10fA at 2pA range. And I've paid less for it  .

Cheers

Alex

[e61_phil]

Looking at these numbers I now more and more appreciate the Keithley 263 calibrator I've bought recently. My 617 electrometer is specified at 0.25%+500fA at 2nA range and 1.6% +10fA on 200pA, 20pA and 2pA ranges (1 year accuracy). The 263 however is specified (as a current source) at 0.065% +100fA at 2nA range, 0.25% +30fA at 200pA range,   0.375% +10fA at 20pA range and 0.425% +10fA at 2pA range. And I've paid less for it  .

These are really impressive specs. How do you want to verify them?

[Alex Nikitin]

Looking at these numbers I now more and more appreciate the Keithley 263 calibrator I've bought recently. My 617 electrometer is specified at 0.25%+500fA at 2nA range and 1.6% +10fA on 200pA, 20pA and 2pA ranges (1 year accuracy). The 263 however is specified (as a current source) at 0.065% +100fA at 2nA range, 0.25% +30fA at 200pA range,   0.375% +10fA at 20pA range and 0.425% +10fA at 2pA range. And I've paid less for it  .

These are really impressive specs. How do you want to verify them?

The interesting feature of the K263 is its self-calibrating capabilities. To calibrate the high ohm and low current ranges it uses the "Ladder" technique and the internal electrometer (quite high-performing, based on the AD549L chip). Below is the list of the calibration equipment required  . I can use the HP3458A as a multimeter, though even K2015 should be fine if recently (<1year) calibrated.

Cheers

Alex

P.S. I can additionally verify the 10M and 100M resistors directly with the HP3458A (50ppm and 0.05% 1 year accuracy respectively).

[VintageNut]

Yes the 263 is very good for current.

My quandary is that measuring a 100G or 1T resistor is not going to be much better than 1% using any current source or current measurement instrument.

1% is 10,000 ppm uncertainty. Thats like a crappy handheld DMM performance. 

[Alex Nikitin]

Yes the 263 is very good for current.

My quandary is that measuring a 100G or 1T resistor is not going to be much better than 1% using any current source or current measurement instrument.

1% is 10,000 ppm uncertainty. Thats like a crappy handheld DMM performance.

I think that with the K263 about 0.1% at 100G should be possible, using an accurate 100V source (say, K617) and measuring 1nA by a compensation method.

Cheers

Alex

[TiN]

If you think 1% is crappy, check high-dollar Guildline 9337 10POhm resistance standard, which has amazing 25% spec . I wonder what insulator is used there.

[VintageNut]

If you think 1% is crappy, check high-dollar Guildline 9337 10POhm resistance standard, which has amazing 25% spec . I wonder what insulator is used there.

Good point. Looks like higher resistances have looser specs. I did not know that. That explains the 1% pA uncertainties for most Keithley sensitive current instruments.

I have a KE515 bridge but have not attempted an overhaul. The output connector appears to have been hacked to install a BNC. I think that it is supposed to have a triax connector.

[mrtn]

One of my first jobs was at a company that makes radon detectors. One (not so great) method to detect radon is to monitor the amount of ionized particles that are generated in a volume of radon laden air. The radon decays and produces charged particles.

How do you measure that? Well the cheapest way is to put a voltage across two parallel plates and measure the current between the two. I developed an instrument that put about 35v across two plates (area ~100cm^2) with long teflon insulators (about 2 cm between plates). The current sensor was an opamp transimpedance amplifier with a feedback resistor of around 1 gigaohm. The PCB had guard traces that were kept at the same potential as the input net. The whole thing was in a shielded box with an air path that passed through wire mesh.

It could measure into the tens of femto-amp range. Unfortunately I don't have statistical information as I'm just working from memory.
It was a fun project.

[Jay_Diddy_B]

Hi,

The most sensitive current measurement that I can make in my lab is with an HP4155B Semiconductor Parameter Analyzer.

Here is a photograph showing Id (Yellow) and Gm (Blue) for a small signal MOSFET (2n7002) versus Vgs.



The vertical scale for the current is log from 1fA (not f**k all, femto Amps) to 100mA, The lowest drain current measured was 1pA. The marker is at Vgs=400mV there is Id=16.9pA

The HP4155B has four of these SMUs.

Regards,

Jay_Diddy_B

[TiN]

Retest with 11KOhm BMF resistor in series with DUT.

1st test was for noise floor for 41421B SMU (cap not connected, resistor just hanging in the air in DUT box).
Source 10V, measure current. It happen to be ~160 fA (femtoamps) average, with noise ~400 fApk-pk. 
There are no Teflon plates or insulators in the DUT box, and also GNDU return for ground is routed using 1m long twinax cable, not a low-leakage triax. Don't have extra triax cable with two TRX ends .

Now connect 1300uF 30V wet tantalum XTV cap #1 (same one measured by K2400 at ~3.2nA Ileak before) as DUT and test again.
Leakage around 1.21 nA +/-0.3nA at 10.00VDC sourced, over >6 hours, even with 3C ambient temp change. Cap had to soak at 10V for ~20hours to stabilize at this level, just like Jim Williams observed and hinted in Linear AN124.

Next test - cap #2, same conditions, starts today

[guenthert]

I have a KE515 bridge but have not attempted an overhaul. The output connector appears to have been hacked to install a BNC. I think that it is supposed to have a triax connector.

Without derailing the thread (too much), I have a question regarding those triax connectors:  those connectors and cables are astonishingly expensive (probably due to the niche market and target audience).  Can't one just replace a triax connection with two BNC connections (one being signal shielded by guard, the other return (GND) shielded by, well, shield)?  I realize guard might be on high potential, but the preamps driving guard are limited to quite low currents, so this should be fairly safe (for humans, not necessarily for semiconductors) as long as one is not charging up CRT-sized capacitors and after being zapped for the third time, one will be more careful anyhow.  (where's the SHOCKED emoticon when one needs one?)

[VintageNut]

I have a KE515 bridge but have not attempted an overhaul. The output connector appears to have been hacked to install a BNC. I think that it is supposed to have a triax connector.

Without derailing the thread (too much), I have a question regarding those triax connectors:  those connectors and cables are astonishingly expensive (probably due to the niche market and target audience).  Can't one just replace a triax connection with two BNC connections (one being signal shielded by guard, the other return (GND) shielded by, well, shield)?  I realize guard might be on high potential, but the preamps driving guard are limited to quite low currents, so this should be fairly safe (for humans, not necessarily for semiconductors) as long as one is not charging up CRT-sized capacitors and after being zapped for the third time, one will be more careful anyhow.  (where's the SHOCKED emoticon when one needs one?)

It is known practice to use ordinary coax where the coax shield carries the guard signal. Leakage is theoretically zero in this configuration. I use this configuration for very high resistance measurements inside of a shielded ground-potential box. The guarded box resides inside of a safety ground potential box. This requires an interlock that turns off the driving instrument when the grounded box is opened.

Ordinary coax has a problem with triboelectric induction during ordinary usage. The high dollar triax cable is constructed with lubricant that reduces or eliminates the effects of triboelectric induction.

It is fairly simple to set up your measurement, turn on the measurement while leaving off the source, flex the cable and see if you can measure an error induced by flexing the cable.   

[Alex Nikitin]

Ordinary coax has a problem with triboelectric induction during ordinary usage. The high dollar triax cable is constructed with lubricant that reduces or eliminates the effects of triboelectric induction.

It is not connected to the coaxial/triaxial difference. You can get a "low noise" coaxial cable with a graphite lubricant and you can get a "standard" triaxial cable without any lubricants. Unfortunately, two triaxial cables I can get easily are of the "standard" type. So far I've used the Belden 9222 which works very well as long as you are not moving it during the measurement. I am trying to get some "low noise" triaxial cable but so far without much success!

Cheers

Alex

[VintageNut]

Ordinary coax has a problem with triboelectric induction during ordinary usage. The high dollar triax cable is constructed with lubricant that reduces or eliminates the effects of triboelectric induction.

It is not connected to the coaxial/triaxial difference. You can get a "low noise" coaxial cable with a graphite lubricant and you can get a "standard" triaxial cable without any lubricants. Unfortunately, two triaxial cables I can get easily are of the "standard" type. So far I've used the Belden 9222 which works very well as long as you are not moving it during the measurement. I am trying to get some "low noise" triaxial cable but so far without much success!

Cheers

Alex

For a home hobbyist, the lubricated coaxial cable is probably good enough. The shock hazard remains. I doubt that any of us here employ a truly safe interlocked and shielded enclosure.

The OEM triax cable is low noise and safe when used with a properly interlocked enclosure. 

You can purchase lengths of unterminated triax cable from Tek.   

[plesa]

Ordinary coax has a problem with triboelectric induction during ordinary usage. The high dollar triax cable is constructed with lubricant that reduces or eliminates the effects of triboelectric induction.

It is not connected to the coaxial/triaxial difference. You can get a "low noise" coaxial cable with a graphite lubricant and you can get a "standard" triaxial cable without any lubricants. Unfortunately, two triaxial cables I can get easily are of the "standard" type. So far I've used the Belden 9222 which works very well as long as you are not moving it during the measurement. I am trying to get some "low noise" triaxial cable but so far without much success!

Cheers

Alex

I replaced all Belden 9222 by Keithley cables. You can find them cheap on eBay (new Keithley Low Noise Triax Cable 7078-TRX-5 for $40). This is cheaper than making your own because connector price.
If you needs cable only this has good price http://www.ebay.com/itm/10m-33-ft-Low-noise-Triax-Cable-50-Ohms-RF-Medical-aerospace-camera-NEW-/201706863349?
TiN is using this type ( CNMC is made by quirkwire.com), we discussed it in this thread https://www.eevblog.com/forum/repair/electrometer-output-stage-keithley-6517/
Cable looks like silver plated which can be handy, Tek/Keithley cable is difficult to solder.
The main disadvantage of Belden 9222 is that it is intend to use in video applications and not in low currents measurements.

[TiN]

Where did you see 7078-TRX-5 for 40$? I see only 80$ one, and it's very short..

[plesa]

Where did you see 7078-TRX-5 for 40$? I see only 80$ one, and it's very short..

It was offer which has been accepted, the buy now prices are ridicilous:)
BTW is the CNMC cable silver or tin plated?

[Alex Nikitin]

While I am waiting for the low noise triaxial cable to arrive from the US (as well as a couple of triaxial sockets), I've built a nice 100G+1G resistor box with plain coaxial sockets. I've measured 1G resistor on the HP3458A as 1.005G and the K617 confirms that value. For 100G value I have to rely upon the K263 and k617. This box so far works very well and it did allow me to estimate the temperature coefficient of 100G resistor at about 400ppm/C (there is no figure in the manufacturer's datasheet, only "can be as good as 50ppm/C"  . I also attempted to measure the input impedance of the K617 by measuring the voltage drop on 100G resistor in series with 19V source. Despite me using a non-guarded connection so the insulation resistance of 1m Belden cable was connected in parallel to the K617 input, only ~5mV drop was observed (after ~10 minutes), making the input impedance in-spec at ~400Tohm (0.1T x 4000).

Cheers and Happy New Year to all!

Alex

[plesa]

Nice!
I will expect the burden voltage even lower.
Attached graph is input noise of 6517.IT is close to what Keithley specify in datasheet ( 0.75 fA p-p and input bias 3fA).

[Alex Nikitin]

Nice!
I will expect the burden voltage even lower.

It should be lower with a guarded connection. I'll repeat this measurement soon when I'll put a proper triax socket on the box.

Cheers

Alex

[guenthert]

Youtube thought I might be interested in the "squre root electrometer" https://www.youtube.com/watch?v=xbKiAl2ioBg.  Looks nice to me, but my Keithley 617 just arrived (seemed to be too good a deal to pass on, but I just noticed the busted GPIB connector 

[Alex Nikitin]

Here is a direct input current measurement of the HP3458A, 10V DC range, AZERO OFF, NPLC100 (AZERO ON adds a noticeable current noise on the input in this setup - up to +/-5pA jumps on the Keithley 617, making accurate measurements quite difficult). Without auto-zero the value stabilises to less than +/- 0.2pA variations and I've recorded the mean value at 1V intervals for the input voltage from -11 to +11V. A 1Gohm resistor was placed in series with the input of the meter as otherwise the current fluctuations from the meter's input did confuse the electrometer. A voltage drop on the resistor was used to double check the measured current value with a good agreement (i.e for 10.000V input the measured voltage on the meter was about 9.986V) . This measurement was performed with TEMP?=35.6C and the ambient temperature 23C. Later I've tried a 100G resistor in series with the input and for the measured current 14.5pA the measured voltage was 8.55V. The temperature of the meter (and the ambient) was about 0.6C higher at that time. With 100G in series AZERO ON did not noticeably change the readings.

Cheers

Alex

[TiN]

Interesting. Can you also measure current on 3458A's GOHM range? I wonder how stable that is.
Also does guard switch to LO or to open (with connection to electrometers earth?) changes DCV current?

[Alex Nikitin]

Interesting. Can you also measure current on 3458A's GOHM range? I wonder how stable that is.
Also does guard switch to LO or to open (with connection to electrometers earth?) changes DCV current?

1) The current through the DUT on 1G range is not well defined  . On 1Gohm and 100Mohm ranges the current source used is the same as for 10M range - about 0.5uA, but with an additional 10Mohm resistor connected in parallel to the DUT. As the result, the current near the top of the range, i.e for 1Gohm resistor, is very small, and the short circuit current is the same on all three ranges.

2)  The measurements were performed with the guard to the LO. If I connect the guard to the electrometer earth the current for 10V input increases by about 0.7pA (from -14.7pA to -15.4pA) .

Cheers

Alex

[Alex Nikitin]

Here is a 16 hours run of 1pA current sourced by Keithley 263 and measured by Keithley 617. Temperature shown on the graph is measured by a thermocouple attached to the case of the K263 near the electrometer amp position. The visible variation with temperature is mostly due to the change of the electrometer input current on the K263. If I suppress the "zero" current again, as I've done at the beginning of this measurement, this variation is reset almost completely. This "zero" current at the moment, with the ADA4530-1 installed, is about +5...+10fA, depending on the temperature. This graph shows a very good performance from both units, essentially variations of about 5fA or 0.5% from 1pA value over 16 hours.

Vertical scale is 1fA/div. Measurements are made every 20 seconds but for a single sample with 330ms conversion time on the K617.

Cheers

Alex

[Alex Nikitin]

Another measurement. I plan to do the same measurement on a number of diodes and JFETs connected as diodes. Here we have BF861 JFET connected as a diode. I've used these for the Keithley 2015 repair and happy to see that the leakage is good. The voltage was supplied by the Keithley 263 and the current measured by the Keithley 617. The JFET was placed in a metal enclosure with two BNC connectors. The residual leakage was less than +/-2fA and the enclosure temperature was 24.6C .

Vertical scale is 0.5pA/div, the second graph is a zoomed in part of the first.

Cheers

Alex

[VK5RC]

Can I ask how are you guys cleaning the wires / connectors / components  after handling / soldering etc to avoid surface conduction?
IPA?
What sort of cloth / tissue / q tip??
Thanks Robert

[Alex Nikitin]

Can I ask how are you guys cleaning the wires / connectors / components  after handling / soldering etc to avoid surface conduction?
IPA?
What sort of cloth / tissue / q tip??
Thanks Robert

Hi Robert,

A good description of a cleaning procedure required is available in the data sheet for the ADA4530 evaluation board from AD (page 5). It is also worth noting that it is best to avoid contamination if possible, using minimal amounts of flux and solder, handling all components with clean tweezers only and so on.

Cheers

Alex

[VK5RC]

Thanks Alex,
Thanks for the reference. Data sheets can be a goldmine.
I have done a few trials of some off the shelf high value resistors (>10 GOhm) and a Keithley 619. The techniques to get reliable and accurate readings would appear to be very fastidious in many directions.
I am pretty sure my IPA is not 'clean room grade' hence I have not used it yet - but the no-touch technique is not new to me.
Robert

[Vtile]

These Metrology threads have been eye opening. Thank you all the ones who are participated, especially the certified voltnuts. 

[Alex Nikitin]

Here is the Keithley 617 measuring the Keithley 263 output over about 18 minutes time - first the zero baseline, then +50fA, -50fA and back to baseline.

Cheers

Alex

[VK5RC]

Hi Alex, got to be happy with those results.
I found an old HP "Test fixture" at a price I thought was about the costs if I made one, BUT my Keithley 619 has 2 lug 'triax' connectors and the HP 3 lug.
I took a few brave pills and bought some 2 and 3 lug 'triax' connectors ( I phoned my Bank manager first!) and built an adaptor cable. I had huge doubts that it would be good enough esp with all the touching of the various insulators during the mounting of the connectors on the triax. I washed my hands and tools with IPA before starting and tried to minimise contact but mounting a triax connector I did not find easy.
Below is a quick test measuring a 100G ohm resistor (1%) inside the test fixture, amazing the difference lifting the lid makes. Hopefully my adaptor cable can handle higher resistances accurately.
Robert
PS I have found some Chemtronics clean room IPA pre-soaked foam swabs (Digikey PN = CS25-ND) for not crazy money, which I will try if I think something has got contaminated until then just no touch and keep resistors etc in plastic bags.

[Alex Nikitin]

Hi Robert,

It looks good with that 100G! Here is a comparative measurement of five different makes of the BAV199 "low leakage" diode. As this data was collected at work, I won't specify the manufacturers . As you may see, the spread is quite large, and a lower leakage in reverse direction does not necessarily corresponds to a lower leakage in forward direction. The best samples manage below 10fA at +/-50mV.

Cheers

Alex

P.S. - to put things into perspective, I've added two comparative graphs for the 1N4148 diode and the worst device out of the BAV199 bunch  .

P.P.S. - all measurements done with the diodes temperature between 24C and 27C .

[VK5RC]

It is interesting to see what really happens to components when you can measure at very low/high  levels, assumptions go to the wall.
That data looks like it came from a nice SMU - I would like one but can't justify the extreme $, in what is for me, a pastime.
I haven't really thought much about diodes - every time I do (mainly as an RF 'mixer') my brain hurts! HIHI
Robert

[Alex Nikitin]

That data looks like it came from a nice SMU - I would like one but can't justify the extreme $, in what is for me, a pastime.

Hi Robert,

The data was taken with the Keithley 617 electrometer, used both as the source and the meter, controlled over GPIB. It takes quite a bit of time to get a good accuracy at low currents so each point was taken after a 30s settling time and averaged for 10 consecutive electrometer readings.

Cheers

Alex

[doktor pyta]

@Alex Nikitin
thanks for nice charts.
Could You please type units (V,mV) on the horizontal axes of last plots?
How many measuring points are on these plots?
All were taken for reverse bias ?

[Alex Nikitin]

@Alex Nikitin
thanks for nice charts.
Could You please type units (V,mV) on the horizontal axes of last plots?
How many measuring points are on these plots?
All were taken for reverse bias ?

Thank you. On all graphs the X-axis is in Volts and Y-axis is in Amps. 0.05V to 100 V in reverse bias (24 points) and 0.05V to 0.5V in forward bias (10 points).

Cheers

Alex

[Conrad Hoffman]

Jim Williams advocated the use of diode connected JFETs for low leakage. Every tried those?

[Alex Nikitin]

Jim Williams advocated the use of diode connected JFETs for low leakage. Every tried those?

I've posted measurements for one JFET (BF861) earlier in this thread, taken in a somewhat different setup (K263 as a source and K617 as a meter). JFETs are much better than common signal diodes like 1N4148, however very few would come close to the performance of even the worst BAV199 devices and none will survive 100V reverse voltage or have leakages similar to the best of BAV199. I plan to measure some more JFETs and BJTs soon (the Keithley 617 has some selected BJTs as input protection diodes by the way) and will post results here.

Cheers

Alex

[David Hess]

I've posted measurements for one JFET (BF861) earlier in this thread, taken in a somewhat different setup (K263 as a source and K617 as a meter). JFETs are much better than common signal diodes like 1N4148, however very few would come close to the performance of even the worst BAV199 devices and none will survive 100V reverse voltage or have leakages similar to the best of BAV199. I plan to measure some more JFETs and BJTs soon (the Keithley 617 has some selected BJTs as input protection diodes by the way) and will post results here.

Cheers

Alex

BAV199 diodes are only tested and guarantied to be below 5 nanoamps.  A 4117 JFET is guarantied to be 10 picoamps or lower.

Of course if you need a higher peak inverse voltage than a JFET gate will support, then you have to use the BAV199 or maybe the collector-base junction of a transistor but in both cases in order to have guarantied low leakage, you have to grade them yourself.

What the BAV199, JFET gate junctions, most transistor junctions, and real low leakage diodes have in common is no gold doping which would significantly increase reverse leakage current.  For the BAV199, this is revealed by its relatively high reverse recovery time of up to 3us.

[Assafl]

I don't know if this is relevant but stumbled upon this while looking for ways to stabilize electrometer readings...

The students were trying to measure electron transport in polymers induced under cosmic radiation - using a 1970's vintage electrometer. Their resolution was 400aA - which IMHO is more the crap physicists deal with than electrical engineers - 100's to 1000's of electrons.

The guarded measurement is done in a constant voltage vacuum chamber (duh).

http://digitalcommons.usu.edu/cgi/viewcontent.cgi?article=1075&context=mp_presentations

This was presented as a poster session in some physics gathering...
http://meetings.aps.org/Meeting/4CF09/Session/D1.3
http://digitalcommons.usu.edu/mp_presentations/75/

[David Hess]

Bob Pease wrote an article about the problems they had testing input bias current at the femtoamp level.  The standard method was to configure the operational amplifier as an integrator but the physically large low leakage air capacitor had enough capture area that incoming cosmic rays were a significant problem.  Their solution was write the test software to distinguish cosmic ray events but now I wonder if a vacuum capacitor would have been better or at least feasible.

[Assafl]

Bob Pease wrote an article about the problems they had testing input bias current at the femtoamp level.  The standard method was to configure the operational amplifier as an integrator but the physically large low leakage air capacitor had enough capture area that incoming cosmic rays were a significant problem.  Their solution was write the test software to distinguish cosmic ray events but now I wonder if a vacuum capacitor would have been better or at least feasible.

Pease porridge columns were awesome reads indeed... I think he even had one of his National Semiconductor "Bob Pease" Show demonstrating that measurement setup. If I recall it was a very temperamental setup they had developed for production testing... completely encased in a box made of Copper PCBs - I think it then had another box around it....

Edit: Found it -

Femtoamp stuff starts at 13:00. Minute 38:45 shows the rather simple integrator setup with the 10pF Teflon cable as integrating capacitor with the shielding as guard (small size to reduce cosmic rays and the shape is self guarding).

[BFX]

Something arrived at Friday   
I'm curious how good is it since I don't believe that plastic box. But I should wait for triaxial cable and connector for my Keithley 602.
 

[Assafl]

A calibration box for a hi-pot tester?

http://www.sefelec.com/en/calibration-kit-MG-91

[BFX]

A calibration box for a hi-pot tester?

http://www.sefelec.com/en/calibration-kit-MG-91

We will see how usable it is     

[guenthert]

I like the idea of coiling the leads of the resistors.  Might have prevented my mishap, breaking the glass of a 100GOhm resistor.

[ocw]

There has been slow evolution of my ADA4530-1 low current to voltage converter.  It has selectable 1k, 10k, 100k, 1M, 10M, 100M, 1G, 10G, 100G and 1T feedback resistors.  See its attached picture.

I'm trying to find the minimum current which I can measure with the 1T feedback resistor.  I knew that 1.5 pA was easy by using a 1.55 volt battery which was connected to an external 1T resistor.  I added a 1/10 voltage divider inside of my shielded test box to feed 155 mV to that resistor.  The second attachment shows the 165 fA current produced by that combination (1T 10% resistor is actually close to 0.955).  As is typical with these converters, a positive current is converted to a negative voltage.  1 volt = 1 pA = 1,000 fA.

My test box has a switch to change to a 1/100 voltage divider.  The third attachment shows the approximately 16 fA current produced by that.

Time to build a greater voltage divider.

[VintageNut]

There has been slow evolution of my ADA4530-1 low current to voltage converter.  It has selectable 1k, 10k, 100k, 1M, 10M, 100M, 1G, 10G, 100G and 1T feedback resistors.  See its attached picture.

I'm trying to find the minimum current which I can measure with the 1T feedback resistor.  I knew that 1.5 pA was easy by using a 1.55 volt battery which was connected to an external 1T resistor.  I added a 1/10 voltage divider inside of my shielded test box to feed 155 mV to that resistor.  The second attachment shows the 165 fA current produced by that combination (1T 10% resistor is actually close to 0.955).  As is typical with these converters, a positive current is converted to a negative voltage.  1 volt = 1 pA = 1,000 fA.

My test box has a switch to change to a 1/100 voltage divider.  The third attachment shows the approximately 16 fA current produced by that.

Time to build a greater voltage divider.

Two things that I am surprised about

1. No guard

2. The banana jacks appear to garden variety. These usually have significant leakage compared to pA current

[ocw]

On the ADA4530-1 data sheets Analog Devices warns of possible nonlinearity when using high voltage resistors at low voltages (a poor voltage coefficient).  To try and evaluate the linearity of the 1T ohm resistor which I used as both a feedback resistor, with another used along with a metered voltage to set the current level which is being measured, I fed the external MOX112523100AK resistor with a 5 volt ramp waveform and observed the output of the converter. 

I fed a 3 milliHertz waveform to the converter's input.  While my DC measurements with the battery and resistors located inside of a shielded box produce negligible power line noise, my quick connection to the function generator had some of that noise modulating the 3 milliHetz source.  But, the attached view of the output of the converter still looks like the MOX112523100AK resistor is fairly linear over the 0 - 5 volt range measured.

Regarding:

1. No guard

2. The banana jacks appear to garden variety. These usually have significant leakage compared to pA current

The construction was a learning project.  My choice of construction at least appears adequate for the pictured 16 fA current, besides a significantly higher pA current...

[tggzzz]

On the ADA4530-1 data sheets Analog Devices warns of possible nonlinearity when using high voltage resistors at low voltages (a poor voltage coefficient).

I've picked up a couple of 1Tohm 3% "resistance transfer standards", without having anything to use them for, which implies TEA, FUSS, GAS, but since CAP didn't kick in, neither did DBA.

Anyway, on the side there is a typewritten sellotaped label

	R1	R2
100V	1.026	1.026
200V	1.027	1.023
500V	1.001	1.016

The corresponding 10Gohm 2% resistors lack such a label, unfortunately.

[ocw]

I'm trying to find the minimum current which I can measure with the 1T feedback resistor...
Time to build a greater voltage divider.

Besides the prior 1/10 and 1/100 voltage division, I added 1/1,000 and 1/10,000 options.  That allowed measurements of 1.6 fA and 160 aA via my ADA4530-1 current to voltage converter using its largest 1T feedback resistor.  The ADA4530-1_cal picture shows my converter with my grounded battery powered calibration box attached.  There's not much unshielded space for power line noise to influence the reading.  While the banana connectors aren't anything special, I'm using Teflon insulated wires.

The 1.6 fA measurement didn't look too bad after the reading stabilized--see the 1.6fA attachment.  Due to the large value resistors used with the small current, the stray capacitance in my circuit produced a significant time constant and delay before the reading became steady.  While the reading did stabilize close to the expected 1.6 mV output level for 1.6 fA, it wasn't steady enough for me to rate it as a success.  With the accuracy at 16 fA adequate to satisfy me, about 10 fA may be the limit for my circuit to have reasonable accuracy.  Based on my results in using increased value resistance, feedback accuracy and my recent results using the ADA4530-1 IC, it's likely if I paid $68.96 for a 10T resistor I would be able to have accurate readings down to 1 fA.

With the mentioned results at 1.6 fA, I wasn't expecting much luck in my final measurement at 160 aA.  I'm not used to dealing with attoamps.  As the 160aA attachment shows, it wasn't a complete failure.  The reading did center around 0 to the desired -0.16 mV to indicate 160 aA.  But, it only points to a very vague low current range.  The big jumps on both sets of today's readings could be due to power supply spikes.

I also wanted to take some readings using my ADA4530-1 circuit with my 1T ohm current limiting resistor along with up to 1,000 volts [with 2 mA current limiting] to take a comparative set of readings using my circuit's different feedback resistors.  The difference in the indirectly measured resistance of my external 1T resistor would partially depend on the voltage coefficient of the external resistor and also on the relative accuracy of my at least 1% accuracy feedback resistors (except for both 1T resistors being 10%).  I don't have any way of confirming which was more influential on my results other than what the 1% accuracy resistors suggests.  The results were:

FB R  |   V to R    |   Voltage Out  |  Measured R    1T       1.0000V        -1.0013V           0.9987T  100G     10.000V        -1.0451V           0.9568T   10G      100.00V        -1.0561V           0.9469T    1G       1000.0V        -1.0915V           0.9162T This has been an interesting project.

[PartialDischarge]

The way I measure picoamps and very high resistances, great equipment 
Didnt found anything on this forum about this gear, so I will post a short review about it




Test resistor,




1 Volt across 100G:

[ocw]

While there are always good (and sometimes not so good) deals available, it seems like building a ADA4530-1 current to voltage converter has a hard to beat performance to price ratio for pA and below current measurements and G to T ohm measurements.  Using a much discussed U1252B meter to measure the converter's output, 10 aA (0.01 fA) current resolution is possible on the lower level currents.  One digit increased resolution is available with my 34401A.  Without zeroing either the converter or the meter connected to it, I had about a 2 fA current reading.  Zeroing the meter was only required for readings where that level of current was significant.

As the attachment shows I saw excellent accuracy with the resistors which I used as feedback resistors at 1gigaohm and below.  The 100 pA and above current measurement typically had around 0.3% accuracy and below.  My lower current accuracy verification was limited by the less accurate feedback and the external resistors which I used to set that current level.  I used a voltage level measured by a 34401A meter and a 0.01% 1M and below resistors and 1% resistors with values up to 100M.  I was limited by using a 10% 1T ohm resistor, which actually had a value close to 0.955T.

It was difficult to accurately verify the accuracy of even lower current measurements.

[TiN]

Let's see if people pay attention what I write. Anyone know where these beauties are from? 

[CalMachine]

Let's see if people pay attention what I write. Anyone know where these beauties are from? 

Your Keithley 6621!

[TiN]

You are no fun.

[CalMachine]

Haha whoops!     Very nice though!  How stable is that reading? 

[ocw]

The attachment shows the output of my ADA4530-1 current to voltage converter when it is fed 250 pA AC current at 500 kHz.
At that frequency/current it has lost its conversion accuracy without taking its high frequency roll-off into account.

[ocw]

I recently completed the modifications of my ADA4530-1 based current to voltage converter.  A picture of it and my accuracy verification figures for it are attached.  It has the conversion ranges of from 1 pA input current producing 1 volt output, to 1 mA producing the 1 volt output.  Think of the current resolution which is provided in the 1 pA range.  Rather than the leakage resistance from my Johnson/Cinch banana input jacks or the Grayhill 56D30-01-1-AJS range switch, I had more problems at the lower currents due to noise and other currents coupling to any unshielded wires outside of its enclosure.  My alignment box solved things by having a N cell battery feeding a 1.0 volt regulator and a current limiting resistor inside of an earth grounded metal box.

My first version had two ranges with restricted frequency bandwidth.  Not finding them beneficial I added two non-decade ranges in my available switch positions.

The accuracy of the unit is primarily limited by the precision of the feedback resistors.  I have their values and accuracy shown on attachment.  The currents being measured are voltage dependent currents, NOT constant currents.  That's what I'm most frequently measuring.  I want to know what the worst case accuracy is, not what the accuracy is with ideal constant currents.  The current being measured was produced by a regulated 1.0000 volt and a current limiting resistor.  The accuracy of that resistor is shown on the attachment under the "Standard" "Accuracy" column.  So, the ADA4530-1 Accuracy column includes the insignificant amount of burden resistance/voltage in that IC.  The single digit fA input bias current of the op amp didn't effect zeroing the circuit's output until that amount of current became significant.  The external voltage meter can be zeroed if that value becomes important.

With an added external voltage or resistance and a simple calculation, resistance and voltage can also be measured.  The circuit has a very high input impedance for voltage measurement.  I've measured up to 1,000 volts using it.

I am surprised that a company doesn't use a similar circuit for their commercial circuit.  It would have a much smaller burden resistance and voltage for current measurements than what they are using now.  Greater value resistors could be measured.  And the circuit would have a higher impedance for voltage measurements.  I'd design the meter so that a 10M resistor could be added during voltage measurements for times when its high impedance causes more harm than good.  And, given the relatively low cost of the components needed, it could be used in a multi-meter with a cost significantly less than that of the typical electrometer.

[Echo88]

Are there any recommended high insulation reed relays with >= 10 Teraohm, which are actually available and reasonable priced? Of course a few old COTO 1240 are available via ebay, but they are NOS. I was rather thinking about the HI-series from Meder, e.g. HI05-1A85, but those are non-stocked by Digikey/Mouser.

Maybe someone has measured a few common reed-relays which proved to have >= 10Teraohm and can make suggestions?

Reason for this question: Im building a multiplexed electrometer-box, to measure the leakage current of a few components automatically. At the moment im testing a few MEDER NP03-1A81-BV247, but they are barely reaching 1 Teraohm after a long time and id like to get better results.

http://www.farnell.com/datasheets/1486738.pdf COTO 1240 Datasheet
https://standexelectronics.com/wp-content/uploads/datasheet_reed_relay_HI.pdf
https://www.mikrocontroller.net/attachment/279340/3103181100D.pdf Meder NP03 Datasheet

Edit: Meder claims that the HI/CRR/CRF/KT-series are made to be high insulating, while the datasheets are very conservative with mostly "minimal 1 Teraohm spec": http://4a30d8fd18dae1bf393d-df49f4cedb726ad03ad145d2e3d346bd.r41.cf5.rackcdn.com/fileadmin/meder/pdf/en/Technical_Documents/Technical_Write-Ups/Electrometer_Detection_Circuits_Use_Reed_Relays.pdf
I will order a few of the CRR and KT-series-relays, measure them and then report back.

[antintedo]

I've been looking for them recently as well. HI05-1A66 is stocked by digikey. There are also a few others from the same manufacturer that have 10T minimum insulation.

[Echo88]

Only the HI-series with the -85 suffix, like the HI05-1A85, claims 10 Teraohm insulation. And this type is not stocked by Digikey/Mouser. They are available from other US-sources, but cost ~20$ each. Therefore i will first test the CRR/KT-relays.

Is it possible to buy the reed-relay-casings (case with electromagnet and shielding, were you insert the reed-contact) somewhere? That would allow one to use good reed-contacts like the mentioned MAAR-5: https://www.littelfuse.com/~/media/electronics/datasheets/reed_switches/littelfuse_reed_switches_marr_5_datasheet.pdf.pdf

[e61_phil]

I don't know if one can buy such housings, but I know a company, which provides 10T amplifiers which are switchable, very good . And they did it by their own, because you don't want to touch the sensitive net with anything.

[chuckb]

I checked one of the MEDER reed relays (Digikey) for current leakage (PN HI05-1A66) with the K617 Electrometer. The DUT is suspended in air inside a metal box. Test conditions 23 deg C, 35% RH.

With 31V across the contacts I had 12 fa flowing. That worked out to about 2,600 Tohms. With 100V it was 42 fa. The coil to contact leakage was 4.6 fa with a 31v Bias. It looks like a very good high isolation relay.

[serg-el]

D series Cynergy3 Components Ltd
https://www.digikey.com/product-detail/en/cynergy-3/DAR71210/725-1047-ND/751989

[e61_phil]

I checked one of the MEDER reed relays (Digikey) for current leakage (PN HI05-1A66) with the K617 Electrometer. The DUT is suspended in air inside a metal box. Test conditions 23 deg C, 35% RH.

With 31V across the contacts I had 12 fa flowing. That worked out to about 2,600 Tohms. With 100V it was 42 fa. The coil to contact leakage was 4.6 fa with a 31v Bias. It looks like a very good high isolation relay.

Did you try to change the polarity to verify the measurement?

[chuckb]

I did this test several months ago. I typically Suppress (zero) the 10 fa of offset current before applying the voltage. The zero is usually stable within 0.5 fa after the machine is warmed up and stable.
I will recheck the leakage across the relay contacts with both polarities later this afternoon.

[chuckb]

With +31 V bias the MEDER HI05-1A66 had 10.5 fa of leakage across the contacts. With -31 V bias it had -12.5 fa of leakage. So the average leakage at 31 V bias is 11.5 fa. This equates to over 2000 T ohms isolation resistance. I have previously measured the contact to coil leakage current being 1/3 of the contact to contact leakage current. This relay has been sitting on my bench for several month with no cleaning before this test. Hope this helps.

[chuckb]

I have also tested the  COTO 9012. This is a very small relay for less than $10 USD. The contact isolation resistance is 4000 T ohms. The coil to contact resistance is 1000 T ohms at 100V.

[Vtile]

With my Marconi / Meratronic Meratester I'm in like a Flyn. Sensitivity (noticeable relative change) is somewhere in region of 50 fA. 

https://youtu.be/3HFX_nMm6D0

[branadic]

We do have one of those brown bricks, Keithley 617. Nice piece of kit. I ordered some Ohmite high value 1% metal oxide resistors (100meg, 1G, 10G) and build a small setup with the resistors mounted on teflon standoff. After all these years without being calibrated the readings on the meter are still consistent.

-branadic-

[TiN]

Accidentally bought myself Xmas gift... 



Source is Keithley 6221.

Oh well... I hope buying not broken modern equipment will not become a habit. 

[Inverted18650]

TiN, I am going to message you about the Keithley 417 PicoAmmeter w/ 4170 input head.

[TiN]

Thanks for offer, Inverted18650, but perhaps some other eager amp-nut would like that 417. I already have enough of backlogged projects. 

[PartialDischarge]

That is a beast, can handle fA like nothing wow, minor teardown at least coming?

[VintageNut]

I believe that I have shared these resistors on a different thread. For this thread, these resistors are intended to be used with an electrometer to check calibration points.

One end of each resistor has a BNC jack. The jack is for the voltage source of the electrometer. Any leakage from HI to LO through this jack is not measured. So, this jack can be a junk BNC with no negative effect on it intended use.

The other end of the resistor is a 3-lug triax jack. This jack is for the triax jack of the electrometer.

These resistors can also be used with a sourcemeter. Care has to be taken to account for any leakage (if any) associated with the BNC jack.   

I had these calibrated at a primary lab in 2017. I will be taking them back again for a cal in February of this year.

I have no idea about temp dependence of these. That info would be needed for any rigorous cal usage. 

[e61_phil]

Could you say some words about the calibration in the lab? How did they cal these resistors?

[VintageNut]

Could you say some words about the calibration in the lab? How did they cal these resistors?

I can sketch the methodology, but I do not have total command of the calibration procedure.

It was explained to me that the calibration system is a bridge. The bridge uses a very old collection of Keithley 5155 resistors that have a long calibration history. The bridge also has an old Keithley electrometer. The remainder of the bridge is unknown to me and the exact procedure is also unknown. I will try to find out more details when I visit in February.

Attached are the calibration reports

[VintageNut]

I double posted and missed one. Here is the missing one.

[VintageNut]

Since you asked about resistor calibration, attached is a calibration report for my Fluke 5450A. It was calibrated against two Measurements International bridges. The low ohms are calibrated with a constant current bridge and the high ohms are calibrated with a constant voltage bridge. You can see that the 100M resistor in the 5450A is measured with better than 10X uncertainty compared to the 100M stand-alone resistor that I built.

Every measurement on the MI bridge requires a soak time of about 30 minutes. Very time consuming.

[Inverted18650]

That is a beast, can handle fA like nothing wow, minor teardown at least coming?

We're you asking me or TiN? 

[PartialDischarge]

That is a beast, can handle fA like nothing wow, minor teardown at least coming?

We're you asking me or TiN? 

Oh I was asking TiN, that shiny new equipment calls for a teardown

[Echo88]

Question to the Electrometer-ninjas about the K617:

As far as i understand guarding is only usable when measuring voltage sources with high source resistance and when measuring high resistance DUTs with the electrometer-internal-current-source up to 200GR.
With the internal voltage source the K617 is capable of determining the resistance/leakage current of higher resistance DUTs like a reed contact, but without guarding it takes a great amount of measurement settling time.
Is there a way to get faster settling time with additional equipment like an external current source or would i need a better Electrometer?

Attached is the measurement of a reed contact MDSR-10, where the long settling time is visible. Cable used: 1.5m Keithley-7078-TRX

[guenthert]

Question to the Electrometer-ninjas about the K617:

As far as i understand guarding is only usable when measuring voltage sources with high source resistance and when measuring high resistance DUTs with the electrometer-internal-current-source up to 200GR.

I'd rather think of guarding being needed only when very low currents are to measured (you could have very low currents with voltage sources of low source resistance, but those would be very small voltages which then pose additional challenges).

With the internal voltage source the K617 is capable of determining the resistance/leakage current of higher resistance DUTs like a reed contact, but without guarding it takes a great amount of measurement settling time.
Is there a way to get faster settling time with additional equipment like an external current source or would i need a better Electrometer?

I don't quite follow.  Why can't guarding be used here and in which way would you want another instrument to be 'better'?

[Echo88]

The K617 manual (page 30) advises to not use guarding in Amps or Coloumbs-mode and it also doesnt show a how a guarded charge/current (when using the internal voltage source instead of the internal current source)-measurement would look like.
Thats why i thought the K617 isnt capable of guarding when measuring current/resistance using the internal voltage source.
The idea of "better Electrometer" meant that newer electrometers are capable of measuring higher resistance with their internal current source while providing guarding, which reduces the settling time.

[guenthert]

  Yeah, sorry, not sure what I was thinking there.  In current measurements, one can expect a very small difference in potential across low and high inputs (<1mV "burden voltage") which makes guarding typically unnecessary. 

  Do I understand the problem now correctly as being that the resistance of the DUT is expected to be so high (>200GOhm) that direct resistance measurements (where the 617 offers guarded operation) is not possible and for the U/I mode (which allows for notably higher resistances, but where there's no option to use the guard) measurements are slow (due to stray capacities)?

[Echo88]

Yep, thats the problem.

[MiDi]

The guard is driven from the preamp output.
In Volt and Ohm mode the preamp output is just the buffered input.
In Amp (V/I) and Coulumb mode it does not follow the input voltage.
As the input is held close to 0V, that should act as a guard to bootstrap input/stray capacitance.
See p. 116/6-4 in the manual.

[MiDi]

The last post is simply wrong, next try well-rested

For Amp (V/I) and Coulumb the inner shield is tied to signal gnd - for both normal and in guarding mode.
It would not make any difference, except for active guarding there has to be a return path to signal gnd outside of the triax cable.

[Alex Nikitin]

Attached is the measurement of a reed contact MDSR-10, where the long settling time is visible. Cable used: 1.5m Keithley-7078-TRX

I would advise you to repeat this measurement with just an air gap of roughly equivalent capacitance as the reed switch. You may find that the settling time is considerably reduced. If that is the case, the apparent settling time is the result of the dielectric absorption of the reed switch glass, so it is a real effect of the DUT, not of the electrometer used.

Cheers

Alex

[Echo88]

Thanks Alex.
I did the following tests:
-reed-contact-surface covered ~80% with aluminiumfoil and foil connected to V+ while sitting on a teflonplate
-reed contact airwired between electrometerinput and voltagesourceoutput and without sitting on teflonplate
-electrometerinput and voltagesource-output sitting on the teflonplate without reed contact

You were right: the settling comes directly from the DA of the reed-contact, since i can see only a very short settling time when no DUT is present and the teflon-plate isnt the DA/charge trapping-culprit.

Im a bit disappointed that reed-contacts arent behaving better and would have expected that the aluminiumfoil-cover would help.
Ive seen that you measured reed-contacts with same results regarding settling-time, any hints on what can be improved?

[MiDi]

When I tested the old brown relays of my unit, the better ones had settling times in the order of an hour for more than 100T
In comparison the measured Meder Relay took less than half an hour to reach 10P
Besides these measurements I have no experience, but I would guess that there are huge differences in DA for different relay types.

[Vtile]

Can these electrometers measure the leakage of typical small engine spark plug (dry&clean, but USED). I'm just wondering... "Metrology object". 

The backstory is that some time ago I was repairing two motorsaws of mine, which both had problems, one later was diagnosed a failed magnetocoil. The thing related to this threas was the idea to just quickly measure the difference between the really used and soothy plug and the new replacement with only a few tens of minutes runtime.

...It was not as easy as I thought, not a quick check of relative leakage current with my Meratester in series with 10Vdc source and dut (fA range). Well I did crank up my lab psu to full 60V ... nothing.

Not only after I changed to 1000Vdc source in series I had some form of reading. IIRC somewhere at Teraohm region.

Which made me wonder if I reached the limits of my setup, but obviously they must have a good insulators after all, but still.

[HighVoltage]

I have tried used the Keithley 617 and the 6517B to measure the insulation resistance of new spark plugs and that failed, because the pure Aluminium Oxid (Al2O3) insulator is better than the ATTO Amp resolution of the meter.

But the 6517B and also the 617 can be very well used to measure the insulation resistance of ignition coils.

[bsw_m]

the ATTO Amp resolution of the meter.

Resolution != sensitivity. 

I'm measure leakage current for PTFE and saphire insulators in BNC triax connectors with test voltage 15V.
Leakage current for PTFE insulator is about 10E-18A...25E-18A(noisy reading). And 16E-18A for saphire insulator. But PTFE is worst for noise and settling time.
And I think that no problem to measure insulator leakage of engine spark plug.

[MiDi]

Why would one want to get exact numbers for leakage resistance of spark plugs?
If it cannot be measured with electrometer, it simply is high enough 
More relevant for spark plugs is if the spark ignites where it is supposed to be and nowhere else...

[guenthert]

     That and further the 'leakage' of a bad spark plug will be break-down, not ohm'sche resistance.  So if the plug looks ok at 100V the electrometers can drive, it isn't necessarily at 10...100kV operational voltage.

[HighVoltage]

Well, I must clarify that of course measuring a used spark plug for insulation resistance does not make any sense at all.
At least not for the purpose to find out if the spark plug is good or bad.

The only reason why I tested some new spark plugs was for pure science to find a high enough resistance to see where the limit of the instrument was.

[Vtile]

Why? Because one can (to try at least ). ...and pure technological interest what (not pure) CARBON deposition on high insulator surface will do etc...

Ps. I do not have electrometer nor picoammeter. Just household analog multimeter (which just happens to be one of the most sensitive chopper-Fet multimeter models ever build). Made in Poland 1980.

Pps. One could assume that surface that is as black as seen on picture would be something like 1Gohm at max... NOT.

[HighVoltage]

Vtile:
Best to open a new thread for testing insulation properties of spark plugs.
This discussion does not really fit in this metrology section.

[Vtile]

Testing insulation of spark plugs/gaps via resistance makes absolutely no sense and is just a waste of time. It should be done like a dielectric strength test (IEC 60243) with a high voltage source (meaning a source of at least 10kV) until discharge occurs

Most of the things in life do not make any sense, metrology included tbh. I was not interested at all of how the townsend avalance forms on sparkplug, or how the dielectric strength of the material might statistically behave.

Vtile:
Best to open a new thread for testing insulation properties of spark plugs.
This discussion does not really fit in this metrology section.

I might (but do not know where as of there is no "measurement and technologies" section other than metrology), but there were reason I did use this particular thread to begin with Measuring nanoamps and below like a Ninja is not the most scientific subject window if I'm not terribly mistaken.

Besides the outcome is the same "spark plugs/gaps via resistance makes absolutely no sense", we definitely would need a "measurement and technologies" section where people could actually discuss measurements without loosing scientific credibility.

[Vtile]

I did create a sparkplug interest group as this seems to generate so much discussion.
https://www.eevblog.com/forum/repair/sparkplug-resistance-and-dielectric-and-surface-avalances/

[Alex Nikitin]

I think it is better to continue in this thread, to keep all this information in one place. Recently I've saved from the bin a small number of proper electrometer grade reed relays, custom made by Crydom/synergy3 probably  about 9-10 years ago. I've measured (using the Keithley 617) the leakage across the open contact at 100V for this relay and for the Meder HI05-1A66 already mentioned in this thread. Results are below, 5 min at 0V, then 5min at +100V and  5min at 0V. The electrostatic screen on the Crydom relay was grounded, on the Meder I've grounded the coil at one end. As clear from the graphs the Crydom has about 20 times less leakage than the Meder and also almost no visible dielectric absorbtion. I have also tested both relays switching 100mV through 10G resistor = 10pA current and both relays performed well.

Cheers

Alex

P.S. The room temperature was ~26C, RH ~ 55%

[1audio]

To add a small datapoint to this thread I just noticed that my HP 4329A High Resistance Meter has a current range that goes down to 50 pA (with significant limitations).It uses a parametric amplifier for the low current measurements. https://www.keysight.com/us/en/product/4329A/hr-meter.html

[HighVoltage]

@Alex Nikitin
Thanks for the nice graphs and pictures.
May I ask, what is your method of getting the data from the 617 to such graphs?

[Alex Nikitin]

Oh, normally I would just use GPIB directly from the 617, however right now the GPIB interface on my 617 doesn't work  , so I use one of my two Keithley 2015, connected to the 2V analogue output on the 617, and read over GPIB, the data is then imported into Excel.

Cheers

Alex

[guenthert]

Oh, normally I would just use GPIB directly from the 617, however right now the GPIB interface on my 617 doesn't work  , so I use one of my two Keithley 2015, connected to the 2V analogue output on the 617, and read over GPIB, the data is then imported into Excel.

Cheers

Alex

     When I got my 617, the GPIB connector was bust.  On my 181 (same vintage) however both output drivers needed to be replaced (and fortunately nothing else).

[bsw_m]

@Alex Nikitin, your post piqued my interest in measuring custom relays produced by MNIPI, which were used in their electrometers.
I checked how the relay commutes the current 1pA.
Leakage of an open contact at a test voltage of 100V(But here it is more likely injection through the capacitance of the contacts, and not leakage. Since in fact, structurally leakage in this relay is impossible. The absence of leakage was checked using a V7-45 electrometer, the sensitivity of which is not enough to measure any leakage at a voltage of 100V ).
The measurements were made with a V7-49 electrometer.
Charts were built according to data from V7-49
The resistance of the closed contacts of these relays is 50mOhm
Everything else is presented in the graphs.

[bsw_m]

Measured the input bias current of the op-amp ADA4530-1
And the input bias current of the NT002 - an old electrometric assembly of transistors designed and manufactured by MNIPI. This assembly of transistors was introduced in the early 80s.
ADA4530-1 was used as buffer (measured Ib+ current) and was powered from ± 6V
NT002 was used in a similar connection as it is used in electrometers developed by MNIPI. Any input bias current compensation not used.
T ambient during measurement: 22℃ 

Blue line - is ADA4530-1
Please excuse me for the x-axis in samples. To translate into seconds, divide by 10 (10 samples is equal to one second)

[bsw_m]

NT002 teardown and DIE photo: https://www.eevblog.com/forum/projects/transistors-die-pictures/msg3713347/#msg3713347

[Echo88]

While trying to find some N- or P-Channel Enhancement MOSFETs with the following specs Udsmax >=20V, RDSon <=20R and Ids-leakage (with gate-source shorted) <=1pA at Uds=10V i tested a few already and just wanted to list them here with measured leakage values.
I didnt find any similar examples for mosfet leakage (Art of Electronics X-Chapters has a JFET/Bipolar leakage chart on page 102 though), so those may be of interest to people who want to know which round about real values can be expected compared to the very conservative/automatic test equipment limited values in the mosfet-datasheets. 
Leakage measurements were generally done with Uds = 10V, as thats the maximum voltage in the circuit that the mosfets would be used at in my circuit.
Sometimes i tested also at lower voltages to see how the leakage behaves.
Maybe you guys also have mosfet-measurements, especially of those with <=1pA?

IRLR2908: N-Channel 80V 30A 28mR D-Pak
0.5V: 10.7pA
1V: 28.9pA
2V: 71.8V
10V: 69pA

SUD50P10: P-Channel -100V -35A 0.04R TO252
-10V: -196pA

SI4410DY: N-Channel 30V 8A 14mR SO8
1V: 395pA

SI4467DY:  P-Channel -12V -12A 11mR SO8
-10V: -1040pA

SI9407: P-Channel -60V -3.5A 0.15R SO8
-10V: -244pA

SI2327DY: P-Channel -200V -0.5A 2.4R SOT23
-10V: -68.6pA

SI2337DS: P-Channel -80V -2.2A 0.2R SOT23
-10V: -31.6pA

2N7002L: N-Channel 60V 115mA 7.5R SOT23
5V: 14.65pA
10V: 16pA

2N7002L: second specimen:
10V: 22.1pA

2N7002L: third specimen:
10V: 22.8pA

NTS4409: N-Channel with Gate-ESD-protection 25V 0.75A 250mR SC70
1V: 2.9pA
10V: 3.2pA

PMV130ENEA: N-Channel with Gate-ESD-protection 40V 2.1A 120mR SOT23
10V: 40.6pA

BSS131: N-Channel 240V 0.1A 14R SOT23
10V: 0.6pA
20V: 0.73pA

All measurements done with my trusty K617. 

[bsw_m]

PVT412PbF mosfet solid state relay:
Leakage:
at 20V: 6.5pA
at 300V: 107pA
EMF at 10mA led current: 0.6uV

Measurement done by my V7-49 electrometer and V2-38 nanovoltmeter.

[bsw_m]

Tests for AGN210S4H relay.
Leakage current after 20min with 50V test voltage is 1E-16A
1pA current commutation very good, but with spikes from charge injection.
Charge injection is big, about 12pC.

[Echo88]

@bsw_m:
My tests on mosfet solid state relays indicated that a lower LED drive current reduces the mosfet output EMF.
When the led current is just above the threshold to switch on the mosfet output it should produce the lowest EMF.
In my case the measured EMF was <100nV afair at about 1 or 2mA LED drive current.

I also measured the PVT412 a while ago and came to nearly the same leakage value at 20V for three tested samples: 5pA 
Heres the thread with my leakage results https://www.eevblog.com/forum/metrology/measurements-of-leakage-current-and-offset-voltage-on-some-optofets-and-relays/

[David Hess]

My tests on mosfet solid state relays indicated that a lower LED drive current reduces the mosfet output EMF.
When the led current is just above the threshold to switch on the mosfet output it should produce the lowest EMF.
In my case the measured EMF was <100nV afair at about 1 or 2mA LED drive current.

Is that a thermocouple effect from self heating of the optocoupler by the LED?

[bsw_m]

Is that a thermocouple effect from self heating of the optocoupler by the LED?

I think, yes.

[Echo88]

Yeah, i think its coming from the heat, leading to a thermal offset.
Datasheets like the PVA3054-DS also refer to it as "thermal offset voltage" http://www.irf.com/product-info/datasheets/data/pva30.pdf

[David Hess]

That would explain why I have seen old precision designs using optocouplers with the emitter separated from the detector by a light pipe, although this is also done to reduce coupling capacitance.

[Kleinstein]

For photo-mos couplers the separation can also be at the detector side. There are 2 seprate parts: some photovoltaic opto-coupler to produce a voltage and than back to back MOSFETs that are controlled from that voltage.  The PV coupler and the MOSFETs can also be a bit apprat. It still is a copromise with size and leakage.

[Vgkid]

Nice.
Looking forward to the updates.

[MegaVolt]

Looks like sweep error into range +10V...-10V(20V span) with programmable speed 1mV/s is less 100nV/s

What were the measurements taken in relation to?

[julian1]

Looks nice!. What is the purpose of the blue (film caps?) on the toroid? To reduce diode reverse recovery ripple on the rectifier?

[Kleinstein]

The capacitors at the transformer are a bit unconventional, but they are a good idea, filtering some of the mains and in addition providing some of the magnetization current from the secondary side too.

The usual way are capacitors (usually smaller ones) in parallel to all diodes at the rectifier. With the capacitors at the supply the shown caps are mostly equivalent.

With a low voltage they don't have to be X2 type, but it does not hurt if they are.

[Kleinstein]

There may be a tricky point with some of the pA meters: some of them want a minimum resistance for the current source and they may not be stable with too much capacitance at the input. The fault is on the side of the meter, but still a point to check and if needed add some series resistance.

[MacIntoshCZ]

I ve just seen that video and truly experienced RUSIA POWEER
Palpatine would be proud of you.

[HighVoltage]

On top - P597 100pF air capacitance standard and Micron-GLIN.

You have interesting things !

[miro123]

The capacitors at the transformer are a bit unconventional, but they are a good idea, filtering some of the mains and in addition providing some of the magnetization current from the secondary side too.

The usual way are capacitors (usually smaller ones) in parallel to all diodes at the rectifier. With the capacitors at the supply the shown caps are mostly equivalent.

With a low voltage they don't have to be X2 type, but it does not hurt if they are.

Thanks for sharing - I enjoyed to follow your post. I even tried to read the Russian text without translator.
1. Do you try to follow the datasheets e.g. LM317 chapter 9 all figures has 0,1u capacitor. Chapter 11 gives layout example
https://www.ti.com/lit/ds/symlink/lm317.pdf

2. LM750l05 is LDO - I cannot find the datasheet. Generally speaking LDO don't like low ESR caps at output. normally LDOs datasheets shows safe operation area Capacitance vs ESR vs Load current
3. X or Y caps are used at line inputs they have special safety properties.  The art of electronics fig.9.49-> Damping circuit has Rs and Cs, you hac only simple  low ESR Cap like #1

[SilverSolder]

I thought the capacitors are supposed to be the same value in a Sallen-Key section?

Edit:  I guess you can do it either way!

[David Hess]

I thought the capacitors are supposed to be the same value in a Sallen-Key section?

The capacitor values are only equal for a specific gain of the amplifier.

[Kleinstein]

Is it sure the nonlinearity is from the GLN and not from the DMMs ?
A simple test to do for this would be to repeat a test with the DMM reversed. This would at least change some of the DMM caused INL, especially the even powers.
The rather wide curve from the 6500 looks a bit like this could be an effect of the meter, maybe an effect of a limited number of digits or internal resolution limitiation.

I would expect the time base errors to be very small, so no worries there.

Another possible error source may be DA of capacitors in the filter. There is a large 10 µF MKS cap on the board. Is that doing ouput filtering (could cause trouble) or is it more reference filtering (would be OK) ?
10 µV are still quite a bit and usually more than to expect from thermal EMF,  and the loading error to an OP like OPA140. For the OP error I would be more afraid of output cross over, that may cause some 1 µV step when the polarity of the current changes, though still a relatively small error.

[KT88]

My preferred idea for today: op-amp SHE(self heating effect) and load problems of OPA 4140 which is used as R2R matrix power driver. Op-amp error may expand DAC errors.

An easy litmus test would be an additional load resistor...

[KT88]

I was thinking of a parallel load for the opamp just to increase the self heating of the opamp die...
If it changes you identified the culprit - if not the root cause is elsewhere.

[Kleinstein]

With the given INL problem, the first point would be to identify the cause of the INL.  Chances are high that this is not the clock reference or the votlage reference. A new version may be needed to fix the INL problem, but not for a better voltage or clock reference.

The candidates for the INL are more the DAC,  self heating and maybe a routing problem (e.g. coupling the variable reference current to the DAC to the signal or reference).  The exact timing when the DMM actually does the measurement could at least be a factor for the DMM6500, causing the wide band with moire like patterns.

[ramon]

Yes, that is a 1W heater. Three years ago there was someone on the action site selling 'Ultra Low Power OCXO' (150mW).

[Kleinstein]

To make full use of the DDS part one should have at least some filtering between the DDS chip and the comparator for the sine to square wave conversion. So the minimum would be some capacitor in parallel to R2. With a relatively low ouput frequency to look at a relatively simple filter can be sufficient.

I think there should be additional decoupling / filtering (e.g. a seires ferrite and / or resistor) at the supplies, espeically for the oscillator. Without a clean supply a OCXO may not help very much and chould show off jitter and FM effects.

The auxiliary clock output may want a series resistor to get an approximately 50 ohm output impedance.

Edit:
For the sine to square wave conversion I would prefer a true comparator or maybe just a schmidt trigger logic chip like LVC1G14. The OP would have a relatively slow transition.

[MegaVolt]

Colleagues advise: NC7SV04, inverter, with a resistor about 10K between the entrance and exit. Noise at -175DB / Hz.

[Kleinstein]

In the first curve with the DMM6500 there is a broad band, but it does not really look like noise, more like a moire pattern and thus more like a systematic error. I could imagine a problem with the time stamps. The keithley meters also tend to show some odd low frequency extra noise that seems to be related on how the AZ mode is done and this can also effect the time stamps. This can especially occur with higher PLC values that are averages over multiple shorter readings and thus a bit tricky with the time when the measurement is actually done.

The newer curves show some odd periodic pattern with some 0.2 V "period". This looks odd and may indicate some rounding problem with the corrections.

[Kleinstein]

I totally agree that the  I = C * dV/dt way is very nice for low currents, lets say below some 100 pA or 1 nA.

I think one would definitely need to have some low pass filtering - othewise the DAC step would result in current spikes - in the extreme case the current is more like a series of charge packets from the steps in the DAC. This may be OK for some ampmeters that measure the current with the integrator method, but could be a problem for other, especially faster meters, that can resolve the spikes or get to a nonlinear range.

In some cases one may need some series resistance anyway (some pA meters need it) and this resistor may do part of the filtering.

I have not used the capacitor way to generate small currents, but I like it as a convenient way to measure small currents (pA range). Compared to resistors in the Gohms range small capacitors (e.g. 100 pF range) are much easier to get and also quite stable and there is less noise.

[Kleinstein]

The LTC2057 may not be that happy with an input impedance of some 660 K. Its voltage / current noise is matched for some 70 K and it would thus be dominated by current noise. Chances are it can still be good enough, at least with a higher voltage.

R16 and R17 look like they could run quite hot. I don't think the Zener diodes at teh protection would need that much current and the resistors may be better a bit larger than 0603.

R19 is quite low in values: this would be quite some load (up to some 2.5 mA) to the output driver and thus quite some power due to the relatively high voltage.

I think the gain switching would interfere with the low pass filter function at the input. With gain the SK filter changes and in some cases can become unstable.

The rather high voltage could be a problem for the LT5400, as it would run rather hot. Ideally one would want higher resistance, though the 100 K version would be a bit on the high side.

[Kleinstein]

With there nominal load and normal PCB layout SMD resistors get uncompftably hot. With leaded solder there is even a chance they may melt the solder. So it is a good idea to stay well below it.  The power rating is valid if used with quite some copper or thick traces. Not sure the 1206 size when used with 8 mil traces would be OK with 100 mW. Also power from neighboring parts add up.

For R17 at the force output one could consider a polymer fuse / PTC instead of a normal resistor. This would also limit the worst case load to the OP in case of a short and reduce the chance for overheating there. I am not sure the LTC2057 is safe with a near short with a +-30 V supply.

Anyway for a precision circuit it helps to keep the overall power consumption low to keep the whole circuit cool.

The load due to R19 is laoding the LTC2057. With 15 V ouput this are 1.5 mA and thus 22.5 mW extra for the LTC2057. With a thermal resistance of some 100 K/W for a SO8 case this would be an extra 20 K temperature rise for the already warm OP due to the high supply voltage.
For best precision chopper OPs should normally be used with light loads and for higher loads one may consider an extra driver (e.g. transistors as emitter followers).

The filter with gain can be still stable in the used configuration. The change in gain would still effect the cross over characteristic. So with changing the gain the cross over frequency and Q values would change. This may not be a big problem, but one should be aware of this effect.

The input current to a chopper is mainly as current spikes up and down from the chopper action. Here asymmetry gives an average input bias and much of the energy is at high frequencies, but there are also variations in the chopper spikes that result in an
about white current noise part that also extends down to low frequencies.

The capacitors help with the chopper spikes - though they still load the output. However they help little with the lower frequency part of the current noise. Due to the low bandwidth this noise may still be acceptable. Changing the resistor values to a lower value is still a reasonably easy change if needed, though this means a higher filter cross over.

A 1 Hz cross over for the fitler may anyway be relatively low for some cases as it means that one would have to wait several seconds from the start of the slope before getting the final slope (in theory SW could speed this up).
On the other side the 766 Hz filter at the amplifier may not have much effect with an update frequency for the DAC in the 100 Hz range. It can still be useful to dampen the DAC switching spikes. I think the relevant frequencies for filtering would be about 5 Hz to 200 Hz.

[SilverSolder]

My dear friend, please don't worry about load of precision circuits... a few days ago i try use my new precision resistors as led current driver, they works fine!


 

I am beginning to like the Russian approach to precision components! 

Can a precision circuit be powered by one of those cute little nuclear reactors that were used to power lighthouses? 

[TimFox]

LED driver:  32 mA will make most LEDs bright.

[Kleinstein]

The BAV99 diode is a fast diode and thus relatively high leakage.  If lower leakage is needed one could use a BAV199, that should come in the same package. It is a slower diode and also higher forward voltage, but I think this should not be a problem.

[RandallMcRee]

The mkp4 seems to be “bad”. DA sb around 0.0002

My 10 all measure similarly.  Also my Russian K71-7 is lower ( but larger ) as you demonstrate.

Randall

Edit
My bad—Shodan said DA not D, dissipation! Ignore me!

[ZhuraYuk]

I think DA means dielectric absorption, but D on LCR meter stands for Dissipation factor. Which is a different thing

[fcb]

The mkp4 seems to be “bad”. DA sb around 0.0002

My 10 all measure similarly.  Also my Russian K71-7 is lower ( but larger ) as you demonstrate.

Randall

How are you measuring DA?  My understanding is this is not a trivial thing to measure and not available on your typical LCR meter?

[TimFox]

There is a mil-spec measurement for dielectric absorption (DA):  a specific sequence of applying a certain voltage, disconnection, waiting, and re-measuring voltage.  It is good as a "figure of merit" to compare different capacitors with each other, but I don't understand how to use that standardized value to predict the effect in a practical circuit, with different voltages and times, since it is a non-linear effect (analogous to ferromagnetism).
"D" = 1/Q is closer to being a linear effect, but depends on frequency.  For non-linear dielectrics, such as Z5U ceramic and electrolytic capacitors, it also varies with voltage.

[TimFox]

The US standard test for DA is similar.
Do you know any method for using that value to predict the error in a different circuit with different timing?

[Kleinstein]

There is a mil-spec measurement for dielectric absorption (DA):  a specific sequence of applying a certain voltage, disconnection, waiting, and re-measuring voltage.  It is good as a "figure of merit" to compare different capacitors with each other, but I don't understand how to use that standardized value to predict the effect in a practical circuit, with different voltages and times, since it is a non-linear effect (analogous to ferromagnetism).
"D" = 1/Q is closer to being a linear effect, but depends on frequency.  For non-linear dielectrics, such as Z5U ceramic and electrolytic capacitors, it also varies with voltage.

With film capacitors the dielectric absorbtion is usually still linear. So more voltage gives more recovered valtage. The DA itself may be nonlinear mainly with electrolytic and class 2 /3 ceramic.

The standard DA test measure the effect for a certain time window (time constands from some 5 seconds to some 3 minutes for the russian version and a little longer for the US version). They still don't seprate it for the different times and a calculation is thus only approximate.

Form the tests I have done so far the PP and PS caps the DA seems to be distributed over a wide range of time constants and usually not just a few discrete ones.  An about even distribution on a log time scale seems to be a reasonable assumption to extra polate to shorter and maybe slightly longer times.

To do a calculation of can use the equivalent model with extra RC elements in parallel, e.g. with 1 or 2 capacitors per decade (in time/frequency). The effect is quite broad in the frequency and thus no need for more elements, as we don't have more specific information anyway. The standard tests covers a little less than 2 decades in the time/frequency. So the size of the extra capacitors would be about 60% times the measured standard DA value for 1 RC element per decade and than sereis resistors 1 decade. Alternative half the size with 2 RC elements per decade. One could than use this equivalent ciruit in a spice simulation of the circuit in question.

[TimFox]

to Kleinstein:
In my career, DA was something to avoid rather than quantify, so I don't have personal experience with the parasitic R-C networks that are often used to discuss DA.
However, in that model, which is linear, DA would not cause any harmonic distortion in the voltage across a practical capacitor driven from a pure-frequency sinusoidal current.
If DA is from ferroelectric hysteresis (presumably true with electrolytics and the crummy ceramic dielectrics), it becomes a non-linear problem.
Respectable audio literature has many examples of careful measurements of audio-frequency THD from capacitors, possibly the result of DA.  One series of measurements showed 0.01% THD on some metallized-polyester devices, where the same measurement on a polystyrene capacitor or a C0G ceramic showed unmeasurable THD.  The same author, M. Blencowe, points out that silver-mica capacitors have mediocre DA but do not show THD in audio tests.  I remember reading the General Radio literature on the choice between polystyrene and mica in standard capacitors and decade boxes, where the micas show an increase in apparent capacitance at low frequencies, which GR attributed to "interfacial polarization" at the mica-silver interface.
Before retirement, my main concern with DA involved the hold capacitor in sample-hold circuits.  Lacking a good quantitative handle on the problem, we started with polystyrene back in the early 1980s with TH boards, later changing to polypropylene, but had to settle for C0G ceramic in the later SMT boards.  (We could not justify using PTFE.)

[Mickle T.]

GOST 28885, DIN 41328-4 etc are useful for capacitors selection, but almost useless for any other thing, like precision circuits design. However, there are a large number of publications (mainly by American and Soviet authors) that describe other methods for modeling dielectric absorption and measuring the parameters of its approximation models.

[TimFox]

Thanks for the references.  I quickly read the third one with its methodology for estimating the parasitic R-C networks, but it was confined to a linear model (since it was based on vector network analysis).  I still am interested in the effect of hysteresis.

[Kleinstein]

I looked at low DA capacitors for my multi-slope ADC design. There  I tested  polystyrene, polypropylene and different C0G versions. Found these I found the lowest DA with TDK brand C0G caps, while the other C0Gs I tested were similar the polystyrene and PP.  The good C0Gs were about 1/5 the DA of PS or PP capacitors. So the need to change to SMD may not be so bad.  I did most of the testing with a shorter time scale (e.g. 1-40 ms ) like relevant for the ADC. So things may be a bit different in the more standard tests.

With a modern and thus relatively fast reading DMM/electrometer one can do the initial test also at a shorter time scale.  The faster test is less sensitive to voltmeter input bias current and is thus more practical also for small capacitors.
If one shortens the discharge time one can get data for quite some time range from a single run. The expected curve for a longer discharge time can be calculated quite easy: just take the voltage at that time as "Zero".

A consequence of the DA is that the capacitance goes up for lower frequencies, as more of those RC elements in the equivalent circuit contribute to the capacitance and not only to the losses. The loss and frequency dendence of the capacitance are linked, by the Kramers Kronig relation. The DA can be the main part of the loss in a capacitor. So capacitors with low dissipation factor are also low DA. The differences is mainly in the frequency range. D is often given for 1 MHz or 1 kHz, while the DA is more like 1 mHz range.

[TimFox]

The exception to "capacitors with low dissipation factor are also low DA" seems to be silver-mica, which has low D but highish DA.

[mawyatt]

Doesn't Keithley use ceramic capacitors in their latest DMMs?

Best,

[Kleinstein]

Essentially all the newer MS- ADCs in the modern DMMs (e.g. from K2000 or 34401/3458 on) use ceramic (likely C0G) capacitors. The ones I use are about 1/5 the DA of PP caps and AFAIK were some 10 cents each. So nothing expensive, one just needs to know which ones are good and which ones are not (about the same as PP and should still be good enough for the HP/KS meters). The HP3458 has a PTFE cap for the S&H stage, but not sure of they really need it there - the coax form may have additional effects and could be more than just a capacitance.

I don't know how good the higher capacitance G0Gs are - they may still be an alternative with 100 nF available in relatively small size (e.g. 1206 and small THT).

[mawyatt]

Essentially all the newer MS- ADCs in the modern DMMs (e.g. from K2000 or 34401/3458 on) use ceramic (likely C0G) capacitors. The ones I use are about 1/5 the DA of PP caps and AFAIK were some 10 cents each. So nothing expensive, one just needs to know which ones are good and which ones are not (about the same as PP and should still be good enough for the HP/KS meters). The HP3458 has a PTFE cap for the S&H stage, but not sure of they really need it there - the coax form may have additional effects and could be more than just a capacitance.

I don't know how good the higher capacitance G0Gs are - they may still be an alternative with 100 nF available in relatively small size (e.g. 1206 and small THT).

Before retiring a few years ago, we used mostly TDK and AVX ceramic chip capacitors, one of our concerns were stress induced effects (vibration sensitivity) as well as very low ESR and ESL. These brands were "discovered" way back in 80~90s, and they seemed to still be the best overall for our use much later. Samsung may be another brand with good overall characteristics?

Some of our work involved telemetry from an active shell in flight which induced well over 50,000Gs when fired

Best,

[Kleinstein]

Measuring the step response is a good idea. It should get a little better (less noise from the DMM / source) when looking at a step down from some 10 or 20 V back to zero.

[KT88]

It would be interesting to see how a vacuum capacitor behaves...

[TimFox]

Is the noise from your large Teflon capacitors due to their physical size (presumably pickup from the environment)?

[Kleinstein]

I did a quick (did not take the 4 hous discharge to serious - it probably was) DA test on a relatively cheap epoxy dipped 1 µF PP caps from Panasonic. It turned out a bit higher in DA with around 0.07% (5 sec. to 3min like russian standard). So not the best choice if low DA is needed.

The higher noise with the large PTFE cap is likely due to hum pick-up and maybe other capacitive coupling.

The noise seen for the PP cap is already at about the level expected for LM399 references.

Settling to 1 ppm after about 15 seconds is still not that bad. With the slope test the excitation would be a bit less than a full jump.

[SilverSolder]

I never realized that blue electrical tape outperforms the black version - good to know it is worth a ppm or two! 

[TimFox]

In high-impedance circuits, I used to use the polyethylene tape that is sold for sealing the caps on reagant chemical bottles.
Unfortunately, Cole-Parmer only stocks this in expensive case lots:  https://www.coleparmer.com/p/polyethylene-tape/4130

[Kleinstein]

Lowerring the impedance makes some sense, as long as the capacitors don't get extraordinary large. However I see little effect of the DA if the cross over frequency is kept constant. The effect of DA gets less important (faster settling) if the cross over frequency goes up. It depends on the update frequency what cross over frequency is actually needed. The initial plan 1 Hz were quite slow. the 2.8 Hz from the next step look more reasonable and one may be able to go a little higher, like 10 Hz. However I am afraid much higher may not give enough attenuation of the DAC steps for very slow ramps.

The ideal filter frequency may depend on the ramp speed, as faster ramps would need faster settling, but less critical filtering, while slow ramps would also need more filtering, but have plenty time for settling.  So one could consider a relay to swich 2 speeds.
Another option would be to generate the very small ramps with a divider after the filter.

I may be worth looking at the filter like used in the Fluke and Datron PWM based DACs in the calibrators. They can get higher order (e.g. 5 th order) fitlers with only 1 DC critical OP and they may only need 2 critical capacitors. The other capacitors in these filter type would not see much DC voltage and could be less critial with DA. The other OPs in the circuit are less critical and could be normal +-15 V supply.

[KT88]

Bob pease once published a compensation circuit for DA. Unfortunately I wasn't able to find it anymore - maby someone recalls that RAP stuff anyhow....

[Echo88]

Couldnt find the exact RAP-document, but heres the circuit mentioned by KT88:
https://electronics.stackexchange.com/questions/289312/how-do-i-find-capacitors-for-low-dielectric-absorption

[MegaVolt]

Bob pease once published a compensation circuit for DA. Unfortunately I wasn't able to find it anymore - maby someone recalls that RAP stuff anyhow....

[KT88]

But we are in to digital era, wasn't it?

Yeah, digital capacitors without DA - what is the DigiKey order-No? 

[SilverSolder]

But we are in to digital era, wasn't it?

Yeah, digital capacitors without DA - what is the DigiKey order-No? 

You could run a model of the RAP compensation circuit in software, instead of realizing it as a physical circuit?   Yay, love digital! 

[SilverSolder]

Hopefully the sun will shine again one day, but right now the skies are very dark. 

Three steps forward and two back,  is the history of the world.

[Vgkid]

Looking forward to seeing this in use.

[SilverSolder]

Vgkid, device will be completed in next month... _{If world not be destroyed in this period...}

Haha better work fast, the world is powered by misdirected testosterone at the moment!

[MacIntoshCZ]

Sanctions makes russia stronger  ...
Good luck

[macaba]

Watching with interest

Couple of things:
1. If you have a chart showing step response of all the capacitors you tested, zoomed into the first few seconds, I would like to see it. It is an interesting compromise between the physical size (Teflon > polypropylene > C0G) and DA.
2. I prototyped an integrating electrometer (like this project but in reverse!) with LMC662 and about 10pF of parallel AVX MLO capacitors. There is a PDF that shows these have extremely low DA, and I could see this on the output of the electrometer circuit - after a step change of 2V, it settled to less than near-zero fA of leakage in 10-20 minutes, much faster than the C0G caps I tested. Might be an interesting capacitor to try for the main current-generating capacitor.

[macaba]

I'm planned use 0.01pF to 10pF capacitors with air or nitrogen dielectric, to generate current. They must be extremely low DA.
About 100pF and 1000pF i'm not sure... maybe its be K71-7 or K72P-6 or some else...

It is understandable that you want to use air or nitrogen, I am looking at the same thing myself, maybe using 2 PCBs as parallel plates capacitor (with guarding, and low DA Rogers 4350B PCB laminate) but the uncertainty due to thermal/moisture expansion probably makes that a bad idea.
It still might be worth trying AVX MLO in your prototype as it is cheap & the DA seems extremely low (not as low as air/nitrogen... but no manufacturing required!). There is also possibly a volume advantage; small 0603 is not going to capture as many cosmic rays!

What is your air/nitrogen capacitor plan?

[macaba]

Near-zero because I don't have the ability to measure it any better! 

[TimFox]

Are you concerned about noise injected into the power supply from the neon bulb?

[TimFox]

Of course, a neon bulb run on 120 V or 240 V rms needs a large series resistor, but the current waveform is "rough", certainly when compared with an incandescent bulb or LED, due to the interesting behavior of the glow discharge.

[mawyatt]

Gas ionization is a good source of broadband noise, even with a large series resistor the distributed capacitance will charge up to the ionization voltage and very quickly discharge into the tube. This will produce a very quick and sharp current peak/spike rich in spectral content.

Edit: Instead of a neon bulb why not use an LED (maybe multiple LEDs) on one, or more of the DC outputs, just use the series bleed resistor(s) already in place.

Why such large capacitance on the transformer secondaries? This isn't necessary and even increases power losses and these large capacitors have a large effective inductance which hinders their ability to provide a low impedance shunt at higher frequencies where they help reduce noise feedthru.

BTW, those custom toroidal transformers look nice!!

Best,

[mawyatt]

And what test result would that be??

Best,

[mawyatt]

What type of "noise performance" tests did you conduct?

I'm curious since I can't recall OEMs using such large capacitance values directly on the secondary of mains transformers which are used for linear type operation (read not SMPS), at least I haven't seen this on the equipment used here in US and may have missed something!!

Best,

[Kleinstein]

The capacirtance on the secondary can serve 3 purposes:
1) They limit the speed on how fast the voltage changes and this way limit / avoid reverse recovery spikes at the rectifiers. This may need quite some capacitance if the transformer has tight coupling / low series impedance.
2) Together with the transformers parasitic series inductances they act as higher frequency fitlering for EMI from the grid.
3) The capacitors can provode some of the magnetizing current for the transformer. This can lower the no load current and actually redue the losses a little - though not much as the resistive loss is only a small part of the no load loss. Transformed back to the primary side the 1 µF at a 25 V winding is only a little more than 10 nF.  The question is more why use so large capacitors for C1 and C2 - here I would consider less (e.g. 100 nF), so that the circuit is safe even when the bleeder resistor fails and the link to the transformer is lost.

The capacitors on the secondary are not standard, as they add costs. This is more like volt nut terotory where low costs don't matter that much.
One finds such capacitors sometimes in audio equipment, not just audio-foolery.

[mawyatt]

I've never seen a 4.7uF directly on the secondary of a transformer for linear type use, usually much smaller like 100~470nF. As OP just stated added series resistance per Horowitz does make some sense as this gives a known impedance for the shunt cap to work into when acting as a HF bypass shunt, and also provides a startup surge impedance for the transformer, rectifiers and filter caps and must be sized for such.

Best,

[macaba]

Yes I'm understand, but 10-20m to "near-zero fA" may increased to hours with "near-zero aA".
Anyway... when my smugglers buy some sanctioned toys for me, i try get a couple MLO caps for testing.

I attempted to measure it as best I could by using the integrator circuit itself. After 0.6V step change, it is 1fA after 5 minutes, 100aA after 1 hour, 10aA after 3 hours.

More detail here

[Kleinstein]

The residual settling after the step still looks quite significant. 100 aA is not a large current, but the charge transferred over 1 hous is still not that small compared to the initial jump.  It is still impressive, but a low DA capacitor should be way better than that. So chances are the observed settling is more something other than the the capacitor, e.g. the PCB, case of the LMC662.  Epoxy and the usual FR4 PCB material is more like poor DA performance.

[macaba]

chances are the observed settling is more something other than the the capacitor, e.g. the PCB, case of the LMC662.

I agree. I used air wiring on the input pin & consideration of cleanliness during construction, so this could be package related (and probably not contamination or FR4).

[Echo88]

Were there ever leakage measurements done on MEMS switches (ADGM1004 for example), just to get a rough estimate in reality compared to the usually conservative leakage spec?
Are there any other MEMS switches that are capable of DC-range with interesting specs compared to normal analog switches?

https://www.analog.com/media/en/technical-documentation/data-sheets/adgm1004.pdf

[TimFox]

I couldn't read the dial:  what temperature did you set the oven to?

[TimFox]

An amusing coincidence for us in the Fahrenheit domain is that Sn50-Pb50 alloy melts at 212^o C.

[tggzzz]

Were there ever leakage measurements done on MEMS switches (ADGM1004 for example), just to get a rough estimate in reality compared to the usually conservative leakage spec?
Are there any other MEMS switches that are capable of DC-range with interesting specs compared to normal analog switches?

https://www.analog.com/media/en/technical-documentation/data-sheets/adgm1004.pdf

https://menlomicro.com/products/signal-relays

0.5mm pitch BGA package, claims 25pA at 150V, 90fF, both typical, note hot switching and floating limitations.

[HighVoltage]

You find the most interesting stuff
What was the purpose for these vacuum capacitors, if you know?
At +/- 5% they are most likely not for reference.

[bsw_m]

10pF and 100pF air capacitors with sapphire insulators that used in differentiator of a EK1-6 and NK4-1 low current calibrators.
A 10pF capacitor is used for ranges down to 1E-13A.
A 1pF and 0.1pF capacitors are used for ranges down to 1E-17A (sorry no photo).

[Kleinstein]

The dielectric absorption makes the capacitance to go down when the frequency goes up.
The measurement at 50Hz is somewhat faster than the use with the ramp.

So part of the difference can be due to the DA of the PS cap. The usual tabulated amount of DA for PS is at some 0.05%, which may be a coincidence because of the different time scales and not all PS caps are the same.

[Kleinstein]

0.05 % accuracy for 1 nA looks already quite good. This is especialy knowing that there a known effect to cause a bias in that direction.

[guenthert]

     I noticed that the Russian word for voltage has been translated into 'tension'.  I've seen the same in other documents (google-)translated from Chinese.  I found that amusing, as the German word for both tension and voltage is 'Spannung'.

[TimFox]

"Tension" is sometimes used for voltage in English, especially in the phrase "high tension".

[Atomillo]

In Spanish too. "Alta tensión" (high tension) is a lot more common than "Alto voltaje" (high voltage).
In all other cases they are used pretty much interchangeably.

[RoGeorge]

In Romanian, too, it's tension.  Usual (and academic, too) is to say, for example for AC voltage, "tensiune alternativă", as opposed to "voltaj alternativ".  "Voltaj alternativ" just sounds wrong, though the word "voltaj" is listed in our dictionary as a neologism, imported from the French "voltage".

[MegaVolt]

     I noticed that the Russian word for voltage has been translated into 'tension'.  I've seen the same in other documents (google-)translated from Chinese.  I found that amusing, as the German word for both tension and voltage is 'Spannung'.

Yes. The translation into Russian of these two words sounds the same.

[2N3055]

In Romanian, too, it's tension.  Usual (and academic, too) is to say, for example for AC voltage, "tensiune alternativă", as opposed to "voltaj alternativ".  "Voltaj alternativ" just sounds wrong, though the word "voltaj" is listed in our dictionary as a neologism, imported from the French "voltage".

Same in Croatian.  Official name is "napon" which is tension. There is a colloquial word "voltaža", a voltage, but that is used very rarely and is usually a sign you don't really know much about electricity .... :-)

[TimFox]

In elementary physics class (English-speaking), one of the tricky bits for new students to understand is "tension", for example the force in a string in an "Atwood machine" (two unequal weights connected by a string running on two pulleys).  That is the same meaning as in "tensile strength" of a material being stretched.  Therefore, English-speaking physicists usually will not use "tension" for a voltage.  God knew what he was doing at Babel...

[RoGeorge]

We have that, too, tension as a mechanical stretch. 
https://www.etymonline.com/word/tensile
https://www.etymonline.com/word/tension

Same word (tensiune) in Romanian can be used for electric tension, mechanical tension, suspense (e.g. in a movie scene), sometimes the word is used to denote a stressful situation, psychological pressure, blood pressure (popular way of saying "high blood pressure", aka hypertension, in Ro is to say somebody "have tension"), etc, while saying voltage (instead of tension) usually indicates the speaker doesn't know much about electricity, just like in Croatian.

[SilverSolder]

Scandinavian languages also call it "tension"...   seems popular!

The concept of "electrical tension" does kind of make sense in English too, but it just isn't customary to call it that.

[TimFox]

In careful American physics usage, one says "electric potential", sometimes the antique term "EMF",  not to be confused with "potential energy".
In English, there are too many meanings for "tension", leading to ambiguity.  https://www.yourdictionary.com/tension

[SilverSolder]

"Electric tension" seems to be fair game, though?  It is unambiguous and seems to be actually used, if you google it!

[TimFox]

"Electric tension" seems to be fair game, though?  It is unambiguous and seems to be actually used, if you google it!

Depending on the reference source, I find that in US usage, it is only common in the phrases "low tension", "high tension", etc., not for "normal voltages".
Tension gives me headaches...

[guenthert]

What have I done?     I didn't mean to derail this thread.  So sorry.   

[Kleinstein]

The capacitors look a lot like polystyrol.  Polystyrol is a really good insulator: I have an old analog meter with the front "glass" made from polystyrol and by some rubbing of dirt from the front I got charge traped on the inside that was strong enough to nearly trap the needle. The charge stayed there nearly unchanged for some 2 years, until removed with water to fix the meter.

A nice point is that they are polystyrol all the way to the wire, so no extra sealant or potting needed.

Just keep in mind these caps are sensitive to solvents (including alcohol).

[MrYakimovYA]

I'm still searching different types of large capacitor in range 10pF-5000pF

What about sovietn green (primarily green) and big K71-7 capacitors? As far as I know their insulator is also polystyrol.

[Atomillo]

Yesterday I came across this article:

Frequency dependence of gas-dielectric capacitors used in sub-nA reference current generators
By S. P. Giblin; G. D. Willenberg; N. E. Fletcher

https://ieeexplore.ieee.org/document/5543771

Which highlights the importance of low frequency testing, even in air dielectric capacitors like your standards.

[HighVoltage]

Nice, looking forward to your video

[RoGeorge]

first try to make realtime CPLD software

If it's first time using a HDL (Hardware Description Language) like VHDL or Verilog, keep in mind that that's not for writing software/programs.  You do not write real time software for a CPLD.  It's hardware, just like you would design with flip-flops and logic gates.  The fact that HDL is text, and that it looks like a program, is very misleading.  That's not software, it's a schematic.  It's hardware, just that in HDL one describes the hardware with HDL text, instead of drawing a schematic on a paper.  HDL is text, but it represents a schematic, not a program.

[MrYakimovYA]

Soldering complete... ready to program...

Wow! It looks like a highly precision metrology equipment! Nice design! I'm looking forward to first run!

How do you verify its precision? Is it based only on theory? Or do you have any ability to measure the current accuracy?

[Zoli]

measure the current accuracy?

It's complicate.... as any SI units which converts from others.
GLIN(ГЛИН - Генератор Линейно Изменяющегося Напряжения also called ultra-low speed ramp-generator) is responsible for part "dU/dt" into formula "I=C*dU/dt". Other important things is accuracy of "C".... capacitance accuracy on the infra-low frequency it is a cutting edge of newest metrology science... many powerfull metrology laboratories investigate that. You can see publications of the TÜBİTAK UME, Physikalisch-Technische Bundesanstalt, VTT MIKES Finland, and others metrology centers.

Accuracy of extremely high currents  (like as 1nA and more) i can compare between "I=C*dU/dt" and "I=U/R" methods trough resistance standards and Keysight B2985A electrometer, as shown earlier.
Unfortunately, common "old school Soviet electrometers" can't do high precision comparison of currents which generated through different methods. But they can be very sensitive on ultra-low current ranges like as several attoamperes, it helps when need to do a component testing, material's selection and etc. (V7-45 ADEV noise floor is around 10^-19A which is 100 zepto-ampere)

Hmmm...
I think "low" and "below" would be a much better fit.
/nitpicking

[Atomillo]

In order to see secondary electron generation the vacuum would indeed need a turbo, but the system would probably also need baking and very careful consideration of leakages, both real and virtual.

I have no practical experience with vacuum but might be the vacuum part of fusor.net (amateur community of nuclear fusion researchers) might be of interest.

[HighVoltage]

Sorry no.
1nA it is around 6 241 509 074 electrons per second. It is a extremely big value for true Ninja... equal to population of the planet earth, each second....
True electrometers can be sensitive to 6...7 electrons per second(1aA), it is low.

Even the Keysight 34470A DMM can measure 1 nA very well and stable, no need for any special instrument in the nA range.

With my Keithley 6517B Electrometer, I was able to get to the atto rage and was able to measure 100 aA kind of stable. (with lots of shielding)

[MrYakimovYA]

Micron-GLIN v.2 PCB repository: https://github.com/shodanx/Micron-GLIN-v.2
Micron-GLIN v.2 schematic book: https://ampnuts.com/wp-content/uploads/2022/06/Micron-GLIN-v.2-schematic-book.pdf

Hi!
Could you tell me what are you using as PE (Protection Earth) in your wall socket? Has your home PE conductor? As far as I know only modern russian home buildings have an earth protection system... Mine doesn't have it(((

[macaba]

If you're enjoying it, then don't worry about it, I'll read your updates for the next 4 years!

Your modular approach means that you can get v1 working quickly and if you've still got motivation, carry on with upgrades.

Do you know which low INL ADC you might use?

[HighVoltage]

You have very interesting projects
Thanks for sharing

[ch_scr]

Amp-porn movie: testing user interface and main functions of new GLIN with 1nF capacitor.
https://ampnuts.com/micron-glin-v2-porn/

Seems like your effort paid out and this one works even better than the last one, good job! Would you share a picture of how the capacitor was hooked up?

[ch_scr]

I use this simple tool. One jack is BNC for connect to GLIN, other is Molex 3-lug triax, connected via 1.5m cable to electrometer (Keysight 16494A-001 cable).

Wow, and for ground return current, just screws / washers biting through aluminium oxide layer are enough? 
That's... mostly an accidental path? Is there actually a DC connection between the two outer connector shells?
Or will return current take long way around device grounding loop? I'm surprised it's this good like this?! 

[ch_scr]

just screws / washers biting through aluminium oxide layer are enough? 

You means resistance of Aluminum oxide layer?
Wall end of aluminum box parts is vanished from paint to provide good connections between outer shield of a triax and shield of GLIN cable.
Aluminum oxide layer is may provide problems, but magic is - you can put some KOhm  resistor between signal ground of a GLIN and ground of a electrometer without any problem. You may add any offset between grounds... Electrometer can contains any input bias.... Ramp through capacitor is eliminate any offset to negligible.

Otherwise - if you generate current with HiMeg resistors, electrometer input bias is important error source.

Thank you, that answers my question.

[branadic]

What's the purpose of magic voodoo slots around OCXO? Is it just for the fancy look of it? 

-branadic-

[SilverSolder]

I guess it is possible to be a "nut" at high frequencies too!  e.g. impossibly accurate frequencies and phases etc.

[Kleinstein]

The frequency part does not really need supper high accuracy.  Even the accuracy of a normal crystal oscillator would be good enough and even an OCXO or similar would be overkill. The accuracy of the capacitors is not even close and in many cases the  I = C du/dt part is used relative to compare currents of different scales (e.g. 1 fA to 100 nA) and not as an absolute from a given capacitance, voltage and frequency.

For the satelite stabilzed clock there are plenty of solutions / plans available. No real need to build everything from scratch here. For the antenna the position (outside with unobstarcted view) may be more important than the actual antenna part.

With the relatively low orbits of the satelites there is not much optimization for regions, as the earth is moving and the orbits are not in sync. It is only a bit the question if they want good coverage also for the polar regions.  GPS currently may have some additional errors over russia, but even this should not effect the frequency part that much, more like some added noise, but not a frequency bias.

[Ole]

I just had a look around ebay and found some reasonably priced triaxial cables.
While I do not know the exact quality of these cables I thought that it might find some interest here.
( https://www.ebay.com/itm/255603797766 )

[Hydron]

I just had a look around ebay and found some reasonably priced triaxial cables.
While I do not know the exact quality of these cables I thought that it might find some interest here.
( https://www.ebay.com/itm/255603797766 )

These are twinax, not triax, despite the auction title. Would stay far away!

[Ole]

Thanks for the warning

[MrYakimovYA]

Dear shodan@micron, you have nice project!

Is USB interface fine to control equipment? GPIB is old, maybe LXI?

Does USB allow to build up of automated system that includes another equipment (DMMs, powers supplys and so on)? I mean almost any equimpent has GPIB/LXI and you GLIN has USB))

[MrYakimovYA]

Ok, USB is good) I'm also developing my highly precision and highly unabtanium device (picoammeter/voltage source). And now I'm thinking about connection interface. Ethernet based on W5500 (Aliexpress have low prices again) looks very very very attractive.
It has:
1. Built-in galvanic isolation.
2. Standard connectors and network structure.
3. Allows to have console (one socket) and up to several data stream (the other sockets).
4. May be the LXI can be added (I don't know about license).
5. W5500 (Wizner) give you hardare TCP/IP stack.

Don't you thing about such solution?)

[Kleinstein]

My completely illegal's, sanctioned refs will be delivered soon... So... it's time to start thinking about new Low-Noise, Mirror-free reference concept...
Earlier version use R+R mirror for making -7V reference for DAC, but i want try new mirror-free concept to do not amplified noise and instability.

Using separate references for the positive and negative side is a 2 sided thing. It avoids the inverter, but it also means that at zero output there is noise from both references and not the usually lower noise of the inverter.  For most uses I would consider the inverter the smaller evil.  One could still build the reference in a way to relatively easy switch between the 2 separate references and a more inverter like version:  the inverter like version would have the point between the 2 references not connected to AGND, but driven by an OP-amp to force the average of the 2 outputs to be at AGND. Functionally this is essentially the same as the inverter version, just a little more symetric.

With the extra buffers for the references, one could consider some easy to add RC fitlering for the references.

The reference amplification to provide the reference current should not use AZ OP-amps like the OPA189. There is no real need for them and they can do more harm than good from EMI and possible beating. So I would prefer something like OP07 or a more modern (and lower power) OPA202. In theroy even an LM358/RC4552 could do.

[Kleinstein]

The DAC kind of uses resistors to set the ouput. So one has that kind of resistor related drift anyway.  For the resistors it is not just tempco but there can be other effects like humitidy in the case and mechanical stress through the PCB. The humidity part can lead to rather long time constants to stabilitze.

It is a good question if the resistors are more stable than the references. At least for the noise part the resistors are usually better, but with ovenized references this is not so sure for the temperature drift.
Another possible alternative would be a charge pump to do the inversion, though this also comes with some problems in the form of EMI and switching transients.

[Kleinstein]

U3 and U8 seem to be overkill, but there is no need to remove them alltogether. One can go one step down and use simper OPs for them instead. Even an LM358 should be somewhat more stable than the LM317.
If overkill comes at the price of 2 relatively simple OP amps (LM358, op07,...) this is still moderate.
Depending on the layout there could be a little more additional resistance than the 25 mohms.

I don't see a real need for U6 and U7. The ground currents should compensate quite well and any residual current would be relatively stable. For thermal reasons it would make sense to have the reference close together anyway. So not very much extra resistance for the traces needed.

I would consider to have to option for one OP-amp in combination with 2 resistors to enforce symmetry.  So one could either use a direct link to ground or that one OP-amp and 2 precision resistors to set symmetry. So the same PCB could be used for either case with 1 bridge  or the optional OP-amp+ resistors populated.

It may not be needed, but for peace of mind it may be good to have simple RC fitlering between the references and U2 / U5, to isolate the reference from possible current spikes.

[Kleinstein]

Especially with the LCC8 version it would make sense to have the 2 reference chips rather close together, to minimize thermal loss. It is less important with the to46 version, but it still does not hurt.
So I don't see a large problem with thermal EMF from directly connecting the 2 references together. The thermal situation is a bit asymmetric, but ideally there should be some thermal isolation / cut-out around the references (can be both together). Much of the heat flow would be through the traces and it would be not so difficult to make the ground trace correspondingly thicker than the ones for the output sides.

For the thermal EMF the 95°C are not that relevant. It is still only the variations from the outside temperature that matter. Any constant thermal EMF part can be seen as part of the reference voltages. Thermal EMF from a fixed temperature setting was actually one of the very early stable (and able to provide significant current) voltage sources.
I don't see a problem having the reference close (a few mm) togther, possibly nearly touching at the no 8 pins (no need to solder pins 8 and 4).

The OPA170 is not the best choice, but should still be more stable than the LM317/337. Good symmetry would help with ground current if the center is directly connected to ground.
Having both options on the PCB is probably a good idea.

What is the idea behind the diode D1. If at all I would expect a need for protection with the negative side to make sure that the heater it the most negative. The may warrant a schottkey diode from GND to -15 V to make sure the -15 V does not go much wrong polatity during start up.

[Kleinstein]

With a classical 50 Hz transformer there is little need for a common mode choke at the input. It does not hurt, but no real need. The filtering shown is more like typical for a SMPS.

I see little sense for the extra resistors in series to C1 and C2. A small series resistor (could be in the form of a fuse) in sereis with the transformer secondary would be more useful. It could slow down mains transients and reduce the current peaks to the rather large filter capacitors.
The combination of 2200 µF filter caps and tiny case LM317 regulators is a little strange.
Just in case there is a problem with the negative supply, it would help to have a reverse schottky diode at the negative supply to preven the heater negative side to go much more positive than the reference part.

C6 and C16 are with a rather small form factor. Especially small form factor X7R capacitors loose quite a bit of there capacitance under a DC  bias. So I would suggest adding a 2nd footprint in parallel to have the option to add more capacitance. The 1 µF is in my view more on the low side to start with. 2 x 1 µF should be no problem.

The heated references would not really like a large ground area. The more normal use is with a little extra thermal insulation, like cut outs.
With the SMD case the soldering would cause some stress, that may reverse some of the burn in effect. So if possible I would consider doing the burn in with the final reference PCB.
If build as a pure burn in board, it would make sense to also have a socket for the TO46 version.

[Kleinstein]

The copper area around the reference causes lower thermal resistance and thus more required heater power. With the socket for some added thermal resistance this may still work, but is not ideal.

With the currently slighty difficult situation to get chips I would not exclude a 2nd case option. I think there should be enough PCB space.

It may be a reasonable idea to have a conncetor for the power - so the part could also be used as just a +-15 V supply if not in use for burn in.

[dcarolan]

Quick question. Where does one get a triaxe shorting cap

Thanks

Don

[MiDi]

If triax shorting plugs exist, so far I haven't seen any.
You can DIY one with a TRB connector.

[MegaVolt]

Keithley sells them as a set with a connector 7078-TRX-TBC:
https://il.farnell.com/keithley/7078-trx-tbc/3-lug-female-triax-bulkhd-conn/dp/2672714
https://www.ebay.com/itm/323974010369

Only a cap Trompeter TAI 14949:
https://www.ebay.com/itm/265814239428

[alm]

That looks like a dust cap to me, which only contacts the outer shields and shields the inner conductors from both dust and external fields. Not a short that would actually short the inner conductor to the other conductors.

[Kleinstein]

There is not really an advantage from using capacitive feedback, especially not with high gain. In most aspects a more normal, non inverting, low noise amplifier with resistor for the gain settting is preferable: usually less drift. No need for a Gohms resistor and without the 12 K resistor in series there is less noise.

I would consider this more like a curiosity, but not really a useful circuit. The good performance is from the high end OP-amps, not the dominantly capacitive feedback.

[Kleinstein]

The large capacitor is only needed with BJT based OPs. With an OPA140 or similar a relatively moderate 1-10 µF can do the job. If one gives it time for settling a MKS type can do.
A not so well known trick is to have the lower frequency limit set in a separate later stage and not directly at the input. This helps to suppress the noise of the resistor to ground.

The amplifier part is a relatively small project, so why not have one extra version. I currently don't see the advantage of the version with capacitor feedback, but there may be one. The characteristics is a little different. It may be usefull in the more differentiator mode.

For the high ohms feedback resistor switching would likely be a bit tricky due to extra leakage. For just the bandwidth one could have to option to extend the frequency range to down lower with a T circuit. Switching for this would be on the lower impedance part and not critical. I don't think the resistor noise would be that critical, as it is more on the ouput side.

[bobAk]

You are trying to prove your approach to building an amplifier using our words and wishes, under your desires and restrictions that you have imposed on yourself.  You live in Russia and it is easy for you to get first-class parts that will easily fit the amplifier into a 100x100x40 mm box with batteries and a weight of 150-250 grams.

[bobAk]

You do not understand that this is a description of the size and opportunities? Your experience?  Okay, I agree.  But then again, why was it necessary to build a deliberately meaningless justification misleading less experienced people?) think should be more professional.

[Kleinstein]

The resistor noise from the input AC coupling mainly effects the frequency range at lower end and below the transition. With higher frequencies the capacitor and source impedance reduce the noise. Similar the 1 GOhms resistor in the capacitive feedback version does contribute in the same low frequency range. The 1 Gohms resistor noise is relative to the output and thus less relevant.

One can get around the resistor noise with an even lower cross over frequency for the input and thus larger resistor and than setting the lower frequency limit in a later stage (could be software).
An alternative viewpoint is having the resistor as a source of current noise that than sees the input capacitor as "load" impedance. A higher resistor gives less noise.
Trying to go with a relative low resistance and large capacitor is more causing problems when connecting to the DUT and the lower impedance loads down the source.
The inverting configuration also has more loading of the DUT. The extra series resistor to set the upper frequency limit also contributes to the noise - this time for the full band and not just at the lower transition range.


It absolutely makes sense to have battery power for a low noise amplifier, but for some uses mains powered operation also helps.

[Kleinstein]

The somewhat jagged pattern of the noise in the time domain could be due to DNL errors of the DAC. It looks like there is repetitive part with some 14 seconds period, that can be seen as some 16 steps. For comparison it may help to know at which time the DAC values are changed. 

I don't see much effect of interference or an indication of a firmware problem. It is more a lack if low pass filtering and a limitation of the DAC.

[Kleinstein]

It is well possible that this are the DNL limitations:
The DAC11001 data-sheet gives a typical (with respect to the codes used) DNL error of some 0.1 LSB or 0.1 ppm (with 20 bit resolution). With a 10 V full scale range this would be around 1 µV.
So the observed steps are about what is to be expected.

The interface of the high resolution DACs is seriell and the digital interference would thus be faster. In addition the main part with digital interference would be capacitive coupling and thus more very fast spikes well beyound the BW of the test. If not carefulll with the software, the INL correction could result in some steps getting twice as large, using 2 LSB jumps. The current observed steps however are smaller, more like 0.1 or 0.15 LSB.

Some lag / rounding is expected from some analog low pass filtering after the DAC. Not sure how much filtering is present in the test.

A ramp of 1 mV/s is relatively slow. With a 10 V full scale range this would allow a 3 hours ramp and this a quite long averaging time. If one does not need the full 3 hours time window, one could divide down the voltage signal and run the DAC with a faster ramp. For testing the slow ramp is of cause a good idea.

The instruments to measure pA range currents are usually not very fast and would average over much of the modulation on top. So the jagged curve is not a real problem, more a sign that the amplifier to test it is good.

[Kleinstein]

Doing the correction via the DDS part is a good idea, as it should give plenty of resolution. The DAC seems to be quite good and quantization there already an issue.

It looks like one can measure the DNL steps. So in principle one could use a correction also for the DNL steps. So a correction value for the 20 bits, or at least most of them - the higher bits are the more critical ones not so much the lowest ones, though they are more visible. There are still limits to this method as the bit values may not be all that stable. Especially the MSB can be drift limited.

[3roomlab]

The somewhat jagged pattern of the noise in the time domain could be due to DNL errors of the DAC. It looks like there is repetitive part with some 14 seconds period, that can be seen as some 16 steps. For comparison it may help to know at which time the DAC values are changed. 

I don't see much effect of interference or an indication of a firmware problem. It is more a lack if low pass filtering and a limitation of the DAC.

is that the similar error as fig2 in this link?
https://www.analog.com/en/technical-articles/16bit-parallel-dac-has-1lsb-linearity-ultralow-glitch-and-accurate-4quadrant-resistors.html

[Kleinstein]

The curve measured over time corresponds to the DNL over code from the link.  Fron the curve it looks like 16 (maybe 32 with a very good LSB) code steps every 14 seconds. The part in between is just noise and there is also some extra noise from the amplifier.

For the correction one could use the same measurement as before: just correlate the steps with the bits changing. So it is only some 20 constants for the correction, not some 15000 points for the center part.
It still makes sense to measure over a certain range to get some averaging also for the higher bits.  For the start one could likely use only the lowest 2 to 4 bits and see how much the "noise" improves.

[Kleinstein]

The popcorn noise of the LM399 often mainly effects really low frequencies and may not show up very much in the frequency range of the preamplifier. The curve beyound some 20 seconds is to a large part effected by the frequency response of the amplifier. So the curves going down beyond some 10 s is no from the references.

For the ramp, there is a part from the DAC and amplifiers and a part from the reference. The ref. part depends on the voltage and not just the ramp speed.

[Kleinstein]

I would somewhat doubt, that the BW really goes down to 0.02 Hz. At least the curve for just the amplifier starts to go down from 20 s on. With a normal circuit the expected shape ot the Allan deviation curve is to finally go up on the long run. There is essentially no reasonable way (other than the frequency response) to let the curve go down to very long times (corresponds to low frequencies). White noise would give a horizontal curve and usually the noise is more going up to low frequencies, but not down. Getting the lower frequency limit from a resistor in the 10s of Gohm range is somewhat prone to errors (e.g. from leakage).