Fun With Low Leakage/Bias Current: Femtompere, Electrometer, Keithley 617

[_Wim_]

Time for settling after change of the the range from current to ohms and back takes the same time (not measured) as with old relays, as nothing in rest of the device was optimized for this function. Maybe that in my instrument it will be a little different, as I changed the type of end stage transistors.

When the input amplifier is saturated, especial in the ohms range, the max output voltage will appear on it's output if terminals are open. It takes a while until it gets back to normal working conditions. This behavior also described in the manual of the device.

Guido

Thanks for your reply. The manual indeed talks about a "number" of minutes, but I was wondering if everybody experiences such long settling times as with my unit.

[zrq]

I'm going to receive a 617 from ebay and I'm planning for the ADA4530-1 mod. From the posts by Midi, it seems a 10 nF capacitor is needed on the feedback to stabilize the loop when the input FET and LT1012 are replaced. I'm kind of worried if this capacitor leak, it can directly contribute to the input bias current. Does anyone did the mod have recommendations on this capacitor? Or is it better to add another buffer stage?
Besides, I have access to a 4200-SCS with the remote amplifiers at work, so maybe I can do some "serious" characterization of the leakages of the COTO 9002-05-10 and Standex HI05-1A66 later when I got time.

[Kleinstein]

I have not found the exact ADA4530 replecement ciruit. But it looks like the compensation capacitor is at the other input. So leakage there is not critical and relativel to a resistor on the order of 100 K.
It would still not hurt to have a PP or alternatively NP0/C0G  (is available as SMD) capacitor.

[r6502]

Hallo all,

as "Kleinstein" concidered, leakage is not critical, as this input is not connected to the electrometers main input, where leakage is critical.

Here you can see my modification, including the Coto relays:
https://www.eevblog.com/forum/testgear/fun-with-low-leakagebias-current-femtompere-electrometer-keithley-617/msg4391443/#msg4391443

here you can find the schematic by MiDi:
https://www.eevblog.com/forum/testgear/fun-with-low-leakagebias-current-femtompere-electrometer-keithley-617/msg2501178/#msg2501178

Guido

[zrq]

Thanks for the information!

I got time to play with my 617. It seems the input offset current can never be zeroed, drifting from -1fA to -100fA in a few hours, from post by fellow members, it seems that's usually due to relay leakage and/or broken input FET. I may have a stupid question, but from the schematic, I cannot figure out where is leaking into the feedback path if the relays are bad. My naive thought is that's not very different from having a lower valued feedback resistor, which will only reduce the reading but not give a offset. And the guard is supposed to drive the shield at the same potential (+-1mV), so leakage from there also will not contribute to offset. What did I missed here?

Another question, after modding with ADA4530-1, is it still necessary to install the bias current compensation R332, R348? From the datasheet, the input offset of ADA4530-1 at room temperature is sub-fA, maybe we can leave that out to reduce the noise (thermal from 250G R332) by a bit? Also Q311 can be removed if we trust the internal protection in the opamp?

[MiDi]

I may have a stupid question, but from the schematic, I cannot figure out where is leaking into the feedback path if the relays are bad. My naive thought is that's not very different from having a lower valued feedback resistor, which will only reduce the reading but not give a offset. And the guard is supposed to drive the shield at the same potential (+-1mV), so leakage from there also will not contribute to offset. What did I missed here?

That is definitly not a stupid question!
Leakage between the contacts would only effect the gain, but not the offset.
Leakage into the input gives an offset, e.g. 1mV with 1TΩ guard (ES/preamp out) to input gives 1fA.
If both leakages would be constant over time & temperature, they could be compensated/calibrated.

Crunching my numbers, the critical relays combined (K307, 9, 10, 12) had at least 2TΩ from ES to input (Relay resistance) and the offset current was ~50fA (from here).
That would need ~100mV between guard (ES/preamp out) and input, this does not compute with expected 5mV (50fA through 100GΩ) .

Another question, after modding with ADA4530-1, is it still necessary to install the bias current compensation R332, R348? From the datasheet, the input offset of ADA4530-1 at room temperature is sub-fA, maybe we can leave that out to reduce the noise (thermal from 250G R332) by a bit? Also Q311 can be removed if we trust the internal protection in the opamp?

It would be counterproductive to have that bias compensation with ADA4530 (it includes bias TC compensation).
ADA4530 has internal protection, so Q311 is not needed anymore.

[zrq]

Just finished swapping the relays to COTO 9002-05-10 and replacing the front end with ADA4530-1. Cleaned the sensitive part with ULSI grade IPA (overkill) and sonicated the air-wired opamp. I used the relay variant with magnetic shield, the shield is a exposed metal cover which turned out to be a annoyance when dead-bug soldering, some electrician tape had to be inserted to insulate it from the board. Applying epoxy to fix the relays on to the board is very helpful reducing the sensitivity to vibrations.
The performance looks really impressive after warm-up, with the bias current compensation circuit disabled, input connector covered by aluminum foil, <0.5fA bias current can be achieved. Did some simple noise characterization by simply reading from GPIB at raw speed (3.13 Hz). The current noise is still 1.75x above the thermal noise floor for the 100G feedback resistor, through. Interestingly the excess noise does not appear as 1/f but a white noise, wondering what can be the source. There is also a popcorn noise, which I don't think originated from ionizing radiation.
Another feature is it requires a really long warm-up to reach the 0.000pA reading. 2 hour is barely enough. I put a thermal couple on top of the preamp shield and saw the temperature raise from 22.4 degC up to 38.7 degC, the timescale of the temperature stabilizing roughly agrees with the time when the reading settled to zero. Maybe it's thermal EMF? The popcorn noise also seems to have a strong correlation with temperature.
Edit: 5fA at 100GOhm is 500uV, that's too high for typical thermal EMF, so there likely to be something else.

[zrq]

Took some thermal images. Apparently the JFET Q302 for the ohm range current source is the biggest heat source inside the shield. Inserted some plastic foam between the opamp and the JFET in attempt to isolate the heat a bit, and the noise did get better! The pop-corn noise that goes beyond 2 fApp appears less frequently, especially before the instrument fully warms up. (please ignore, this effect disappeared after longer 16000s acquisition, it maybe just a coincidence). So the next thing to try is move the JFET out of the shield... Any suggestions on better thermal management?

[Kleinstein]

The whole part around U304 / Q302 and VR301 (6.3 zener ref) is not that sensitive and thus does not absolutely need to be insider the enclosure. It looks like a reference of some 10 V ( could be more ?) or so.
It is not only the JFET to produce heat, but also the OP-amp (a 741 is not really low power).  The supply is also not well regulated and thus possibly a change in the heat with mains variations.

A lower temperature for the critical input amplifier could help. However on the other side it also helps to have a low relative humidity and for this a slightly elevated temperature (e.g. 10 K above ambient) helps.

[zrq]

Now I have a new suspect of the origin of the popcorn noise, is it possible it's the input glitch caused by the ranging amplifier, chopper ICL7650? Although the 100 Ohm output impedance of the preamp is not favoring this, maybe it triggered some ringing?

[Kleinstein]

I would not expect the ICL7650 to contribute much to the noise. As an AZ amplifier it has essentially no extra 1/f  or similar noise.
The ADC may have relatively poor suppression of supply noise from the + 5 V supply. The 5 V supply to U126 is used as reference to the ADC itself and the result is only relative to the main zener from an extra measurement of a 2.5 V signal dirived from the main reference. Excess ripple on the 5 V (e.g. aging capacitor) can thus cause quite some extra noise that somethimes comes up when mains is low.


One should see if the noise is from the input amplifier part or the ranging amplifier and ADC by looking at the noise at different gains.
For the 200 V range, the gain at the ranging amplifier is low and the noise of the input amplifier should be no longer relevant.

[zrq]

I'm thinking about the spiking similar to Art of Electronics Third Edition Figure 5.50, maybe in rare cases, the switching of amplifiers can randomly get a bigger mismatch?

The reason that make me suspect this is, when using my NI DAQ USB6363 to sample from the the Analog Out, I do see spikes happening at ranges other than 2pA, like 2nA,  200nA, 2uA or even in zero check  . At these ranges, the glitches are much faster (tens of us) and happens at the same rate (one per ~200s), so will be ignored by the built in ADC. The screenshots are captured at 2nA range. So I tends to believe it's not caused by ionization radiation, dielectric absorption or other high impedance magic, but something simple. One thing I'm kind of sure is these spikes have nothing to do with the ohm range, it persists even with Q302 removed.

Also, similar glitch noise appears on the preamp out when on 2pA range, verified by acquiring with Keithley 2001M, didn't try with the DAQ to see if also on other ranges as I'm afraid of frying it (although got it from scrap dump for free). So if it's the ranging amplifier stage causing it, it must happen at the input and can be transduced to the Preamp Out.

[Kleinstein]

The spikes could be due to switching of the signal to the ADC input.  When the ADC input is switched from a zero or 2 V (I don't know the sequence) reading to the input reading there would be a charge spike from the CMOS switch (U145). The internal ADC would ignore that spike and only measure after this.
Ideally there would be a resistor (e.g. 3 K) between the amplifiers output and the CMOS switch (pin 1 of U145), not to load down the OP-amp so much.
So the glitch could be a thing of the analog output only.

[zrq]

Thank you for the analysis, but this cannot explain the randomness of such glitches right? And they do appear on preamp out at least for 2pA range.

[Kleinstein]

The pulses from the CMOS switching should be at a fixed frequency, with the normal 330 ms reading cycle. As the pulses can be short (e.g. 1-10 µs) even the Ni DAQ card may miss some of them.
From the manual the sequence is 0 V , -2 V and than the actual signal in a +-2 V range.  So the expected pulse would be negative (may be with ringing in some ranges, but not much expected) at the analog output.

I would not expect much effect backwards from the ranging amplifier to the preamplifier, as there is a 200 K resistor between them.  So there could be another source of glitches too.

More random pulses in the 2 pA range would be something else. Maybe some radioactive decay or maybe some discharge happening somewhere.
A point to check would be if they also happen in the 2 nA range (so the next smaller FB resistor) - they may be quite a bit smaller.

[zrq]

The DAQ ran at 2Msps, so it's not likely that it missed these short pulses.
Well, I still tend to believe the random pulses on the 2 pA range have the origin with what's seen on DAQ, as they have similar randomness  (comparable event rate in Poisson distribution and visually comparable peak height distribution). Of course, I may be wrong. Can it be power supply glitches? I'm not aware of such sources.
One thing I'll probably try this weekend is to lift a pin from R126 to disable the ranging amplifier and following circuitry, and then sample at the Preamp Out with some protection to see if the pulses disappear or not.
BTW, I check earlier posts in this thread, not only Midi's data, the Alex's original data of LMC662 mod also features similar noise, so likely not a fault of ADA4530-1 alone... Even more mystery.

[Kleinstein]

The pulses shown look like they come in pairs, first negative and than positive. So something like capacitve coupling to some digital pulses or maybe a signal (e.g. supply) with strong random telegraph (popcorn) type noise.
The positive/negative pairs are something that does not fit to radiation effects.

[_Wim_]

There is also a popcorn noise, which I don't think originated from ionizing radiation.

Why do you think that?

[zrq]

Because I also see suspicious spikes with correct amplitude at ranges other than 2 pA, where it supposedly be much less sensitive to all these high impedance problems. One should reproduce on their meters by monitoring the 2V analog out with a deep memory oscilloscope for 600 s and using a timebase suitable for 10 us pulses, maybe for example on 2 nA or  200 mV or other ranges with x10 ranging amplifier gain. Those spikes can have 0.05V to 0.12V amplitude and are actually small bipolar oscillations.

I also tried the switching to Ohms range and back to Amps experiment yesterday, I may have stayed in the Ohms range for longer than 60s, but it took >25 min for my unit with COTO 9002 relays to settle back to <10 fA bias. I'm also suspecting C312, C318 can have more dielectric absorption than they should.

[zrq]

Today I quickly tried removal of C312 and C318 and do the test exactly as Wim described. The meter appears to settle a bit faster (Orange curve) then before after switching back from Ohms range, around 2x faster, but still needs 13 min to reach 10 fA.

[_Wim_]

Today I quickly tried removal of C312 and C318 and do the test exactly as Wim described. The meter appears to settle a bit faster (Orange curve) then before after switching back from Ohms range, around 2x faster, but still needs 13 min to reach 10 fA.

Interesting! I will try to make a comparison run related to the popcorn noise when I can spent some time in the lab...

[_Wim_]

[ Specified attachment is not available ]

One should reproduce on their meters by monitoring the 2V analog out with a deep memory oscilloscope for 600 s and using a timebase suitable for 10 us pulses, maybe for example on 2 nA or  200 mV or other ranges with x10 ranging amplifier gain. Those spikes can have 0.05V to 0.12V amplitude and are actually small bipolar oscillations.

Hi, I first made a 10min log of the 2V output (with range set at 2nA), but did not see any spike. After that, I configured my scope to trigger on a spike, and let it run over night (the buffer/segmented memory of the Picoscope was configured to store at least 500 triggers).

Upon leaving the lab I saw the first trigger when I turned of the lights. This morning I reviewed the 8 triggers, and I only saw triggers until we left home, and again saw a trigger when we arrived back in the middle of the night.

So, these "popcorn" spikes are spikes that enter via the AC into the 617 and make it through to the 2V output (at least in my case).

[zrq]

It seems I have a bigger trouble to solve with my meter before pursuing the noise. After transporting the meter between the day job lab and home lab (public transportation so exposed a little bit to the weather), it got unhappy and the bias cannot settle to 0 fA again even after a overnight warmup. While the bias level is always around 5 fA after stabilization, huge spikes of hundreds of fA in the reading can be appear.
Tried disconnecting relays but later discovered it's probably the triax connector went bad. Unplugging the input immediately reduced the bias back to <2 fA level before a full warm-up. Out of curiosity, tried again Wim's test of charging for 60s in Ohms range and switching back to Amps with the input unplugged. Wow, the connector is also contributing a lot to the "dielectric absorption"! In this configuration the meter settles back to <10fA in 400s.
Well, I have sonicated the connector assembly today and let's see if it gets better after soaked in IPA for a day. Still want to rescue it before paying 107 CHF for the 7078-TRX-TBC.

[zrq]

Finally made up mind and modernized the connector to 3-lug triax Pomona 5219, cannot afford a 7078-TRX-TBC. It didn't go through so well, as the hole on the panel is smaller than necessary for 5219. One need to brutally enlarge the hole to insert the Pomona. After I'm reconnecting the wires, I realized there is a possibility that it's not the connector is leaking, it can be the wires connecting the center pin was touching the metal tube for the guard. The coating on the black wire may simply not have good enough insulation. Anyway, already spent the money, so I moved forward. I salvaged some PTFE tubing from a dead gas chromatography vaporizer in a local trash dump, insulated the connection at the panel connector with that and made a simple few-pF PTFE coaxial capacitor replacing C312. While the meter is still nosier than it was before, the crazy bias current is gone and the dielectric absorption is much less obvious. (<7.5 min down to <10fA, doing Wim's test).
Also recapped the electrometer board, although it turned out every cap removed look fine and have rated capacitance.

Now I have a spare Pomona 5090 2-lug triax to BNC adapter and the removed 2-lug triax for sell. Well, probably not many people want them, but let me know if they can help someone.

[analogNewbie]

remove one lug, k617 can fit into modern triax connector.