Keysight B2980A series electrometer/picoammeter review/teardown

[TiN]

maxwell3e10
My 6430 has lower noise on most low ranges, much less tempco and it's also a full 4-quadrant SMU, not just electrometer with decoupled separated voltage source. Keithley alternative for B2987A would be 6517B. 6430 also has proper external remote sense for sourcing which can be helpful with leakages.

[HighVoltage]

I have sold my Keysight B2987A because I did not really like it too much.
Instead I got a Keithley 6517B and that one performs very well.

[bsw_m]

Can you describe the shortcomings of your B2987A?
And why K6517B turned out to be better?

[guenthert]

I don't think there is any corrections. Analog output is not output of TIA but generated separately from digitized code by separate DAC, unless I mistake something.
[..]

      Then why digitize that with a 3458A?  The B2987A has GPIB and Ethernet outputs.  Sure, if I had a 3458A I probably would use it any time I get a chance too, but ...

[TiN]

What is wrong with digitizing with 3458? 
I've also tried 1V source now, 10nA measure.
Log data: https://xdevs.com/b2987a_10nA_SRL100Mstd_1V_Hulk1src_megatec_jan2022/

[guenthert]

What is wrong with digitizing with 3458? 

      Er, there's no point in digitizing data which had been already digitized.  I thought it had already been established that the 'analog' output is created by a DAC (unlike the old Keithley 617, where it is the output from the range amplifier which also goes into the meter's own ADC).  If you bin the measured data you'll notice missing codes, from which one can estimate the word width of the DAC.  I'd be surprised if it is more than 12bit.  It's purpose is to drive moving coil meters or those old analog chart plotters, decidedly not to feed long scale DMMs (Keysight doesn't even bother to specify its characteristics).

[TiN]

It was 1 minute task, and bsw_m wanted to see it, so nothing to loose in doing so.

[bsw_m]

I think I need to clarify the reason for my request a bit.
I thought the analog output is the output from the transimpedance amplifier. But here I was wrong. Unfortunately, due to the fact that the analog output is implemented on the DAC, this request does not make sense.

[matches]

Nice tear down TiN.

The device has a real time noise monitor. For this function, the adc must run with a high sample rate. The monitor can be activated even with the longest integration times.
Therefore the integration must be numeric in contrast to the convetional integrating DMM's

I think I've read somewhere the ADC is always running at 100 kS and for longer aperture/integration times, the reading is simply averaged.
Does anyone know more about this?

[EC8010]

I borrowed a Keysight B2980A for evaluation. It was much nicer to use than a Keithley 6517B. Unfortunately, the key thing with an electrometer is its input leakage current/offset, which can be evaluated by capping the input and logging for a while. The B2980A didn't do well, so I bought a 6517B despite its appalling user interface. I later also bought a 6430. See the graph for a comparison of their drift.

The 6517B is sensitive to mains transients (you can see some disturbances on the graph). The 6430 is much more tolerant, possibly because of its median filter, or perhaps its power supply was designed better. The B2980A has a battery power option, so unplugged from the mains and on battery power, it should be immune to mains spikes. Realistically, a decent electrometer ought to be immune to mains disturbances, so in that respect the 6517B is flawed.

The good thing about the 6517 series is that they have a "2V OUT" connector that you can (and should) monitor on an oscilloscope to prove to yourself that you are measuring a genuine current, not the DC component of distorted mains hum. The 6430 does not have that option, so you have to take its numbers on trust.

Finally, there are four versions of Keithley 6517; 6517, 6517A, 6517B with vacuum fluorescent display, and 6517B with LCD display. The 6517, 6517A, and VFD 6517B all suffer hum (don't know about the later 6517B), with the 6517 being the worst because they put the EI mains transformer close to the input electrometer!

[Alex Nikitin]

Just for fun of it here is my Keithley 617 (the one in my profile picture) results for 6 hours today (the colored line is 1min averaging). I had to use Keithley 2015 meter to log 2V analogue output of the 617, as GPIB doesn't work at the moment on 617 (waiting for the GPIB chip to arrive). 1fA/div vertical scale.

Cheers

Alex

[EC8010]

That's really good on the long term drift (which is what's important). Do you know what the input stage is?

[Alex Nikitin]

That's really good on the long term drift (which is what's important). Do you know what the input stage is?

I've modified it in 2015. And the unit was recapped recently after some of the electrolytics failed. I have also the 263 to complement 617 and to have all kinds of perverted femtoamp-related fun .

Cheers

Alex

[EC8010]

Thanks for that. I've made a sub-femtoammeter (similar performance to 6430) using ADA4530-1 but only a single unswitched current range and had wondered about LMC662. Looks like it's time I ordered some.

[MiDi]

I borrowed a Keysight B2980A for evaluation. It was much nicer to use than a Keithley 6517B. Unfortunately, the key thing with an electrometer is its input leakage current/offset, which can be evaluated by capping the input and logging for a while. The B2980A didn't do well, so I bought a 6517B despite its appalling user interface. I later also bought a 6430. See the graph for a comparison of their drift.

The 6517B is sensitive to mains transients (you can see some disturbances on the graph). The 6430 is much more tolerant, possibly because of its median filter, or perhaps its power supply was designed better. The B2980A has a battery power option, so unplugged from the mains and on battery power, it should be immune to mains spikes. Realistically, a decent electrometer ought to be immune to mains disturbances, so in that respect the 6517B is flawed.

The good thing about the 6517 series is that they have a "2V OUT" connector that you can (and should) monitor on an oscilloscope to prove to yourself that you are measuring a genuine current, not the DC component of distorted mains hum. The 6430 does not have that option, so you have to take its numbers on trust.

Finally, there are four versions of Keithley 6517; 6517, 6517A, 6517B with vacuum fluorescent display, and 6517B with LCD display. The 6517, 6517A, and VFD 6517B all suffer hum (don't know about the later 6517B), with the 6517 being the worst because they put the EI mains transformer close to the input electrometer!

Just for fun of it here is my Keithley 617 (the one in my profile picture) results for 6 hours today (the colored line is 1min averaging). I had to use Keithley 2015 meter to log 2V analogue output of the 617, as GPIB doesn't work at the moment on 617 (waiting for the GPIB chip to arrive). 1fA/div vertical scale.

Are all those plots from a cold start?

[EC8010]

Mine had a one hour warm-up, just like the graph said. That allowed the really fierce drift to settle.

[Alex Nikitin]

As a rule I run my K617+K263 24x7 . Next time I'll power it down to replace the GPIB chip, I'll try to remember to check the warm-up drift.

Cheers

Alex

[EC8010]

I notice the unipolar peaks on the 617. That's something I've seen on my ADA4530-1, and which I've attributed to cosmics. Not a lot you can do about cosmics other than reducing the size of the ion chamber - making the screened box containing the electronics smaller. Although a median filter would help.

[bsw_m]

Just for fun I did a similar test for my DIY picoammeter.

[EC8010]

And that's even nicer. Warm-up?

[bsw_m]

Warm-up?

The warm-up test is still in progress, but initial data can already be published.

[bsw_m]

I've made a sub-femtoammeter (similar performance to 6430) using ADA4530-1 but only a single unswitched current range and had wondered about LMC662. Looks like it's time I ordered some.

The ADA4530-1 has better performance than the LMC662.
The OPA928 also looks promising, but unfortunately it is not available to me and I have no opportunity to test it. If you make your own meter design, pay main attention to the construction of the input circuits.

P.S.
This DIY picoammeter, which is a research project rather than a finished instrument uses the ADA4530-1 as the input amplifier.

[EC8010]

Indeed the ADA4530-1 is better, but you pay for that low leakage current with increased voltage noise, so for measuring larger currents (pA) a larger area FET is better. And utmost cleanliness is essential at low currents.

[bsw_m]

The voltage noise of the ADA4530-1 is only about 4µV p-p. For currents in the range of units to tens of pA, the transistor will have no advantage. But for currents above 10...100nA the noise will already be an advantage for example for OPA140.

P.S.
I would also like to draw attention to the effect of ADA4530-1 noise on the resulting uncertainty of the meter, against the background of other sources of error and uncertainty.

[Alex Nikitin]

K617 measuring +/-10fA current supplied from K263 over 1.5m of a low noise triaxial cable. "Zero" current - 5min +10fA, 5min -10fA, back to "Zero".

Cheers

Alex