Ed,

Nice job, please keep us posted ! Are you planning to publish some
additional data, such as a circuit diagram ? I am thinking about doing a
similar project on my Keithley 616, which currently has approximately
25fA of residual input current (the spec says it should be less than
5fA). And no amount of cleaning seems to bring it down.

I also hope a redesigned input stage using a good op-amp will have lower
offset voltage drift.

Thanks,

Joel Setton

Joel, I too have a K616. When I get a chance, I will pull it down and
look at the bias current. I also have a K610CR, and a K602, which I
looked at last week. These all use the same front-end module based on
P-channel MOSFETs with unprotected gates, differentially clamped (gate
to gate) with external diodes or transistor junctions. And, they all
have the same 5 fA bias spec.

Of all the MOSFETs from the old days, I know of only one type so far -
the 3N163, I think - that has the unprotected gate. The early MOSFETs
(called IGFETs back then) needed special anti-static storage and
handling, then later models added built-in Zener gate protection for
more durability. This is what spoiled them for ultra-low current
applications.

So, I'm pretty sure Keithley used selected and matched 3N163s or an
equivalent, for the inputs, and selected small-signal bipolar
transistors for the gate clamping B-C junctions.

My 610CR fired up OK after many idle years, and settles down to right
around 5 fA, so OK there, but is very jumpy and sensitive to every
mechanical effect like tapping the controls and cabinet. I need to do a
thorough cleaning and Deoxit treatment on all the controls, and tighten
things up mechanically. I also found either a packaging design flaw, or
that there may be a missing shield part under the input section at the
front panel. There is a huge open area around the input connector at the
bottom, where it's exposed to the bottom cabinet cover. The cover edge
is just placed in a slot in the front panel, and has a couple clips
retaining the back. The slightest vibration or shifting position causes
huge jumps that would be gone if proper shielding was inside. I'm
assuming there's a missing part on mine, more likely than Keithley
missing this vital issue - either way I need to make an inner shield
part for it eventually.

The 602 also fired up OK, after changing a couple of its six 9-volt
batteries. It settles to around 1 fA, and is not very jumpy, but the
zero drifts quite a bit. I think this is mostly due to the battery
voltage dropping during use (there is no regulation), and maybe would be
much better with all fresh batteries. BTW I modified this unit years ago
to replace the big old-style C-Zn block batteries with modern 9V ones,
and eliminated the Hg cell, by using a 1.22 V IC regulator and different
scaling resistors to get the 1 V reference for ohms mode.

These models both are subject to rather large voltage offset drift, so
need quite a bit of tweaking to keep everything good, even though the
bias currents are OK. It's interesting that there are three levels of
zero adjust - H, M, L - to accommodate some amount of MOSFET variation
and operating temperature range. But, that's what it took back then, and
it's still pretty respectable, especially considering these are about
fifty years old.

Anyway, regarding getting the bias current back to spec on a 616, a
thorough cleaning of all high-Z insulating structures should do it, but
don't forget to do the input MOSFETs. Their weak point is the package,
and possible gate leakage on the surface, from the leads to each other,
and to the can, which I think is the substrate and source, tied together
in this application. If you've tried every other possible thing but not
the MOSFETs, then I think it will work wonders. Gain effects tend to be
from leakage on feedback elements, while bias current and offsets tend
to be adjacent-node problems with DC present - even seemingly tiny
amounts. Also, since there's a digital voltmeter measurement behind all
this, part of getting things zeroed may include what's going on in there
too. I don't know offhand.

Speaking of device packages and leakage and such, if there's enough
interest, I can tell my "AD542 story" from decades ago, which explains a
lot about this subject, and why I always pay lots of attention to it,
and how to avoid or improve it. And it would include a free rant about
how crappy device pinouts are for some purposes, and how the standard
advice in every high-Z data sheet and app note ignores the biggest problem.

I'll report some more about the 417 project later, and maybe see what
the 616 looks like by then too.

Ed

Joel, did you let your 616 warm up for a good long time? I fired up mine
yesterday and it read over 50 fA, then increased to over 130 fA after a
while. I figured maybe there was something wrong with it - yet another
repair project. I studied the manual a bit and noticed it said that if
the 616 has not been used in a long time, that it could take very long
(unspecified) to stabilize when first put back in service. I left it
running overnight, and got around to checking it late this afternoon. It
read about +4 fA, but then I found the offset voltage had drifted up a
bit. After re-zeroing, it now drifts around between +1 and +2 fA, with
occasional jumps. This is at fairly high ambient - it's somewhere in the
80s (F) here now. So, after about a day of burn-in, it's well within
spec, and quite respectable. Later I'll try shut down and restarting
tests to see how quickly it recovers to spec with more normal use.

Maybe you should let yours cook for a while, if you haven't already,
before assessing it.  Good luck.

Ed

One more thing - be sure you're interpreting the readout properly. I got
quite confused at first, trying to figure out which digit is how much
current. At maximum sensitivity, the last digit is 100 aA, so my reading
for example is 10 to 20 counts, or 1.0 to 2.0 fA. That last digit is
mostly noise, but gives some idea of what's going on in between.

Ed

Ed,

Thank you for the tip on "warm-up" time -- or rather in this case, time
for leakage to cool down. I will definitely give the 616 a chance by
leaving it on for the next 24 hours.

Last time I did a measurement, it settled in the +20 fA range, and I
confirmed this number by doing a dV/dT measurement in the charge mode.
In this configuration, the input current plus leakage gets to charge one
of the built-in capacitors (I selected the 1E-9 range) and the rate of
voltage change (dV/dT) is a good indicator of current.

I once had a K610B electrometer, which as you know used a differential
pair of 5886 tubes. It worked, but zero drift was very bothersome. I was
also weary of using a tube-based instrument, which at the age of 50
could fail at any time with a dead tube. I finally sold it and bought
the K616.

That's all for now, I'll write again tomorrow with measurement results !

Joel Setton

Hello Ed,

I left the '616 running for a full day, and the residual current remains
remarkably stable, between +24 and +27 fA, even with ambient temperature
ranging from 68F at night to 94F in the afternoon. I was expecting wide
changes with temperature, but this was not found to be the case.

I may try some additional testing to determine whether the leakage is in
the input switch or in the input MOSFET. This needs some further thinking !

Joel

Update.
I think I'm done with the overall new design, except for fine tuning and
compensation, which will have to wait until I can build an extension
cable to work on the plug-in out in the open. I did a lot of
improvements mostly in the power supply, and added four new multiplier
ranges, adjustable bias cancellation, and added a switch and new
circuitry to run the remote-zero solenoid. It's looking pretty good now,
burning in out in the garage.

For ranging, I built a new switch with eight positions and a couple
extra wafers for signal routing, extending the multiplier step sequence
out to 0.03, 0.01, 0.003, and 0.001, providing maximum sensitivity of 1
fA FS. I originally planned to add two steps, after seeing how well the
new IC-based front end was working, bias-current-wise. Searching my
switch parts inventory, I could only find a compatible one with eight,
without doing extra mechanical mods. So figured I'd go with it and have
two left for experimental setups.

The extra gain to go ten and a hundred times more is provided by a new
meter drive circuit. I did some experiments with the original ranging
feedback divider, which looked promising at 10X more, but made it very,
very slow in response, without fixing up the overall compensation. So, I
opted instead to change the meter sensitivity by 10X. In principle, this
was no sweat, but the meter in this machine is 1 mA FS, so needs fairly
low network resistance, making clamping and protecting it more
complicated. It would require choosing between probably inadequate meter
protection, excessive clamp leakage error, and not overloading the
electrometer's output, which would cause instability on recovery from
large excursions. This was solved by using a separate opamp follower
circuit to buffer the original meter signal, and provide limiting to
protect the meter movement - no more than 2-3 full-scales (mA) worth of
drive current can be applied, under any condition. This is about the
same as with the original meter setup. This worked out quite well, once
I ran the meter circuit from separate supplies. When it was from the
same ones the electrometer circuit used, it made a relaxation oscillator

• the slight changes (maybe 10 mV) in the supplies during overload were
  enough to upset the front end, so it would just bang back and forth
  between extremes.

Once I had committed to the opamp follower setup, the next natural thing
was to consider getting 10X more still, by having a 1X/10X amplifier
gain mode, and that's what I did. So, the first 10X shift is from the
low resistance meter drive loop, providing 300 mV FS, which is a tenth
of the original 3V. The next 10X comes from switching the follower's
feedback for 10X voltage gain, while the meter is still set for 300 mV,
so it reads 30 mV FS from the electrometer output. The other, original
ranges are the same as before, with about 3 kohms of meter and resistor
load. The range switch is set up to continue the original 0.3 and 0.1
multiplier feedback pattern, and route the meter to the original path or
the new path for 10X, and activate the 10X opamp gain in the last two
settings.

It all worked out well, especially the new 30 fA and 10 fA ranges. With
bias current looking like less than 1 fA most of the time, it's ten
percent or less of the 10 fA scale, and with the new bias cancelling
deal, it can be pretty well zeroed out, for short term operation. The 3
fA and 1 fA ranges are of course so sensitive that they're not really
usable for external measurements without a lot of watching over and
tweaking - it can go off-scale at any time from random events, but
gradually drifts back. They are experimental for now, but very useful
for making assessments and further improvements - I can really see
what's going on now.

The bias cancelling scheme is similar to one I made for an old K410A
tube unit (now long gone), except that the tube ones tend to have only
unidirectional grid current, while for this I needed bidirectional
control. This was done with a ten turn pot driving the LEDs of two
opto-couplers differentially, and connecting their E-B pins
anti-parallel so their currents are opposite, and cancel when equal. The
CTRs of the optos don't necessarily match, but with the right setup,
there is a range of adjustment near the middle where there's no net
current out, and at either extreme, a maximum of each polarity. This one
is set up to deliver about +/- 1 uA. The voltage is set by the load
resistance, in this case 5 kohms, so about +/- 5 mV out.

The receiving end of this signal is the high megohm feedback resistor
array, but only the highest two or three may need it. In the lower
sensitivity ranges, the bias current is insignificant (even though it's
still there). The small voltage signal is superimposed on the feedback
signal voltage at each big resistor's end - the low-Z, feedback end - in
accordance with the scaling needed. In this case, the +/- 5 mV is added
at the E12 resistor, and that signal divided by ten is added at the E11
resistor, and so on. The divider is a 9k, 900, 100 ohm setup, but so far
I have only connected the E12 and E11. The E10 circuitry would need a
bit more hacking up of the board, so I skipped it for now but left the
provision for it. This divider is insignificant compared to the huge
feedback resistors, and with no current applied, the tiny feedback
currents go through as usual. When correction current is applied, the
result is a small voltage causing an excess small current in the
resistor, that can cancel the input bias current. The E12 one, for
example, will produce 1 fA/mV, which can cancel 1 fA at the input, if
the polarity is right. So, the floating output of the opto rides on the
feedback, and tweaks the bias according to the pot setting. A load trim
resistor on the opto output sets the desired voltage range. I'm leaving
it wide for now, but may eventually go narrower. Wide is good for
experimenting like setting it at zero-center in the 3 and 1 fA ranges,
where the needle bobs around a lot in both directions. Narrow is better
for setpoint resolution, maybe after things are refined enough. It's
also open-loop, of course, without the operator to turn the pot and
adjust it accordingly - and slowly - not quite tedious, but very slow.
Also, as the input bias drifts it needs readjustment. You can't do much
about big, random hits, except wait for the needle to return.

I'll have more to report after I see how it's actually running later.

Ed

Ed,

I can't believe you're working on 1fA full-scale ! This is impressive
indeed and even though your message gives some very useful explanations,
I can't wait to see your schematic and learn from it. Of course I'm
staying tuned.

Then, at the end of your project, we may have to change the name of this
list to include the "amp-nuts" side of things !!!

Joel

Joel and Lou,
I will be sending you pdfs of the preliminary schematics and notes soon,
probably this weekend. This all is particular to the 417, of course, but
some of the principles can apply to other electrometer designs and upgrades.

Ed