Cooling the electrometer front-end

Here's my fA parts tester. The critical node is mid-air.

The crummy RatShack binding posts were leaky so I had to add the polycarb.

I was measuring down to about 10 fA. There are better opamps around now.

Cooling will cause condensation. Bad.

Thank you, John. So I am on the right track.

IMHO, if you glued the opamp upside-down, the remaining small components could be floating in mid-air without that supporting structure. Was high-g a real concern?

LMC6001 is not that bad. The ADA4530-1 supports input guarding, though.

Best regards, Piotr

ADA4530 Ouch. $29.370, Single Amp

How about the LMC662 - $1.800 at Rochester, qty 1, Dual, Input Bias Current

2 fA

As far as cooling, I don't think you see much until you get down to liquid nitrogen temperatures. Then you need ventilation to avoid death.

I had good fun with LPC661 amplifiers. They're specified as having a 2fA bias current, but the ones I measured only had about 150aA, more than fifteen times better. I had an tiny ionization chamber attached, so the real bias current may well be lower still.

This was at room temperature, with the critical node wired in the air. Cost not even 1$/pc for a reel-full.

Jeroen Belleman

You could surface-mount it and bend the +IN pin out into free space.

The driven guard is very nice. That should block leakage across the SO8 package.

Ib is about 600 electrons per second.

Don't expect anything to happen fast.

Air is good, as is a PTFE-insulated standoff or ceramic; some PCB materials with guard-rings are good, too, I suspect, but this is a TSSOP package. It might be hard to fit in a guard ring, but you can still bore a hole under the high-Z pin, and carefully hand-wire it to a thin coax with shield driven appropriately. It doesn't matter if the PCB leaks, if the sensitive lead isn't in contact.

It might be possible too, to plunge the coax center wire THROUGH the PCB and butt to the lead wire ; a snug fit will keep the wire steady during soldering.

Most of the PTFE insulators I see in my junk box are BNC jacks. Solder type ends for those are circa 2mm, and TSSOP gives you about 1mm to work with. You could mount such a jack next to the chip, and just flying-wire the last centimeter or two.

It's actually a regular SO8, so it makes things simpler. But, given the QTY 1 and the inadequacy of FR4, I am not considering making a PCB for this front-end. The current idea is to enclose it in a metal box and machine a 0.6mm copper plate with a cavity in the middle of the bottom, edge, just big enough to embrace the SO package tangentially. This U-shaped plate will then be soldered to both GRD pins, splitting the SO8 into two halves and providing reasonable stiffness. This plate, in turn, will be then soldered to the enclosure using a small number of those nifty SMD AlN thermal bridges due to their "infinite" resistance and solderable ends. This levitated and GRD-screened op-amp should take full advantage of the superior resistance and low soakage of air.

In addition to the 1x, 10x, 100x and 1000x selectable gains I would like to be able to connect one of several shunts too. Heck, what is the off-state resistance of a reed relay? :) The specs say 10G+, but I believe they were too lazy to measure it.

And what triax connector is actually good?

No problem with flying parts and connections, this is a lab unit not subject to automated assembly anyway.

Best regard, Piotr

The thermal bridges sound like a nifty repurposing; those are usually kinda short, though, so cleaning is important. I've heard of using small-value ceramic capacitors, and a burnt-out ceramic fuse should have some standoff virtues: that's something a good electrometer can test!

Pease had a jig for low-current testing, using plain old springy pushbutton switch contacts and off-the-shelf relays

For quick-connecting a shunt, you might consider the magnetic-power-connector gizmos like here

which eliminates the spring tension uncertainty.

None, probably; to active-shield a long wire, I'd use coax for signal (center) and shield (outer), with a braid sleeve overall for a ground. The logic is, capacitance signal-to-shield is constant (and has null effect due to driven shield) while capacitance shield-to-ground is high impedance compared to shield drive, so it can vary without harm.

My solution, of course, is a quick and dirty one-off.

The old electrometers used sapphire insulators, and instructed on careful cleaning; I'd think quartz or white ceramic is good, too. The epoxy of an IC package will set a leakage minimum.

Ceramics and glass are generally disappointing insulators in the femtoamp range. PTFE is very good, and a clean plastic DIP package is also very very good. Anything hydrophilic such as polyamide (nylon) is a disaster.

Glass can have low leakage, but the soakage (dielectric absorption) of a glass capacitor has to be seen to be believed. :(

I do electrometer stuff using small through-hole parts assembled dead-bug style. Good Medicine.

Cheers

Phil Hobbs

They come in various sizes, but there is another option:

These are pretty big. One can solder them to a copper base plate and then apply precision abrasive Dremel trimming to move their resistance a bit closer to infinity. Same can be done to any SMD resistor, so I don't know if in this application there are any inherent advantages of AlN as compared to Al2O3. I don't know if grinding would not contaminate the substrate, though. Not an issue for generic structural support, but might be for sensitive wire holders.

They discuss 1TOhm resistors and 100TOhm leakage in the datasheet, so the IC package is not a primary concern IMHO.

Thank you for your hits, much appreciated!

Best regards, Piotr

Axon Instruments cools their "Axopatch" front end ...inside a tiny gold-plated DIP14 box, epoxy sealed, N2 charged, with a huge piece of dessicant pellet sealed inside. Their TEC is internal, few mm on a side, with a heavy copper finger against the bottom of the little box. -30C operating temp. w/thermistor, for shutting down power until leakage gets low enough.

See photo

Gold-plated box is a driven shield. The input is via a gold pin through a little ceramic disk, then a large ?hdpe? threaded part. They used JFET dies on ceramic, deadbug leadbond, also no feedback RC, instead it's operated as an integrator, with a slow periodic FET reset, then a differentiator to get back the DC signal. Supposedly this thing can watch the state of single ion channels in neurons. $5000 ea, iirc.

Neurons would freeze solid at -30C. Not much ion channel activity at that temperature.

On a sunny day (Wed, 25 May 2022 16:27:58 -0700 (PDT)) it happened Bill Beaty snipped-for-privacy@eskimo.com wrote in snipped-for-privacy@googlegroups.com:

For cellphone towers there exist superconducting filters, so cooled to a few degrees above absolute zero, I happen to have one:

dewar with filter on the left side Use it for all sort of experiments..

Jan Panteltje snipped-for-privacy@yahoo.com wrote in news:t6n998$6ak$ snipped-for-privacy@dont-email.me:

snip

The coolest thing in the universe...

Heh, better check photos before commenting, and for info, search "patch clamp," capillary microprobes of in vitro cell surfaces. The -30C cooled JFET front-end is inside Axon Instruments "Axopatch" electrometer.

And no, you don't put your microscope slides INSIDE the tiny electrometer!

Only the dual JFET itself is cooled, inside the sealed DIP14 package, inside the "headstage" housing, which is typically bolted to a microscope stage.

I don't know about that particular part but some do improve a lot with active cooling well below ambient. Same is true for CCD operation.

This patent exploited the effects of cooling certain instrumentation opamps for mass spectrometry Faraday collectors back in the 1980's.

The improvement in LF noise was a bit of a surprise. Later versions of the kit cooled things a bit more aggressively. It doesn't give much away ;-)

I can't recall if it was granted or not.

Things were made a bit more interesting by it all having to work in a high vacuum and also withstand periodic baking to 150C.

1/f noise is caused by conductance fluctuations, which basically require mass motion. (At very low temperature there are long-lived traps that contribute as well, iirc.) So for old-timey parts that were made on less-clean processes, I can easily believe that cooling helps the 1/f noise a lot.

If you go too cold on a JFET, the transconductance goes into the tank, so that the voltage noise gets worse.

The noise voltage is ideally sqrt(8 kT/( 3 * g_M)) in 1 Hz, i.e. sqrt(2/3) of the Johnson noise for the same conductance. A BJT is slightly better, at sqrt(1/2)--a BJT emitter effectively has a noise temperature of T_amb / 2.

Cheers

Phil Hobbs

ISTR we used PEEK which is a more nearly engineering grade plastic than cheaper PTFE which tends to deform and creep under stress. Although it is a thermoplastic it could survive the normal baking for standard mass specs. It doesn't outgas much at all after the first heating session.

I recall several tricks of that sort relying on the package starting out clean and staying clean (not always a safe bet in routine manufacturing). Easy enough for a one off being made carefully though.