sensitive electrometer

BTW, Daniel, you do know the 640 vibrating membrane approach was superseded by low-gate-current MOSFETs in the 642, which was Keithley's next high-performance remote-head model? Now similar fA capability can be had with a few NSC MOSFET opamps.

I read in sci.electronics.design that snipped-for-privacy@web.de wrote (in ) about 'sensitive electrometer', on Thu, 6 Oct 2005:

The 400 kHz acts as a polarizing voltage, and variations in its amplitude are amplified by the AC amplifier much more accurately (and/or easily) than could be done using a DC polarizing voltage (or none at all) and a DC amplifier. As the manual implies, it needs to be a frequency much higher than the resonance frequency of the membrane.

You could make a toy one from an electret microphone capsule with the polarising electrode attached over the sound entrance but insulated from the capsule body.

I read in sci.electronics.design that Winfield Hill wrote (in ) about 'sensitive electrometer', on Thu, 6 Oct 2005:

You can get fA capability much easier than that. In fact, there's a glut of it.

Hello

I know of the 642 but the paper by Dimeff talks about 10^-19 A current sensitivity and the NSC OPs only have bias current of around 2 fA. (in fact I was not going to build one but the mechanical principle is fascinating) I am just wondering about the principle of the 642 because the Dimeff thing needs an excitation frequency of exactly the resonance frequency. Without DC bias the membrane will not resonate, because force is proportional to squared voltage and as a result there is only an excitation at DC and at the double resonance frequency, where the membrane will not follow the signal.

Does anybody know about a paper explaining how the 640 works?

Many thanks

Daniel

Robert,

I am sorry about this, but publishers are criminal in my opinion. If I download the paper my IP and my institution are included in the pdf and I am not allowed to give this to anyone.

Daniel

These are not mutually exclusive. All sensitive electrometers have input leakage, which they deal with by means of an input- current offset control, on the front panel, or invoked under uP control during an autozero cycle, etc. This means the ultimate sensitivity is in effect determined by the noise and drift rate away from the auto-zero value. Moreover, if the open-circuit drift rate can be modeled, the claimed measurement sensitivity can be further improved. NSC's opamps are much better than 2fA, assuming you can create an appropriate environment for them.

Differential pair of selected MOSFETs. Sapphire insulators, etc.

That "scitation" reference is not too useful; Ya gotsta reggiesterr eN *pay* !!

Who makes 3N155A these days?

Regards,

Boris Mohar

Got Knock? - see: Viatrack Printed Circuit Designs (among other things)

void _-void-_ in the obvious place

The 10^14 number is clearly far too conservative, e.g., having been thoroughly beaten by NSC. My comment about using NSC's chips to make measurements below 1fA refers to their low leakage drift rate, which allows you to correct for the standing level and then observe changes from that with a sensitivity below 1fA.

As far as a PCB is concerned, no-one would let this measurement node anywhere near the PCB: I've used small teflon standoffs, as well as raised-leg soldered-in-air connections, which works well.

I'm curious about this comment, Win. If I recall correctly, I remember reading something about the IC packaging having a bulk impedance figure around the vicinity of 10^12-10^14 ohms. If so, this would seem to be a barrier and would suggest to me the asking for unpackaged dice. But is it truly possible to achieve "much better" than 2fA in packaged components automatically placed on common (perhaps with cutouts in them) PCB materials, even assuming a sealed metal can bonded to the PCB?

Jon

They do not, to my knowledge.

Right, creating low-leakage protection diodes is the trick. There's reason to believe the plastic package may be "perfect."

This would be a useful idea, but only for specific configurations for the opamp. Bob Pease isn't telling how they accomplished the low leakage level, but it seems clear they haven't used a guard, at least in the conventional sense (i.e. from the "+" pin). It's possible each gate pin guards its own ESD diode, ultimately from the same MOSFET's source, but IIRC, there's reason to believe they don't do that either.

Specsmanship. Inability to create a perfect plastic *all the time* - who knows.

In fact there are about 20 problems you may run into in the under- 100fA region, so watch out and don't have your expectations up too high. I have written about 6 to 10 of the problems, and solutions, if you search my previous postings here on s.e.d. Add the search word Keithley to your Google advanced usenet search form.

No, they say 2fA on the front page, and the "typical" specs column says 0.002 with the unit of pA on the far right side of the page. So that's 2fA typical for the LMC660 and LMC662. Anyway, ignore all that, we're relying on Bob Pease's remarks about these ICs.

Hi Win, Do you know if NSC bootstrap the ESD proteciton diodes on their op-amps? I'd be pretty sure that the leakage of the ESD protection would be the dominant leakage mechanism on the chip (but perhaps not as bad as the package though). If they were to drive the other end of the ESD diodes with a guard signal, that guard signal being in turn ESD protected with diodes to the supplies, then I would expect that lower leakage could be obtained than with plain diodes from the input pins to the supplies. Perhaps they already do this - do you think so? Although I have never seen it done, they could also drive the pins surrounding the input pins with this guard signal if low leakage through the packace were the problem. I wonder if anyone makes such a thing.

Chris

Isn't it just a matter of balancing the reverse leakage currents- there is no telling what can be achieved with a 4-layer integrated structure looking like two diodes between the supplies and a weakly enhanced MOSFET transversely.

Okay. Thanks. Although I'm very inexperienced in electronics, I understand at least somewhat about the difference between a bias current that can be mostly accounted for (leaving some residual error) and also time drift away from that accounting.

That said, if the package itself is so good (and I read that you used the word 'perfect' to possibly describe it), then why is it that the better COTO relays tend to specify leakage resistances in the 10^12 ohms region? (I'm thinking here that one of the reasons that someone would bother using a COTO relay to switch gains on the 1st stage of a transimpedance amp would be for its OFF leakage, and that COTO would be driven to use some of the better packaging available, if possible

-- at least, that is the inspiration for my further question.) Or is it that NSC truly has done something very special that others cannot achieve in their packaging? I guess I'm still just having a hard time with the idea that bulk impedances in packaging can really achieve that kind of level of insulation, which must be in the 10^16 region or better (yes?) in very tiny packages.

Gotcha. I assume this part is packaged in such a fashion as to encourage these configurations.

Just imagining other problems, it seems like the whole thing with some of the circuit would need to be placed in at least an electrostatic shielding can and sealed dry (hmm, is the packaging hydrophobic?)

...

Finally, I'm at a loss for the part number to look at. I went to National's web site (I'm assuming this is "NSC") and did find dies with an input bias of 2fA, namely the MNLMC662AM-X REV 0B1 data sheet, but this is unpackaged. The packaged LMC662 appears to spec 2pA (they say .002nA which I read as 2pA.) The LMC6001 is also 2pA, not 2fA. So I'm still looking for a packaged 2fA part (rubbishemail had posted this comment "NSC OPs only have bias current of around 2 fA") and wouldn't mind being given a clue where I could read about such a thing. So far, I'm only finding that spec in a national _die_ and not a _package_.

Jon

I should add that the die form specifies 3mV offset voltage with a drift of 1.3uV/C, so it seems to me that a diode detector would either need to be "very, very cold" to get its own impedance very much higher or else the 3mV would need to be tweaked away, perhaps with a DAC to a resistor divider biasing the other, non-summing input node.

Jon

Okay. Trust nothing at these levels, then.

Will do.

I was looking only so far as the web page:

and also the parametric search page (which says the same thing.)

As you can see, that page says: "Max Input Bias Current (nA) .0040, .0020"

So I downloaded the data sheet PDF and see exactly what you say about it. At least that explains my confusion on this point. Thanks.

Jon

No, apparently the tough issue for those parts isn't their performance, it's the difficulty making production measurements at low currents to support a lower "maximum" spec. Such very-low current measurements would require specialized home-made circuitry, rather than standard production test equipment, and would take much more time for each part in the test fixture. The solution was to use a high tested-max spec, and list the much lower typical value and then write and talk about it.

Well for all sensible amplifier configurations, the two inputs have the same voltage, though admittedly it may not be equal to the output voltage. They could make the guard voltage using a diode connected device matched to and running at the same current density as the input pair, with its source connected to the source terminals of the two input devices, assuming the input is a differential pair.

Thanks for the info, especially about Bob Pease's comments. It sounds from what you say that they actually tried to get low input current, whereas I had previously assumed (from the fact that 2fA is always a 'typical' spec, and the max is much higher) that they didn't really care about the low input current and that it was an accident. If it was an accident, then I thought that there might be the potential for it to be even better if they were trying.

With these parts, I have found it impossible to clean the package myself to the same standard as a new part. After I have handled one, therefore, it must be demoted to a lesser duty and a new part put into the low current circuit. Thus for me it seems important that the distributor that I buy from must not handle the parts at all when sending them to me. Maybe I will have to expect a certain percentage of parts will be unusable, and just rely on the statistical chance that there are some parts where the important part of the package hasn't been touched. I might have better luck if I buy a full tube.

Chris