34401A input jFETs - 10uV/C drift?

[cape zoloh]

A schematic of the input stage of the 33401A multimeter is attached. Note the matched jFETs, the first component seen by the input signal.
The component is the Vishay U406. (Schematic can be found here: https://datasheet.octopart.com/U406-Vishay-datasheet-9672.pdf)
This part has an (MAX) offset voltage drift of 10uV/C. I can't say what the drift is typically, but isn't the max rating very high for such a relative precision application? Is the drift in fact some orders of magnitude better, and if so, why is that?

[Kleinstein]

The specs are often limited by the testing. So the typical actual value can be quite a bit better than the test limit. The temperature drift, like low frequency noise and pA range leakage are properties that are slow to test - so the tests are often not very strict and often only typical values are given. The parts used in the DMM may be additional screened parts. So the HP number may have tighter specs - though essentially the same part.

The drift of the JFETs would effect the drift in non AZ mode only. When in AZ mode, drift of the FETs would be corrected.

[tggzzz]

This is discussed in section 5.12 "Designs by the masters: Agilent’s accurate DMMs" of The Art of Electronics 3.

Section 5.12.1 is "It’s impossible!". Section 5.12.2 is "Wrong – it is possible!".

If you are at the level where you are thinking at this level, then you will find TAoE3 invaluable. But expect that every time you look something up in it, your nose will remain buried in the book for several hours

[cape zoloh]

GREAT TIP. Thanks you.

[Kleinstein]

If nothing else works, one could change (adjust) the TC be changing the current to the FETs in the pair. This would change the offset and also the TC. As far as I see the circuit, the 34401 should take care of the fixed offset from the last zero of the input.

[David Hess]

It does not look like they implemented any active correction for the offset voltage drift other than the unshown automatic zero.  It is fascinating to consider *why* chopper stabilization is not used.  Also, a differential stage is not the only way to get a low offset voltage drift FET input stage; Vgs of a single JFET can be compensated and adjusted using the source current with the advantage of lower noise over a differential pair.

As the other posts alluded to, testing for things like offset voltage drift, noise, and low bias or leakage current is time consuming so there is a severe tradeoff between guaranteed specifications for these and cost.  Manufacturers may choose to grade or select parts themselves, or make arrangements with the manufacturer of the parts to do it for them.  Some designs include automatic calibration.

[Kleinstein]

HP chooses doing the AZ mode by switching all the way at the input. So in AZ mode they don't need a chopper stabilized amplifier. The AZ mode is a kind of chopping done with one half done digital.
Using chopper stabilization would either add current spikes from the chopper or additional noise from the filter to suppress these spikes. Especially in the old days chopper OPs were also quite noisy. So there was a good justification for not using chopper stabilization. The AZ switching at the input still produces some comparable (maybe stronger) switching spikes, but these are at a low frequency and with some dead time.

Some adjustment for the TC could make some sense (AFAIK it is done in the ADT 6581, 8.5 digit meter), though maybe not needed for a 6 digit meter (lower resolution at higher speeds were the non AZ mode may be relevant).
Still the adjustment would be time consuming.

[cape zoloh]

One part not mentioned here but in the book recommended above is that the differential pair in the more fancy HP 7½ one is bootstrapped, so it doesn't get affected by the input voltage, as it "moves with the rails".
But OK, the lession here is correction DACs are used, and offsets can definitely be adjusted and corrected, down to silly numbers of precision.

[Kleinstein]

The JFETs in the 34401 also get bootstrapping: Q104 provides a bootstrapped supply to the drains, that is higher by UGS of Q104 than the source voltage.

The 2 OPs to the left are just 2 fancy constant current sources, so that the JFETs work as source followers for the inputs of U106. I don't see a correction DAC for the main amplifiers offset - there is one for the pre-change amplifier in some of the HP meters, but this is not just for the Offset, but more to trim the overall charge injection.
The main offset correction is done numerically from the ADC results.

[cape zoloh]

Which model are you refering to here? And what do you mean by AZ done with one half digital? As in correctional DACs?

best,
Nicolas

HP chooses doing the AZ mode by switching all the way at the input. [......] The AZ mode is a kind of chopping done with one half done digital.

[Kleinstein]

The auto zero mode of the 34401 (and most other high resolution HP dmms) works in a way that the input is switched between the actual input and a zero reading. The result is than the difference of the 2 readings. So for 10 PLC (50Hz) the input is sampled for some 200 ms and the short is sampled for some 200 ms. I don't know which of the 2 conversions is done first or maybe use both the zero before and after the actual input.

This is similar to a classic one side chopper amplifier. In the analog version the difference (demodulation) is done analog, but the principle is the same. This is why I called the AZ mode half digital chopping - the demodulator part is moved to the digital domain after the ADC.

I don't know a DMM that uses a DAC to trim the input amplifier offset (it is still a mechanical trimmer in the ADTV6581). Only the pre-charge phase often has a DAC for adjustment - this is more to trim the current spikes from switching and indirectly the bias current. Some old meters (e.g. Datron) have an adjustment for the input bias current and this may be via a DAC.

With the AZ mode the drift of the amplifier and very low frequency (< 2 Hz) noise does not matter.
A low drift amplifier would be needed mainly for the non AZ mode.

AFAIK the AZ mode has not changed much with the HP/Keysight meters (e.g. from 3456  to the current 34470). The older models (e.g. 3455) and many Keithley meters have also an additional reading of a reference voltage to get a real time correction for gain drift, but the correction of the offset was/is similar - though not always as good.