4 wire zener diode

[jimon]

Maybe I don't understand something, but when I look on voltage references circuits I often see something like this :


And then I read that zener diode is quite stable, but wires around it drift with time (LTFLU case). Ok, wires actually pass some current through zener diode, if wire resistance drift slightly then overall voltage drop will drift as well.

So why not just tap off actual voltage drop on zener diode itself using 4 wire technique ? Using high impedance op-amp it should be possible to minimize drift of sense connectors because minimal current is passing through them. Or am I just missing something ?

[timb]

A lot of higher end voltage references have force and sense connections to compensate for voltage drop.

So, yes, you could use Kelvin connections from a Zener to a low offset, low noise op-amp.

Keep in mind it absolutely has to be low noise, zero drift op-amp or you won't see any benefits.

Basically, your performance is only as good the weakest link in your signal chain. Linear Tech makes some good op-amps for this sort of thing.


Sent from my Smartphone

[w2aew]

Ordinary zener diodes are not that precise or stable.  The temperature coefficient varies with the zener voltage and the operating current.  There are special temperature compensated zener diodes (such as http://www.microsemi.com/document-portal/doc_view/125410-lds-0155), but even these aren't as stable as a good, precise voltage reference.

[jimon]

I mean, why they dont use 4 wire approach in LTZ1000 or LTFLU if they know that wire resistance is gonna drift with time ?

[timb]

Ordinary zener diodes are not that precise or stable.  The temperature coefficient varies with the zener voltage and the operating current.  There are special temperature compensated zener diodes (such as http://www.microsemi.com/document-portal/doc_view/125410-lds-0155), but even these aren't as stable as a good, precise voltage reference.

They can be if you ovenize them. One of those Microsemi diodes is a modern version of the ones used in the old Power Designs precision power supplies (in an oven, of course).


Sent from my Smartphone

[timb]

I mean, why they dont use 4 wire approach in LTZ1000 or LTFLU if they know that wire resistance is gonna drift with time ?

They do. Check the LTZ1000 data sheet. Pretty sure there's force and sense connections on the reference circuit.


Sent from my Smartphone

[SeanB]

The problem is the actual silicon bulk to the diode junction having a varying resistance with both temperature and current. This is solved by selecting the current through the diode so the variation in diode drop with current is exactly equal to the change in resistance, so the voltage stays constant over a small range of temperature. Then you can further select the diodes so the change with temperature and current is roughly the same, but this is not as good as simply selecting the best units at an elevated temperature, and driving them with the correct current. Thus the best are heated to a constant temperature, and the slightly worse ones are not heated but are meant to have a preset current that is individual per diode to get the best performance. 5-6 digits or more need a heated reference, under 4 you can get away with a temperature compensated one.

You cannot make a 4 wire connection on the silicon die itself, any attempt will make a new set of resistors. Best is to make the circuit compensate for the variation.

[jimon]

They do. Check the LTZ1000 data sheet. Pretty sure there's force and sense connections on the reference circuit.

Doesn't look so for LTZ1000, looks like zener diode is just connected with two pins.

It's a bit different with LTFLU : there is a pin between zener diode and transistor, not sure what exactly is going on there.
Attached picture from https://www.eevblog.com/forum/projects/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/

Clearly LTZ1000 and LTFLU doesn't use 4 wire sensing, maybe they just don't need it ? But from this quotes it looks like they do need it :

This was to be pin and function compatible with the SZA263, and they did a pretty good job, with one exception: the LTFLU-1AH uses a different material for the interconnects [aluminum] than Motorola did [a proprietary alloy], and this causes the SZA263 to drift *UP* over time, while the LTFLU-1AH [and the LTZ1000(A) and the LM199/299/399] will drift *DOWN* over time.

This long term [time] drift in these reference IC's is caused mainly by the parasitic resistances of the interconnects and the interconnect/silicon interface-- the "buried" [subsurface] Zener has very little drift over time unless it is damaged during die separation from the wafer.

[timb]

I'll find the circuit for you in a bit. I swear it was the LTZ1000, but maybe I'm losing it!


Sent from my Smartphone

[jimon]

The problem is the actual silicon bulk to the diode junction having a varying resistance with both temperature and current. This is solved by selecting the current through the diode so the variation in diode drop with current is exactly equal to the change in resistance, so the voltage stays constant over a small range of temperature. Then you can further select the diodes so the change with temperature and current is roughly the same, but this is not as good as simply selecting the best units at an elevated temperature, and driving them with the correct current. Thus the best are heated to a constant temperature, and the slightly worse ones are not heated but are meant to have a preset current that is individual per diode to get the best performance. 5-6 digits or more need a heated reference, under 4 you can get away with a temperature compensated one.

You cannot make a 4 wire connection on the silicon die itself, any attempt will make a new set of resistors. Best is to make the circuit compensate for the variation.

Ok, this make sense So 4 wire connection doesn't make sense because resistance of silicon itself will drift, now I can sleep well at night

[timb]

Here you go, the LTZ1000 with sense and force connections.

[Kleinstein]

With a 7 V reference and the relatively small current, there is no real need for a 4 wire connection right of the chip. There are still other parts with more drift. So one has look at the whole circuit - some of the dirft terms can compensate or are trimmed so that they do. For the IC itself there is essentially the same temperature for all parts.

For the circuit beyound the IC package a 4-Wire connection might help, as here temperature may not be that similar in all parts and thus drift term can act separately, though there wil still be a lot of correlation.